Co-production and Uptake of Weather and Climate Services: Evidence of Welfare Impacts Among Farmers in Senegal Using A Dynamic Pan el Data Approach Working Paper No. 355 CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS) Brian Chiputwa Genowefa Blundo-Canto Peter Steward Nadine Andrieu Ousmane Ndiaye Co-production and Uptake of Weather and Climate Services: Evidence of Welfare Impacts Among Farmers in Senegal Using A Dynamic Panel Data Approach Working Paper No. 355 CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS) Brian Chiputwa Genowefa Blundo-Canto Peter Steward Nadine Andrieu Ousmane Ndiaye Correct citation: Chiputwa B, Blundo-Canto G, Steward P, Andrieu N and Ndiaye O 2021. Co-production and Uptake of Weather and Climate Services: Evidence of Welfare Impacts Among Farmers in Senegal Using A Dynamic Panel Data Approach, CCAFS Working Paper no. 355. Wageningen, the Netherlands: CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS). Available online at: www.ccafs.cgiar.org Titles in this Working Paper series aim to disseminate interim climate change, agriculture and food security research and practices and stimulate feedback from the scientific community. The CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS) is a strategic partnership of CGIAR and Future Earth, led by the International Center for Tropical Agriculture (CIAT). The Program is carried out with funding by CGIAR Fund Donors, Australia (ACIAR), Ireland (Irish Aid), Netherlands (Ministry of Foreign Affairs), New Zealand Ministry of Foreign Affairs & Trade; Switzerland (SDC); Thailand; The UK Government (UK Aid); USA (USAID); The European Union (EU); and with technical support from The International Fund for Agricultural Development (IFAD). For more information, please visit https://ccafs.cgiar.org/donors. Contact: CCAFS Program Management Unit, Wageningen University & Research, Lumen building, Droevendaalsesteeg 3a, 6708 PB Wageningen, the Netherlands. Email: ccafs@cgiar.org Creative Commons License This Working Paper is licensed under a Creative Commons Attribution – NonCommercial–NoDerivs 3.0 Unported License. Articles appearing in this publication may be freely quoted and reproduced provided the source is acknowledged. No use of this publication may be made for resale or other commercial purposes. © 2021 CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS). CCAFS Working Paper no. XX DISCLAIMER: This Working Paper has not been peer reviewed. Any opinions stated herein are those of the author(s) and do not necessarily reflect the policies or opinions of CCAFS, donor agencies, or partners. All images remain the sole property of their source and may not be used for any purpose without written permission of the source. Abstract The provision of tailored weather and climate information services (WCIS) to help adapt decision making to climate variability is gaining increasing recognition. This study analyzes the impact of seasonal and daily weather forecasts mediated by a multidisciplinary working group (MWG), a co-production model for weather and climate information. A two-season balanced dataset in combination with panel econometrics was used to explore the impact of uptake of weather and climate information uptake and the impact on farmers’ agricultural income in Senegal. The data were complemented by participatory surveys that provide richer contextual information to explain the impact pathways. Results show that the use of weather and climate information increased the value of crop produce by between 10-25\% for farmers with access to an MWG. Coordinated platforms that involve diverse stakeholders like the MWGs play a critical role in co-producing weather and climate information that are more usable to farmers, thereby improving uptake and livelihoods. The impact pathways and implications for policy are discussed. Keywords weather, climate, fare-casting, impact assessment, impact evaluation, Senegal About the authors Dr. Brian Chiputwa is a Livelihoods and Gender Specialist in the Research Methods Group (RMG) at the World Agroforestry (ICRAF), based in Nairobi, Kenya. His research interests are in the area of impact assessment, adoption of agricultural practices with a focus on developing and testing of methodologies that generate rigorous evidence-based scientific knowledge. He has a PhD in Agricultural Economics and Rural Development from the University of Goettingen in Germany. Contact: B.Chiputwa@cgiar.org Dr. Genowefa Blundo-Canto is a Development Economist at the French Agricultural Research Center for International Development (Cirad). Her research focuses on impact assessment of agricultural research for development (AR4D) interventions. Contact: genowefa.blundo_canto@cirad.fr Dr. Peter Steward is an PhD ecologist (University of Leeds) and CIFOR-ICRAF associate scientist for the CCAFS Evidence for Resilience Agriculture (ERA) project. He has a breadth of scientific experience in areas such as agricultural and conservation ecology, measurement of biodiversity and regulating ecosystem services, human wildlife conflict, agronomic trials for sustainable agriculture, crop-climate suitability modelling, advanced meta-analysis, agricultural ontologies and concept schemes, niche modelling, and prediction of outcome performance using machine learning. Contact: P.Steward@cgiar.org Dr. Nadine Andrieu is a CIRAD senior scientist with 13 years of experience in systemic analysis and modeling of farming systems taking into account synergies and trade-offs between different production activities. Her main research area is the co-design with stakeholders of innovative agroecological farming systems. She has a PhD in agronomy and sustainable development from ABIES (Paris).Contact: nadine.andrieu@cirad.fr Dr. Ousmane Ndiaye works for the Senegalese weather service (ANACIM) as head of the research and application group. He is also an Adjunct Research Scientist at the International Research Institute for Climate and Society (IRI). He is known for innovative work on how to use climate information in rainfed agriculture in the district of Kaffrine to increase farmers’resilience. Contact: ousmane.ndiaye@anacim.sn. Acknowledgements This work was funded under the 2018 call for ex-post impact assessment (ep-IA) of climate change related work in the CGIAR. The call was a competitive bid funded by the CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS) with the aim of providing robust impact evaluations of climate change related work. The work was carried out with support from CGIAR Fund Donors and through bilateral funding agreements. For details see https://ccafs.cgiar.org. Part of this work was also supported by USAID through grant number USAID Contract No: AID-OAA-A-16-00072 under the Climate Information Services Research Initiative (CISRI), A Learning Agenda for Climate Information Services in sub- Saharan Africa. We are grateful to the Meteorological Department in Senegal (ANACIM) for their assistance, to the team of data collectors led by Adama Ba and Souleymane Jules Diol, and to all the respondents in our interviews. Contents 1 Introduction 2 2 Empirical evidence of WCIS impacts on livelihoods 7 3 Research Methodology 9 3.1 Country profile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 3.2 Data and methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 3.2.1 Household surveys . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 3.2.2 Participatory surveys . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 3.2.3 Biophysical data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 3.3 Modeling households’ uptake and use of different WCI . . . . . . . . . . . . . 16 3.4 Empirical estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 4 Descriptive statistics 26 4.1 Differences in socio-economic, institutional, and bio-physical factors . . . . . . 26 4.2 Differences in awareness of, access to, and use of WCI . . . . . . . . . . . . . 29 4.3 Differences in crop productivity and income . . . . . . . . . . . . . . . . . . . 32 5 Empirical results 36 5.1 Econometric results from panel data estimation models . . . . . . . . . . . . . 36 5.2 Participatory pathways linking the use of WCI, MWG and livelihoods . . . . . 39 6 Discussion and conclusions 44 7 Acknowledgements 47 8 Appendix 53 i List of Figures 1 Average and cumulative monthly rainfall in the districts surveyed in 2016 and 2018 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 2 Composition of sampled households according to MWG and use of CIS in the full household survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 3 Farmers being trained on the interpretation and application of seasonal rainfall forecasts in Kaffrine, Senegal. Photo credit: J. Hansen (CCAFS) . . . . . . . . 14 4 Cumulative distribution function for value of main crops in the 2016 season . . 35 5 Cumulative distribution function for value of main crops in the 2018 season . . 36 6 Participatory pathways for users of WCIS . . . . . . . . . . . . . . . . . . . . 41 7 Participatory pathways for non-users of WCIS . . . . . . . . . . . . . . . . . . 42 8 Participatory pathways for non-users of WCIS . . . . . . . . . . . . . . . . . . 53 9 Conceptual schematization of the MWG model . . . . . . . . . . . . . . . . . 57 ii List of Tables 1 General differences in socio-economic, institutional and biophysical factors between WCI users and non-users . . . . . . . . . . . . . . . . . . . . . . . . 28 2 General differences in awareness, access and use to WCI between households in locations with an MWG and with no MWG . . . . . . . . . . . . . . . . . . 30 3 General differences in crop productivity among households that use seasonal forecasts (with MWG) vs. non-users . . . . . . . . . . . . . . . . . . . . . . . 33 4 Estimates of the impact of WCI use from MWG on crop incomes: full SLR results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 5 Estimates of the impact of WCI use from MWG on crop incomes: abridged SLR results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 A1 Estimates of the impact of access to an MWG and use of WCI on agricultural productivity using different models . . . . . . . . . . . . . . . . . . . . . . . . 54 1 Introduction Climate change and variability have been identified as major threats to key sectors such as agriculture that drive economic growth and sustainable development in developing countries. In sub-Saharan Africa (SSA), the agricultural sector contributes about 15% of the total gross domestic product (GDP) (OECD, 2016), with 80% of all farms managed by smallholder pro- ducers and the agricultural sector provides direct employment to about 175 million people (AGRA, 2014). The sector depends primarily on rain-fed agriculture, which is further char- acterized by limited use of improved technologies, and a high prevalence of poverty and food insecurity, low levels of economic diversification and a general lack of adaptation and mit- igation strategies (Alfani et al., 2015; Fischer et al., 2005; Hansen, 2005). This makes the agricultural sector one of the most vulnerable to climate and weather-related risks in sub- Saharan Africa (IPCC, 2014) and highlights the need for improved planning based on climate projections to help reduce risks posed by climate variability (Singh et al., 2017). The provision of tailored weather and climate information services (WCIS) that meet users’ expectations can play an important role in managing current and future climate and weather risks to achieve Sustainable Development Goals (SDGs) under the Vision 2030 (Griggs et al., 2021; Machingura et al., 2018). Most of the 17 SDGs, along with the 169 targets and activities to be implemented are weather and climate-sensitive. The provision of tailored weather and climate information, outlooks, forecasts, advisories, and farmer-focused services can help crop and livestock farmers better manage extreme conditions by making informed decisions to improve productivity and income and consequently reduce poverty (SDG 1) and hunger (SDG 2). SDG 13 recognizes the need for coordination platforms that bring together stakeholders from diverse disciplines in promoting international action and cooperation on 2 climate change through institutional frameworks. 1 Considerable attention has been paid to improving the quality of weather and climate informa- tion (WCI) mainly due to the gap that exists between production and transmission of climate knowledge. Transmission often involves meteorologists and purveyors and users of climate information, who are decision makers in the agriculture, water, health, and other sectors. For weather and climate information to more effectively influence users’ decisions, Cash et al., 2003 proposed three basic criteria that need to be met. First, the information has to be per- ceived as credible by stakeholders along the WCIS value chain; second, it has to be salient i.e., relevant to the needs of users; and third, it has to be legitimate and to take stakeholders’ divergent values, beliefs, knowledge contexts, and interests into account through an open, transparent, and unbiased process. One widely accepted approach in meeting these criteria is the co-design and co-production of weather and climate information, leveraging the exper- tise of different actors to guarantee that the climate science is appropriately tailored to meet the needs of end users (Dilling & Lemos, 2011; Lemos & Morehouse, 2005; Singh et al., 2017). WCIS involve the transformation of climate-related data and other information into customized products such as projections, seasonal rainfall forecasts, advisories, and trends that assist users across different sectors with better planning and decision-making (European Commission, 2015). There have been several initiatives to improve the effectiveness of the provision of tailored WCI to help smallholder farmers in SSA manage climate and weather related risks. through better informed farming decisions (Hansen et al., 2011; Roncoli et al., 2009; Roudier et al., 2014; Vaughan & Dessai, 2014). Examples of such initiatives include the Global Framework for Climate Services (GFCS) that supported a number of African countries in developing na- 1for more details on contribution of weather and climate information services to the SDGs see https:// sustainabledevelopment.un.org 3 tional policy frameworks and action plans for improved delivery of climate services (WMO, 2017).Other initiatives that have supported African countries in building climate and disaster resilience through improved climate services include the Climate for Development in Africa (ClimDev-Africa), established by the African Development Bank (ADB) and its partners, the Africa Hydromet Program that seeks to improve weather, water, and climate services through- out Africa 2, the Climate Change Agriculture Food Security (CCAFS) coordinated by the CGIAR, the Enhancing National Climate Services (ENACTS) developed by the International Research Institute (IRI). 3 While it is widely recognized that the provision of robust WCI can be vital in helping society mitigate, adapt, and build resilience to climate variability and change, the evidence base to support this is still growing. Vaughan et al., 2019 recently reviewed 66 studies on weather and climate services in Africa over the last 40 years, and for relevance to this study, we focus on evidence that can be categorised in two broad strands. The first strand focuses on outcomes that measure the degree to which WCIS interventions lead to changes in awareness, access, and use, including behavioral changes (Amegnaglo et al., 2017; Carr, 2014; Mudombi & Nhamo, 2017; O’Brien et al., 2000; Ouédraogo et al., 2018; Patt et al., 2005; Rasmussen et al., 2012; Roncoli et al., 2009; Tarhule & Lamb, 2003), and more recently (Blundo-Canto et al., 2021; Chiputwa et al., 2019; Chiputwa et al., 2020). 4 The second strand is much thinner and zeroes in on impacts that lead to changes in livelihoods e.g., increased agricultural outputs, income, efficiency, or reduced losses (Anuga & Gordon, 2016; Lo & Dieng, 2015; Ouédraogo et al., 2018; Patt et al., 2005; Rao et al., 2015; Stats4SD, 2017). There are several characteristics of WCIS that present challenges in monitoring and evalu- 2https://www.worldbank.org/en/programs/Africa_hydromet_program 3https://iri.columbia.edu/wp-content/uploads/2013/07/ENACTS7_10v2.pdf 4these later studies were not included in the Vaughan et al., 2019 review 4 ation, e.g., their non-excludable and non-rivalrous nature that make it difficult to establish a counterfactual, and the stochastic nature of weather, which implies that farmers’ adaptive strategies may vary from year to year depending on the variability (see Vaughan et al., 2019 and Tall et al., 2018). As a consequence, while a large number of WCIS initiatives have been implemented in Africa in the last two decades, empirical evidence on their effectiveness in improving livelihoods has been limited to a handful of studies. As Tall et al., 2018 note: ‘In agriculture, many studies have explored the potential value of climate ser- vices, but only a few have evaluated actual services and rigorously tracked their connection to outcomes (i.e., changes in decisions) and impacts (i.e., changes in well-being) at rural household and farm level.’ The present study consequently aims to assess the impact of weather and climate information on the agricultural productivity and household income of smallholder farmers in Senegal. More specifically, we assess the effectiveness of the Multidisciplinary Working Group (MWG) model that co-produces tailored weather and climate information in influencing management responses by farmers, ultimately leading to improved livelihoods in the form of higher crop yields and incomes. Under the MWG co-production model, scientists interact with farmers and exchange ideas on how to integrate farmers’ indigenous knowledge and forecasting meth- ods with scientific weather forecasts, thereby enabling the identification of knowledge gaps at the local level and increasing farmers’ trust in scientific forecasting. Farmers are given the opportunity to present their own climate information needs as well as the way they would like to receive climate information. The local MWGs also manage an early warning system (EWS) based on climate information received from the National Meteorological Agency (ANACIM). The MWG meets every 10 days and provides agricultural advice that is shared with policymakers and farmers through a special program broadcast on community radios. 5 The interactive radio program allows listeners to provide their feedback, including additional information, views, and requests for clarification. Up to 2015, a total of 84 journalists from community radio stations were trained to understand and communicated climate informa- tion to an estimated 740,000 rural households (corresponding to 7.4 million rural people) in Senegal (CCAFS, 2015; Lo & Dieng, 2015). This study expands the limited evidence base on the role of weather and climate services in improving the livelihoods of smallholder farmers in three ways. First, we built on the cross- sectional analysis conducted by Chiputwa et al., 2019 and Chiputwa et al., 2020 by collecting an additional wave of data that enabled us to use a two-time panel approach to reduce selec- tion bias and thus improve the robustness of causal estimates. We are only aware of one other study, Patt et al., 2005, that uses a panel data approach to assess the impact of weather and climate information. Second, the present study goes beyond exploring how WCIS affects out- comes such as awareness, access, uptake, and behavioral changes by exploring causal impacts on farmer’s livelihoods, i.e., crop yields and household income. Chiputwa et al., 2020 found that the presence of MWGs generally increases farmer’s awareness of WCI by 18%, access by 12%, and uptake by 10%, and further demonstrated that this increase in uptake is associ- ated with very significant changes in management responses by farmers exposed to an MWG. Third, we complement quantitative analysis with a participatory approach that captures the im- pact pathways that link outcomes to impacts, seen from the farmer’s’ points of view. This type of blended approach is not only vital in triangulating findings but also provides compelling evidence that explains how the impact pathways work. By combining these two methodolog- ical approaches, the unique perspective of this study contributes to the discourse on rigorous approaches needed to accurately evaluate the impact of weather and climate services. 6 2 Empirical evidence of WCIS impacts on livelihoods The evaluation of whether WCIS actually support agricultural decision-making by small- holder farmers is challenging due to the factors highlighted in Tall et al., 2018. Broadly, assessing the value of using seasonal climate services to support decision-making can be pursued empirically using quantitative and/or qualitative approaches (Bruno Soares et al., 2018). The main advantage of using appropriately designed quantitative methods is that they enable the identification of causal associations between the intervention and impacts that can be replicated and generalized across a larger population. However, they often fail to elucidate the subtle aspects of how these impacts occur and how different members of the sampled com- munities are affected depending on the depth of their vulnerabilities, something that can be better captured through qualitative research involving interactions between users and scientists e.g., participatory and ethnographic methods (see Carr, 2013; Roncoli, 2006); or other par- ticipatory tools such as the participatory impact pathway approach (Douthwaite2003). Most studies of the impacts of using weather and climate services on farmers’ welfare such as agri- cultural yields and/or income, have generally found a positive relationship, albeit with notable variation depending on the agro-ecology, socio-economic, institutional context, crop type and WCI type and accuracy assessed (Vaughan et al., 2019). Ouédraogo et al., 2018, assessed the effects of using seasonal climate forecasts on yields using data from 289 farmers in Burkina Faso. Based on subjective reporting by farmers, the study reported that 78.4% of farmers who received climate information said their crop yields had improved substantially.Stats4SD, 2017, used a mixed-methods approach involving 32 qual- itative interviews and household data from 802 farmers to evaluate climate services under the GFCS Adaptation Programme for Africa in Malawi and Tanzania and more than 90% of 7 those surveyed reported that receiving PICSA training made them more confident in planning and making decisions for their agricultural production and livelihoods. Anuga and Gordon 2016, used data from 320 farmers in Ghana, and, based on multivariate regression analyses, concluded that receiving training on how to access weather information through local informa- tion centers increased yam yields by 17% and that the use of weather information explained about 21% of the variation in maize yields. Ouédraogo et al., 2015 used data generated from an ex-post survey of 170 farmers who were divided into control and treatment groups in Burk- ina Faso. That study showed that access to climate information increased cowpea yields by 24%, and gross margins by 66% (due to savings in seed and pesticide purchases). However, the yields of sesame obtained by farmers that received climate information were 10% lower than yields obtained by the control group. Lo and Dieng, 2015 compared test plots containing groundnut and millet in which farmers used management strategies informed by WCI with plots managed using traditional knowledge. The study concluded that the use of seasonal fore- casts and weather forecasts (2-3 days and 10 days) led to between 15% and 50% increase in groundnut and millet yields, which resulted in increases in the income and resilience of house- holds in Senegal during the lean period. Patt et al., 2005 used a two-year dataset and a control group to estimate the impact of farmer participation in a participatory climate information workshop on yields in Zimbabwe. Their methodology was based on a multivariate regression analysis that controls for use of forecast and location. Although the study found that farmers who participated in the workshops had significantly higher yields, particularly in the second year, no strong links were found between management responses to the forecast and increased yields. 8 3 Research Methodology 3.1 Country profile Senegal is a predominantly rural economy, where rain-fed production systems are the key drivers of economic growth. It is among the fastest growing economies in Africa and in 2016, the gross domestic product (GDP) showed growth rate of 6.7% (Loayza et al., 2018). How- ever, despite the general decline in poverty levels due to economic growth in Senegal (47% in 2010), the poverty rate among the rural population was even higher, around 51% in the same year (Loayza et al., 2018). Agricultural and livestock production are the main economic activi- ties, representing 17.5% of the GDP in Senegal and employing 69% of the population directly and indirectly (FAO, 2015). Although the agricultural sector accounts for a relatively smaller share compared to other sectors of the Senegalese economy, it is key to poverty reduction as it is a major source of employment and income for poor households who are mostly located in rural areas. The country’s agricultural sector grew at an average rate of 3.2% between 2000 and 2016, but volatility around that average was large (Loayza et al., 2018). The significant fluctuations in agriculture growth were mainly the result of weather and climate hazards that severely affected pastoralism and rain-fed crops. Weather is one of the most important produc- tion risks in Senegal due to moisture stress caused either by erratic rainfall, early end of the rainy season, delayed onset of the rainy season, extreme rainfall events or extended drought. More than 40% of the variation in national crop yields can be attributed simply to variation in annual rainfall amounts (D’Alessandro et al., 2015). Hence, for Senegal to achieve and maintain high output growth, more efforts are needed to protect the agricultural sector against climate variability and to enhance livelihood resilience in rural areas. 9 Rainfall is the key to agricultural production as more than 95% of the land cultivated is under rain-fed conditions. The agricultural economy is characterized by the dominance of small- holder farmers cultivating millet, sorghum, groundnuts, maize, and rice for subsistence. To adapt to weather and climate variability, in West Africa farmers use a variety of indigenous and modern coping strategies such as soil and water conservation practices, water harvesting techniques, and more recently, climate information services. While WCI has the potential to provide farmers with timely weather information to help them make appropriate decisions in risk management, most of the initiatives on WCI dissemination in West Africa have failed due to a mismatch between what scientists produce as forecasts and what farmers need at the local level (Ouédraogo et al., 2018; Singh et al., 2017). Senegal is characterized by a hot semi-arid climate with a rainy season lasting from mid-May to early November, and a dry season be- tween November and May. Using data extracted from the International Research Institute for climate and society database, Figure 1 shows the average and cumulative monthly rainfall in surveyed districts for 2016 and 2018, respectively. The rainfall pattern was generally the same for Kaffrine and Kaolack provinces in the two survey years with rains starting in June and ending in October, and the peak rainfall periods were July to September in 2016 and August to September in 2018. 3.2 Data and methods 3.2.1 Household surveys In this study, we use a mixed-methods approach that combines household data from a two- year balanced panel with participatory impact surveys collected from smallholder farmers 10 Figure 1: Average and cumulative monthly rainfall in the districts surveyed in 2016 and 2018 Source: IRI database5 in Senegal. The study builds up initial surveys conducted by Chiputwa et al., 2020 in 2017, under the USAID funded Climate Information Services Research Initiative (CISRI)6. We fol- lowed up by re-interviewing a total of 596 households out of the total of 795 interviewed in the first wave. The households were selected in Kaffrine (which is the treatment province with access to functional MWGs) and Kaolack (which is the control province with no access to the MWG). Within Kaffrine region, we purposively selected two districts i.e. Kaffrine (with ac- cess to MWG). The sampling criteria were informed by spatial mapping of functional MWGs based on their performance (see Figure 4 in Chiputwa et al., 2020). In terms of MWG perfor- mance, the MWGs in the Kaffrine district (our treatment group) had held on average up to six meetings in 2017 alone, while the control group region in Kaolack had no functioning MWGs. 6https://www.climatelinks.org/resources/climate-information-services-research-initiative 11 The questionnaire captured complimentary information to that collected in the first wave, i.e. household information (e.g., demographics, education, asset ownership, income generating activities); farm characteristics and agricultural production (e.g., plot characteristics, crop and livestock data, crop varieties grown); other off-farm crop production livelihood strategies. In addition, the questionnaire collected detailed information on the household’s awareness of, access to, and use of weather and climate services such as seasonal forecasts on the onset and end of the rainy season and short-term weather services such as weekly and daily forecasts. All questions on uptake of WCI and the farm management responses informed at plot level refer to the agricultural season ending in 2018, which was just prior to the household survey. For a detailed account of the sampling design and strategy, see Chiputwa et al., 2020. Figure 2, shows details of the relative number of farmers interviewed and their distribution in relation to access to an MWG and use of WCI use (see also Figure 9 in the Annex for further details on how the MWGs operate). Figure 3 shows pictures taken during a training workshop for farmers on the interpretation and application of seasonal rainfall forecasts in Kaffrine district, the main pilot district for CCAFS Adaptation Through Managing Climate Risk project in Senegal. 3.2.2 Participatory surveys A total of six focus groups with men and women farmers were held prior to the household surveys to try and understand household level impact pathways related to the use of climate information. To mirror the logic of the sampling used for the household surveys, eight to ten participants were invited based on the following criteria: • Promoter farmers (men) from Kaffrine who have access to the MWG and use WCI (8 12 Pooled Sample (n=596) Households in Households in Kaffrine province Kaolack province (n=403) (n=193) Households with access to the Households without MWG MWG (n=403) (n=193) Households Households Households Households Households Households Households Households USING USING USING all CIS NOT USING any CIS USING USING USING all CIS NOT USING any CIS at least one CIS seasonal forecast (n=143) (n=55) at least one CIS seasonal forecast (n=18) (n=65) (n=343) (n=303) (n=106) (n=96) Figure 2: Composition of sampled households according to MWG and use of CIS in the full household survey participants) • Promoter farmers (women) from Kaffrine who have access to the MWG and use WCI (8 participants) • Female farmers from Kaffrine who have access to the MWG but do not use WCI (11 participants) • Female farmers from Kaolack who have no access to the MWG but use WCI (9 partici- pants) • Male farmers from Kaolack who have no access to the MWG but use WCI (10 partici- pants) • Male farmers from Kaffrine who have no access to the MWG and do not use WCI (10 participants) 13 (a) Facilitators leading the training (b) Farmers participating in the training (c) Farmers reading a rain-gauge (d) Facilitators leading the training (e) Facilitator interviewing a local farmer (f) Facilitator leading discussions with farmers Figure 3: Farmers being trained on the interpretation and application of seasonal rainfall forecasts in Kaffrine, Senegal. Photo credit: J. Hansen (CCAFS) 14 3.2.3 Biophysical data One of the key challenges in modeling farmers’ behavior in the use of WCI is that farm man- agement decisions depend on stochastic variables, i.e., rainfall and temperature as well as soil organic carbon, which are important to account for when modeling agricultural yields. The rainfall data used in this study were taken from rain gauges and extracted from satellite observations based on the Climate Hazards Group InfraRed Precipitation with station data (CHIRPS) (Funk et al., 2015). 7 The CHIRPS rainfall data represent a more than 30-year quasi-global rainfall dataset. CHIRPS incorporates 0.05° resolution satellite imagery with in- situ station data to create gridded rainfall time series for trend analysis and seasonal drought monitoring. A temporal down-scaling is applied in order to improve daily estimates. CHIRPS were screened and validated over most part of the world including the Sahel (Funk et al., 2015). Temperature data were extracted using NASA’s Modern-era Retrospective analysis for Research and Applications, version two (MERRA-2) (see Gelaro et al., 2017). 8 NASA’s Modern-Era Retrospective Analysis for Research and Applications, version two (MERRA-2), a global atmospheric reanalysis produced by the NASA Global Modeling and Assimilation Office (GMAO). It spans the satellite observation period from 1980 to the present. The goals of MERRA-2 are to provide a regularly gridded, homogeneous record of the global atmosphere, and to incorporate additional aspects of the climate system including trace gas constituents (stratospheric ozone), and improved land surface representation, and cryospheric processes. It is a widely used data set and captures temperature variations fairly well. Data were extracted from the International Research Institute for climate and society database. 9 For each location, we used data from the nearest grid point. 7https://developers.google.com/earth-engine/datasets/catalog/UCSB-CHG_CHIRPS_DAILY 8https://gmao.gsfc.nasa.gov/reanalysis/MERRA-2/ 9https://iri.columbia.edu/resources/data-library/ 15 Data on soil organic carbon were derived from the SoilGrids dataset (Hengl et al., 2017; ISRIC., 2018). The SoilGrids system provides global predictions at 250m resolution for standard numeric soil properties at seven standard depths (0, 5, 15, 30, 60, 100 and 200 cm). We calculated the weighted mean of soil organic carbon for the 0-30 cm soil layer with a 100 m buffer radius around each household, the reported accuracy for this variable is 63.5% (Hengl et al., 2017). Data on slope were calculated using the terrain function of the raster package and void-filled 3” resolution digital elevation models (Abrams et al., 2020). Values were extracted and averaged for a 100 m buffer radius buffer centered around each household based on their GPS coordinates. 3.3 Modeling households’ uptake and use of different WCI Following Chiputwa et al., 2020, we consider the impact of uptake and use of six different weather and climate information products i.e., total rainfall for the season; onset of the rainy season; end of the rainy season, daily weather forecasts (2-3 days and 10 days); and instant forecasts of extreme events. Firstly, an individual can only access or receive a particular weather and climate information if they are already aware of it. A household is considered to be aware of an innovation when their information level on the technology exceeds a min- imum threshold (Adegbola & Gardebroek, 2007). Therefore, awareness of and having the ability to receive or access a particular WCI are necessary but not sufficient conditions for the individual to be able to use this information to influence his/her farming decisions. Conse- quently, a farmer can only uptake and use WCI to inform decision making and if he/she has access and the means to make alternative decisions and/or to implement alternative strategies. Hence, the decision to use is defined as a binary variable for each individual and takes the value of 1 for a household that uses a specific WCI to adjust their farm management responses, 16 and 0 if the household does not. This implies that for each household, we can see whether they used any one or a combination of the six weather and climate information products. A household is classified as a WCI user if they used at least one of the six types of WCI to ad- just their on-farm decisions in the 2016 agricultural season. Non-users, on the contrary, are households that used none of the six types of WCI to inform their farming decisions. The farm management responses informed by the use of WCI depend on the type and timescale of the information provided. For example, while seasonal forecasts provide a general overview of the season, they cannot inform on day-to-day fluctuations in the weather. It is important to note that the decision to use information depends on (i) the household being aware of at least one type of weather or climate information, its attributes, and its potential net utility; (ii) the household having the means to receive this information; and iii) the household is able to change its practices/actions based on this information. We define awareness, access and use of WCI in line with Chiputwa et al., 2019. Awareness is expressed as a dummy variable for each WCI and takes a value of 1 if the household has sufficient knowledge of any particular WCI to be able to consider this information to inform their decision making, and 0 if otherwise. Access is measured as a binary variable that takes the value of 1 for each weather and climate service that the household is able to receive from one or more sources e.g., radio, extension workers, or from fellow farmers, and 0 if otherwise. The decision to use is a function of the benefit expected from using the WCI, which depends on the attributes of the WCI in question including source and accuracy, as well as other socioeconomic factors (e.g., age and educa- tion level of the farmer, the size of the farm) institutional factors (e.g., access to extension and markets) and the ability of the decision maker to act on the information, to implement alternatives. The impact of a given WCI is difficult to assess because information alone has no intrinsic 17 value. The value is only realized when the information is translated into farming decisions that result in positive benefits or utility for the user. We model a household’s decision to use WCI using household decision making under imperfect information and a random utility framework. In this framework, we assume that a household chooses to use WCI based on the maximization of an underlying utility function,*, which is determined by a set of farm and household variables, X, and can be represented in the form: MAX U = f(X). (1) We assume that household i will use one or a combination of WCI j, where j(j=1,...J), if the utility*8 9 derived is greater than the utility*8< of not using WCI. Since the utilities cannot be observed, they can be expressed as a function of observable elements and can be represented by latent variable model as: ∗ = *8 9 −*8< > 0, for all 9 ≠ <, (2) where ∗ represents the benefits of using WCI j as opposed to not using m. While ∗ is un- observed, we can observe the type of WCI the household uses. The probability that a farmer uses WCI j can be denoted as %A ( = 1), otherwise ∗ takes a value of zero. The utility maxi- 18 mizing behavior of farmers can then be represented as:  ∗ if 8 9 ≥ 0*8 =  (3)0 if 8< < 0. If a linear relationship is assumed, ∗ can be written as: ∗8 9 = V 9-8 + D8 9 , (4) where ∗ is a latent variable determined by a broad set of observed household and farm char- 8 9 acteristics, and institutional factors -8, as well as unobserved factors affecting the uptake decision contained in D8 9 . The households’ demand for WCI (decision to use) is given by  1 if  (*8 9 −*8<) ≥ 0⇔ V 9-8 ≥ −D3 8 98 9 =  (5)0 if  (*8 9 −*8<) ≥ 0⇔ V 9-8 < −D8 9 , where 3∗ is the expected utility differential of using WCI. The use decision is a function of 8 9 the expected benefits from the uptake of WCI, which depends on the attributes of the WCI in question such as source and accuracy, as well as other socioeconomic and institutional factors that may constrain uptake. The fundamental first-step determinants that act as preconditions before a household uptakes WCI are (i) awareness or knowledge and (ii) access to or ability to receive the information. Following the empirical applications of Shiferaw et al., 2015 and Adegbola and Gardebroek, 2007, a latent variable 0∗ can be defined as the level of aware- 8 9 19 ness of a household about a particular WCI. This level of awareness is dependent on the level of information or knowledge acquisition 8 9 that facilitates the household’s awareness and sufficient knowledge of the innovation, that is above the minimum level of information thresh- old 8< to be able to make the decision to use it. This level of awareness is also affected by a set of observed household and farm characteristics, and institutional factors /8. If a linear relationship is assumed, ∗ can be written as:   1 if (0 − 0 ) ≥ 0⇔ U /0 8 9 8< 9 8 ≥ −n8 98 9 =  (6)0 if (0 − 0 ) < 0⇔ U / < −n8 9 8< 9 8 8 9 Similarly, a household’s level of access to WCI through various channels such as radio or extension services can be represented by the latent variable A∗ and can be given by: 8 9  1 if (A − A ) ≥ 0⇔ W 9"A 8 9 8< 8 ≥ −Y8 98 9 =  (7)0 if (A − A ) < 0⇔ W 9" < −Y8 9 8< 8 8 9 In reality, what we observe is the household’s use of WCI which can be expressed as:  3 0 A  1 if (A − A ) ≥ 0⇔ W8 9 8< 9"8 ≥ −Y8 9 =    =  (8)0 if (A − A ) < 0⇔ W 9"8 9 8< 8 < −Y8 9 The uptake of a particular WCI occurs when several factors hold simultaneously, i.e., the household is sufficiently aware of the innovation (0 = 1) ; the expected utility differential has a net positive (3 > 0) and the household is able to receive or access WCI (A = 1). Therefore, 20 the probability of uptake, P(I), of WCI can be given by: %() = %(3) ∗ %(0) ∗ %(A) (9) Such a conceptual framework for farm household decision making under information and access exposure illustrates the importance of the variables that determine awareness and access to information about the innovation, its net benefits, and how these influence the uptake behavior of smallholder farmers. 3.4 Empirical estimation Program evaluation is the assessment of cause and effect factors with the aim of determining the extent to which the net difference in outcomes between users and non-users of an inno- vation can be attributed to an intervention. The main concern is threats to internal validity; these threats are external factors affecting outcomes other than the intervention itself. In other words, the net difference in outcomes could have occurred in the absence of the intervention. The counterfactual approach to impact evaluation pioneered by Rubin, 1974, consists of mea- suring what would have happened to users in the absence of the intervention. However, it is not possible to observe the state of nature of users had they not participated in the interven- tion, implying that the data collected can only be on the factual. Therefore, this is basically a missing data issue. A central tenet of the average treatment effect ()) framework is the potential-outcomes model (also known as Rubin’s causal model (Rubin, 1974) based on the idea that every subject has different potential outcomes depending on the group to which the subject is assigned. In this case, the potential outcomes of a household that uses WCI will 21 differ from those of a household that does not use WCI. The treatment is a binary variable 8 that is set to 1 if the farmer is aware, has access to, and uses WCI to inform his/her farming decisions and 0, if otherwise. The household has two hypothetical scenarios, 0, representing the potential outcome of not using WCI, and 1, representing potential outcomes of using WCI. The potential-outcomes model provides a solution to the missing data problem and allows us to estimate the distribution of individual-level treatment effects. The econometric estimation of generating a counterfactual in non-experimental studies consists of selecting a comparison group with similar characteristics to the treatment group for comparison. In this case, any difference that arises between the two groups can be attributed to the program rather than to other external factors. However, the empirical challenge is addressing the self-selection bias when estimating the average treatment effects on the treated. Most approaches assume that selection into treatment is exogenous after controlling for observed factors (i.e., unconfound- edness of the treatment conditional on a set of observed covariates), or that selection into treat- ment is endogenous (both observed and unobserved factors matter) (Imbens & Wooldridge, 2009; Imbens, 2000). For observations in # households, we denote a binary variable F to indicate the observed status of WCI uptake, where F = 1 if the farmer is exposed to and uses a particular WCI (treated), and F = 0 if the farmer is not exposed to and/or uses WCI. In addition, we also consider exposure to the MWG as also taking a binary variable, with < = 1 if the farmer lives in a location where they are exposed to an MWG and < = 0, if they are not exposed. Hence, 22 we can construct an interaction treatment indicator  0  if F = 0, & < = 0 0  if F = 1, & < = 0 = F<1 = 0(1 − F) (1 − <) + 1F<  (10)  0 if F = 0, & < = 1  1 if F = 1, & < = 1, where  is a dummy variable that is only switched on only if the interaction of WCI uptake and the presence of MWG are both true  = F<1 = 0(1 − F) (1 − <) + 1F< (11) For a population of N households, we denote the potential uptake of WCI with a binary vari- able  with the observed status of uptake being 1, implying they are exposed (treatment) and 0, if the farmer is not exposed (control). Therefore, under incomplete exposure, the treatment effect on a farmer 8 can be measured by the difference 1 − 0. Similarly, the expected popula- 8 8 tion uptake impact of using WCI in the presence of an MWG is expressed as the mean value  (1 − 0) , which in theory, is the ) of exposure. Since it is not possible to simultaneously observe the outcomes of the same individual with and without exposure, 1 − 0 cannot be mea- 8 8 sured. Therefore, since exposure is a necessary precondition for uptake, 0, assumes a value of zero. Thus, the adoption impact of any farmer is given by 1, hence the mean uptake is re- duced to  (1). Therefore, for the sample of individuals exposed, the mean impact of uptake on the exposed sub-population is given by the conditional expected value  (1 |F = 1|< = 1), which is the ) on the treated ()1). Similarly, for the non-exposed sub-sample, the mean 23 impact of uptake is given by  (0 |F = 0 & |< = 0|< = 1), which is the ) on the untreated ()0). This approach can be formally represented with a generalized panel model as shown in equa- tion 12. In this case, we want to estimate the effect of using WCI in the presence of an MWG on agricultural productivity. Using the gross value of crop production per from the three main crops (peanuts, maize, and millet) in the study area as the main outcome variable, we use the Cobb-Douglas production function as follows: ;=(.8C) = U + V;=(-8C) + W;=(8C) + d;=(8C) + a8 + aC + n8C , (12) where -8C is a vector of farmer’s characteristics (e.g., sex, age, and education of household head, access to off-farm income) and farm characteristics (cultivated area, labor, livestock), productive assets and inputs (fertilizer, improved seed and manure), and institutional char- acteristics (e.g., access to extension and markets) for individual 8 in year C. 8C is a vector of biophysical factors such as rainfall, temperature, slope, soil organic matter for individual 8 in year C. a8 the inclusion of time-invariant unobserved demographic characteristics, while a8 are time-invariant district-fixed effects, to control for further geographic differences in each of the two years. The coefficient d, is the effect of the treatment on individual 8 and it measures how the gross value of production for the individual would have differed in any given period when the individual with access to an MWG, uses WCI compared to the alternative scenario where they do not use WCI or do not have access to an MWG. The variable n8C is the error term for which a strict exogeneity condition is assumed to hold; the errors are independently and normally distributed with zero mean and constant variance. 24 When estimating how different covariates affect an outcome variable such as agricultural productivity, often some characteristics remain unobserved (e.g., information asymmetries, skills, and abilities) that are likely to be correlated with productivity. Estimating a model with this type of correlation using ordinary least squares (OLS) regression models yields estimators that are biased upwards or downwards (Wooldridge, 2010). Popular panel data models include the fixed effects (FE) and the random effects (RE) models. The FE model assumes that the individual specific effect is correlated with the independent variables. The RE model assumes that the individual unobserved heterogeneity is uncorrelated with the independent covariates. The RE model can help control for unobserved heterogeneity when the heterogeneity remains constant over time and is not correlated with independent variables. This constant can be removed from the longitudinal data through first differencing, which removes any time invariant components of the model Wooldridge, 2010. Using panel data econometrics as we do in this study, we use the cross-sectional and time- series aspects for which we have data on the same households observed at different points in time. Panel data allows for richer model specifications that minimize some of the bias resulting from unobserved heterogeneity. This study follows estimation approaches used in previous studies, such as that by Amare et al., 2018, , who model the impact of rainfall shocks on agricultural productivity, or by Björkman-Nyqvist, 2013 , who estimates the impact of education on income, both of which use panel data. Specifically, for this study, we use the sequential linear panel data estimation approach proposed by Kripfganz and Schwarz, 2019 that uses a two-stage approach. To account for both time-varying factors and time-invariant factors that affect productivity, the first-stage model allows for time-varying independent variables (e.g., use of inputs like improved seed, fertilizer, cultivated area), while the second- stage model incorporates time-invariant independent variables such as altitude, slope, and soil organic matter. 25 4 Descriptive statistics This section presents descriptive statistics on the differences between WCI users with access to an MWG and non-users in terms of their household, farm, and institutional characteristics as well as their awareness, access to and use of individual WCI. We also present differences in agricultural productivity and crop incomes between users and non-users of WCIs in the two regions of Senegal . 4.1 Differences in socio-economic, institutional, and bio-physical factors Table 1 shows the differences between WCI users and non-users in terms of household and farm characteristics; institutional factors; and biophysical factors. A household is classified as a WCI user in both panel surveys if they used at least one of the six WCI products (and have access to an MWG) to adjust their on-farm decisions in the agricultural season preceding the survey. Non-users, on the other hand, are households that did not use any of the six WCI to inform their farming decisions. This could have been for a variety of reasons such as lack of awareness, access to or knowledge of the value and net benefits to be derived from using each WCI product. In general, households with access to an MWG that use at least one of the WCI products are household heads with a higher education level and who are full time farmers. This was consis- tent in the two rounds of surveys. The results also reveal that households using WCI had sig- nificantly higher scores on the Poverty Probability Index (PPI). 10, implying that households 10The PPI is a user-friendly and indirect tool for measuring household poverty developed by the Grameen 26 with access to an MWG and who used WCI were less likely to fall below the US$1.25 poverty line than non-users. However, more rigorous analysis would be needed to check whether the inverse relationship observed between WCI use (with access to an MWG) and poverty is in- deed a causal one. Our results also showed that significantly more male headed households with access to cellphones, a radio, and group membership used WCI in the second wave than in the first wave. When it comes to institutional factors, a significantly higher proportion of WCI users in both waves had greater access and were located closer to extension services than non-users. In the first wave, farmers in locations with an MWG travelled an average of 8.7 km to access extension services, while those in locations with no MWG travelled an average of 19.6 km. Similarly, in the second wave, farmers in locations with an MWG travelled an average of 8.3 km to access extension services while those in locations with no MWG cov- ered 16.5 km. In the 2017 wave, 26% of households with access to an MWG did not use WCI compared to about 24% in the 2019 surveys. There was also a positive correlation between WCI users with the following variables altitude, farm slope, temperature, and rainfall. Over- all, comparing the first and second waves, we fund no significant differences in the three key factors of production i.e., land under cultivation, productive asset index and the total livestock units. 11 Foundation and measures consumption-based poverty by considering numerous questions contained in income and expenditure surveys. The PPI is a country-specific poverty measurement tool, available for 60 countries available at https://www.povertyindex.org/ppi-country 11The Tropical Livestock Unit (TLU) is a common unit that describes livestock numbers across species to produce a single index weighted according to the specie type and age using the “Exchange Ratio” concept. Live- stock is considered an important source for the supply of energy, food, and support for agricultural production. Among rural families in different parts of the world, livestock is also a store of wealth. The more livestock a household owns the wealthier they are considered in society. (see Njuki et al., 2011 for further details) 27 Table 1: General differences in socio-economic, institutional and biophysical factors between WCI users and non-users . First wave (2017) Second wave (2019) Non-Users/(SD) Users/(SD) Non-Users/(SD) Users/(SD) Household and farm characteristics Male household head (dummy) 0.95 0.96 0.90 0.95∗∗ (0.21) (0.19) (0.30) (0.22) Age of household head (years) 51.56 49.26∗∗ 53.02 51.32 (12.58) (13.65) (13.89) (13.48) Education level of household head (years) 1.47 2.06∗ 1.25 1.76∗ (3.35) (4.06) (3.32) (3.73) Cultivated area (Ha) 8.32 8.84 7.50 8.24 (8.28) (8.94) (7.55) (10.08) Members fully engaged in farming 0.23 0.36∗ 0.36 0.27 (0.79) (1.07) (0.97) (0.92) Full time farming (dummy) 0.83 0.89∗∗ 0.71 0.82∗∗∗ (0.38) (0.31) (0.45) (0.39) Group membership (dummy) 0.52 0.70∗∗∗ 0.70 0.73 (0.50) (0.46) (0.46) (0.45) Productive asset index 22.50 24.84 17.87 19.52 (17.56) (20.26) (15.24) (15.28) Total livestock unit 4.03 4.40 3.12 4.14 (7.38) (13.01) (5.65) (9.21) Poverty Probability Index (PPI) score 20.79 23.21∗ 23.34 26.00∗∗ (15.58) (16.95) (14.54) (17.03) Access to radio 0.82 0.88∗∗ 0.83 0.87 (0.39) (0.32) (0.38) (0.34) Ownership of cell phone by male 0.98 0.97 0.96 0.99∗∗∗ (0.14) (0.17) (0.20) (0.09) Ownership of cell phone by female 0.79 0.72∗ 0.81 0.84 (0.41) (0.45) (0.39) (0.36) Institutional factors Access to extension (dummy) 0.10 0.31∗∗∗ 0.15 0.26∗∗∗ (0.31) (0.46) (0.36) (0.44) Presence of a MWG (dummy) 0.26 1.00∗∗∗ 0.24 1.00∗∗∗ (0.44) (0.00) (0.43) (0.00) Distance to extension (km) 16.55 8.84∗∗∗ 14.75 8.10∗∗∗ (18.57) (13.34) (25.32) (15.87) Distance to all weather road (km) 2.63 3.68∗ 2.48 2.15 (3.65) (8.98) (3.72) (2.57) Biophysical factors Altitude (m) 29.66 25.94∗∗∗ 29.29 26.31∗∗∗ (8.18) (12.16) (7.87) (12.32) Slope (◦) 0.60 0.69∗∗∗ 0.62 0.67∗∗ (0.20) (0.30) (0.23) (0.29) Soil organic carbon (g/kg) 93.36 95.38∗∗∗ 94.13 94.76 (5.53) (9.43) (4.18) (9.94) Rainfall (mm) 518.76 613.04∗∗∗ 487.87 585.97∗∗∗ (52.77) (43.30) (55.24) (27.31) Temperature (◦c) 28.62 28.20∗∗∗ 28.35 27.87∗∗∗ (0.25) (0.00) (0.27) (0.00) No. of cases 193 403 193 403 Standard deviations in parentheses, *** p<0.01, ** p<0.05, * p<0.1 Users: Interaction term of WCI use combined with access to MWG; Non-users: No use of WCI &/or No MWG access 28 4.2 Differences in awareness of, access to, and use of WCI Table 2 also lists the general differences in awareness of, access to, and uptake of WCI be- tween households in locations with or without an MWG. Overall, a high proportion of the sampled households in both survey waves were aware of at least one type of weather or cli- mate information in locations with or without an MWG. Awareness of individual climate services was generally lower in the first wave than in the second. In addition, awareness in locations with no MWG was also consistently lower than in locations with an MWG. For example, in the first wave, only 63% of farmers sampled in locations with no MWG were aware of seasonal forecasts compared to 94% in locations with an MWG. In the second wave, awareness of seasonal forecasts increased to 89% in locations with no MWG and to 96% in locations with an MWG. The trend was similar for daily forecasts and early warning services. Based on the conceptual framework illustrated in Section 3.3, a household’s ability to access a specific WCI, is conditional on that household first being aware of and having enough back- ground knowledge about the specific WCI. On average, 65% of the households sampled in locations with no MWG and 93% of the households sampled in locations with an MWG ac- knowledged accessing at least one out of the six types of WCI in the first wave compared to 89% of households in locations with no MWG, and to 99% of households in locations with an MWG in the second wave. Access to seasonal forecasts in locations with no MWG also increased from 49% in the first wave to 85% in the second wave, while in locations with an MWG, the increase was from 87% to 94%, respectively. The trend was the same when consid- ering household access to daily forecasts and EWS. Similarly, household’s use of a specific WCI is conditional on the household being aware of 29 Table 2: General differences in awareness, access and use to WCI between households in locations with an MWG and with no MWG First wave (2017) Second wave (2019) No MWG/(SD) MWG/(SD) No MWG/(SD) MWG/(SD) Awareness of different WCI types Awareness of at least one WCI 0.82 0.99∗∗∗ 0.93 1.00∗∗∗ (0.38) (0.09) (0.25) (0.00) Awareness of seasonal forecasts 0.63 0.94∗∗∗ 0.89 0.96∗∗∗ (0.48) (0.24) (0.31) (0.21) Awareness of daily forecasts 0.75 0.92∗∗∗ 0.80 0.99∗∗∗ (0.43) (0.27) (0.40) (0.11) Awareness of EWS 0.53 0.81∗∗∗ 0.69 0.93∗∗∗ (0.50) (0.39) (0.46) (0.26) Awareness of all WCI types 0.24 0.43∗∗∗ 0.38 0.67∗∗∗ (0.43) (0.50) (0.49) (0.47) Access to different WCI types Access to at least one WCI 0.65 0.93∗∗∗ 0.89 0.99∗∗∗ (0.48) (0.25) (0.32) (0.11) Access of seasonal forecasts 0.49 0.87∗∗∗ 0.85 0.94∗∗∗ (0.50) (0.34) (0.36) (0.23) Access of daily forecasts 0.60 0.84∗∗∗ 0.76 0.97∗∗∗ (0.49) (0.37) (0.43) (0.17) Access of EWS 0.38 0.68∗∗∗ 0.64 0.91∗∗∗ (0.49) (0.47) (0.48) (0.29) Access to all WCI types 0.18 0.33∗∗∗ 0.34 0.63∗∗∗ (0.39) (0.47) (0.48) (0.48) Use of different WCI types Use of at least one WCI 0.40 0.83∗∗∗ 0.55 0.85∗∗∗ (0.49) (0.37) (0.50) (0.36) Use of seasonal forecasts 0.32 0.76∗∗∗ 0.50 0.75∗∗∗ (0.47) (0.43) (0.50) (0.43) Use of daily forecasts 0.33 0.74∗∗∗ 0.37 0.70∗∗∗ (0.47) (0.44) (0.48) (0.46) Use of EWS 0.27 0.56∗∗∗ 0.32 0.69∗∗∗ (0.44) (0.50) (0.47) (0.46) Use of all WCI types 0.10 0.26∗∗∗ 0.09 0.35∗∗∗ (0.31) (0.44) (0.29) (0.48) Intensity of WCI use Number of seasonal forecasts used 0.73 1.79∗∗∗ 0.99 1.77∗∗∗ (1.17) (1.22) (1.14) (1.23) Number of daily forecasts used 0.50 1.12∗∗∗ 0.53 1.14∗∗∗ (0.77) (0.79) (0.76) (0.85) Number of WCI used 1.49 3.48∗∗∗ 1.84 3.61∗∗∗ (2.17) (2.15) (2.09) (2.24) Behavioural changes made Changes based on seasonal forecasts 1.53 5.32∗∗∗ 3.23 4.55∗∗∗ (3.21) (4.57) (4.07) (3.87) Changes based on daily forecasts 0.88 3.06∗∗∗ 1.28 2.07∗∗∗ (2.49) (3.57) (2.03) (2.16) Changes based on EWS 0.43 1.32∗∗∗ 0.76 1.44∗∗∗ (1.70) (2.65) (1.46) (2.14) No. of cases 193 403 193 403 Standard deviations in parentheses, *** p<0.01, ** p<0.05, * p<0.1 Users: Interaction term of WCI use and access to MWG; Non-users: No use of WCI &/or No MWG access 30 and having access to that specific WCI. Our results revealed significant differences between households in locations with an MWG and households in locations with no MWG. Consis- tently, across the two waves, farmers in areas with access to an MWG used more WCI to make their farming decisions than farmers with no access to an MWG. In the first wave, the pro- portion of farmers in locations with an MWG who used seasonal, daily, and EWS forecasts was almost double the proportion of farmers in locations with no MWG who did so. This difference was significantly reduced in the second wave. For example, in the first wave, 40% of farmers with no MWG used at least one WCI to make farming decisions compared to 83% who had an MWG. In the second wave, 55% of farmers with no access to an MWG used at least one WCI to make farming decisions compared to 85% with access to an MWG. Sim- ilarly, in the first wave, 32% of farmers in locations with no MWG used seasonal forecasts compared to 76% who had an MWG. In the second wave, 50% of farmers with no MWG used seasonal forecasts compared to 75% of farmers who had access to an MWG. All these differences are significant at the 1% level. The use intensity is defined as the total number of different types of WCI a household used to inform farm management decisions in the agricultural seasons preceding the two survey years. In general, households in locations with an MWG used significantly more combinations of different types of seasonal and daily forecasts than households in locations with no MWG. More specifically, in the first wave, farmers in locations with no MWG used on average two different types of WCI to inform farming decisions, while farmers in locations with an MWG used four different types. Similar use intensity was found in the second wave. The last part of Table 2 shows the behavioral changes made by farmers in adjusting their farm management decisions. Farmers with access to an MWG made significantly more farm man- agement decisions after receiving seasonal, daily and EWS than farmers with no access to 31 an MWG. In the first wave, on average, farmers with no access to MWG made approximately two behavioral changes that were informed by use of seasonal forecasts compared to approxi- mately five behavioral changes made by farmers with access to an MWG. In the second wave, farmers with no access to MWG used seasonal forecasts to adjust approximately three be- havioral changes based on seasonal forecasts compared to five behavioral changes made by farmers with no MWG. Similarly, considering changes made by farmers with access to an MWG after listening to the daily forecast, there was a slight decline in the mean value from 3 to 2, implying that in the second wave fewer farmers used daily forecasts to make farming decisions than in the first wave. 4.3 Differences in crop productivity and income Table 3 summarizes the differences in crop production variables among households using seasonal forecasts (with access to an MWG) compared to farmers who do not use WCI in the two rounds of surveys. We focus on the three most important crops in the survey districts, i.e., groundnuts, maize, and millet. For each crop, we present statistics on the following key production variables (i) proportion of households that grew the crop; (ii) number of plots allocated; (iii) area under cultivation; (iv) productivity or yield; and (v) value of production. In the last part of the table, we sum up all the values for the three crops. Generally, we found that more than 90% of the sampled households grow groundnuts. The results show that in the second wave, households that used WCI and had access to an MWG, a significantly higher proportion grew groundnuts and allocated more land than households who did not use WCI. In the second wave, those who used WCI had a significantly higher yield of groundnuts than non-users, whereas in the first wave, non WCI users had significantly higher 32 Table 3: General differences in crop productivity among households that use seasonal forecasts (with MWG) vs. non-users First wave (2017) Second wave (2019) Non users/(SD) Users/(SD) Non users/(SD) Users/(SD) Peanuts production Planted groundnuts (dummy) 0.97 0.98 0.92 0.95∗ (0.18) (0.15) (0.27) (0.21) Number of groundnut plots 1.99 2.19 1.77 2.37∗∗∗ (1.48) (1.55) (1.61) (2.00) Area under groundnuts (ha) 4.85 4.75 3.38 4.38∗∗ (5.84) (4.60) (3.54) (5.91) Groundnuts yield (kgs/ha) 558.16 475.02∗∗ 533.22 681.77∗∗∗ (412.37) (416.47) (421.72) (394.06) Groundnut value (CFA/ha) 121352.37 103276.52∗∗ 100176.78 128119.88∗∗∗ (89655.66) (90548.10) (79611.73) (74562.16) Maize production Planted maize (dummy) 0.31 0.66∗∗∗ 0.96 0.99∗∗ (0.46) (0.47) (0.20) (0.10) Number of maize plots 1.07 1.15∗ 1.88 1.56∗∗∗ (0.25) (0.40) (1.52) (1.22) Area under maize (ha) 1.36 1.55 3.52 2.64∗∗∗ (1.36) (1.25) (4.73) (3.41) Maize yield (kgs/ha) 450.79 531.55 525.03 680.63∗∗∗ (423.38) (506.22) (372.73) (446.91) Maize value (CFA/ha) 81794.93 97748.39 93235.12 115504.41∗∗∗ (79522.42) (94990.57) (86670.77) (76677.01) Millet production Planted millet (dummy) 0.93 0.92 0.23 0.51∗∗∗ (0.25) (0.27) (0.42) (0.50) Number of millet plots 1.35 1.36 1.32 1.23 (0.74) (0.76) (0.56) (0.49) Area under millet (ha) 3.22 2.48∗∗∗ 2.93 2.52 (3.79) (2.02) (2.98) (2.87) Millet yield (kgs/ha) 433.98 438.13 545.48 714.64∗ (470.48) (401.68) (466.71) (636.93) Millet value (CFA/ha) 89362.01 85828.83 98568.06 126149.16∗ (97961.10) (79896.73) (82587.37) (113253.31) Total production Total number of plots 3.54 4.21∗∗∗ 3.86 4.46∗∗∗ (2.16) (2.18) (2.40) (2.44) Total cultivated area (ha) 8.17 8.04 7.39 8.14 (9.75) (6.48) (7.42) (9.89) Value of main crops (CFA/ha) 227915.70 248194.43 210554.08 303278.87∗∗∗ (169026.25) (175715.49) (148438.55) (178289.90) No. of cases 193 403 193 403 Standard deviations in parentheses, *** p<0.01, ** p<0.05, * p<0.1 Users: Interaction term of WCI use and access to MWG; Non-users: No uptake of WCI &/or No MWG access Official exchange rate at time of survey was ≈ 1USD:595CFA 33 yields. Interestingly, concerning maize, in the second wave, almost all households that used WCI grew maize. WCI users had significantly less land and fewer plots allocated to maize but higher yields and value than non WCI users. Concerning millet production, the yields and value of production were higher WCI users with access to an MWG than non-users. Overall, the value of the main crops produced by WCI users was higher than that produced by non-users in both waves. In the first wave, users obtained a crop value of CFA 248,194.43/ha (≈ US$ 417.13/ha) compared to CFA 227,915.70/ha (≈ US$ 383.05$/ha) obtained by non- users. Similarly, in the the second wave, users obtained a crop value of CFA 303,278.87/ha (≈ US$ 509.71/ha) compared to CFA 210,554.08/ha (≈ US$ 353.87/ha) obtained by non-users. Overall, the value of main crops increased from CFA 238,055.07/ha (≈ US$ 400.09/ha) in the first wave to CFA 256,916.48/ha (≈ US$ 431.79/ha) during the second wave. Figure 4 and 5 show the cumulative distributions (CDF) for users and non-users of using seasonal forecasts in 2017 and 2019 surveys, respectively. Starting with the 2019 season, we find that the CDF on the value of crops for farmers that used seasonal forecasts and had access to MWG dominates that of farmers that either did not use CIS or did not have access to the MWG. Based on the Kolmogorov-Smirnov test12, these results confirm that the CDF of users of seasonal forecasts stochastically dominated that of non-users at the 1% significance level. This implies that there is a significantly higher probability that, if randomly chosen, users of seasonal forecasts with access to MWG will on average obtain a higher crop value than non-users. However, in the 2016 season, this statistical dominance was not significant. The descriptive statistics in this section compare the socio-economic, institutional, and bio- physical factors of the sampled households in the two survey rounds depending on whether 12This is a non-parametric test for first-order stochastic dominance, to check for the statistical significance of this dominance 34 Figure 4: Cumulative distribution function for value of main crops in the 2016 season they used WCI and had access to an MWG. The descriptive statistics point to some system- atic differences between WCI users and non-users when disaggregated by access to an MWG. To properly estimate WCI impacts, we used panel regression models that account for both observed and unobserved heterogeneity across sampled farmers in the next section. 35 Figure 5: Cumulative distribution function for value of main crops in the 2018 season 5 Empirical results 5.1 Econometric results from panel data estimation models Here we present the results of panel regression models that analyze how the use of WCI me- diated through the MWGs affects agricultural productivity based on three major crops, maize, millet, and groundnuts. Table 4 and Table 5 show the main results of the econometric model- ing are presented in. These results are estimated based on Equation 12, where the dependent variable — the natural log of the value of crop income per hectare — is used as a proxy for 36 agricultural productivity and regressed against independent covariates that range from farmer and plot characteristics, use of major inputs and biophysical factors, asset index, extension household fixed effects and a time dummy. All continuous covariates are expressed in the natural logs, while dummy variables are not. The treatment variable of interest is the inter- action term that captures two states: (i) that a household has access to an MWG and (ii) the household used seasonal and/or daily weather forecasts to inform farm management decisions. Table 4, lists the full SLR results with covariates for the three main categories of farmers (i) the pooled sample, (ii) the sub-sample in which farmers have access to an MWG, and (iii) the sub-sample in which farmers have no access to an MWG. To compare the consistency of the results across different functional forms, we specify two models of the SLR regression, one without lagged values (SLR) and another with lagged (SLR lag) for each of the sub-samples. For the sake of brevity, Table 5 only presents coefficients for the treatment effects with vari- ous interaction effects and leaves out all the other covariates included in the full model, as in the previous table. The variables of interest in this case are the effects of using the seasonal forecasts on the total value of the three crops, which can be expressed as elasticities and are captured in rows 1 to 6. We start with the first row, which shows that farmers who used sea- sonal forecasts (onset, amount seasonal forecasts (onset, amount of rainfall or end of rainfall) and had access to an MWG obtained 10.6% and 20.7% higher total crop value than farmers who either did not use WCI or did not have access to an MWG. The second row is limited to farmers with access to an MWG and shows gains of 10.4% and 19.3% in total crop value obtained by farmers who used seasonal forecasts over those who did not. Results in the third row show that farmers that used daily weather forecasts and had acccess to an MWG obtained 9.3% and 25.4% higher total crop values (non-lagged and lagged models, respectively) than farmers that either did not use WCI or did not have access to an MWG. The 37 Table 4: Estimates of the impact of WCI use from MWG on crop incomes: full SLR results Dep. var: Value of crop income CFA per ha) Pooled sample With MWG No MWG SLR SLR (lag) SLR SLR (lag) SLR Use of seasonal forecast x: MWG 0.106* 0.207*** - - - (0.0583) (0.0717) - - - Use of seasonal forecast - - 0.104* 0.193*** -0.0312 - - (0.0588) (0.0705) (0.0772) Male household head (dummy) 0.242*** 0.269** 0.189* 0.142 0.287* (0.0929) (0.124) (0.105) (0.140) (0.161) Education of household head (years) -0.0388 -0.0331 -0.0284 -0.0260 -0.0689 (0.0255) (0.0316) (0.0309) (0.0346) (0.0424) Age of household head (years) 0.0328 0.0161 0.0990 0.0358 -0.123 (0.0796) (0.0993) (0.0944) (0.114) (0.145) Full time farming (dummy) 0.113** 0.0388 0.133* 0.0274 0.0925 (0.0522) (0.0615) (0.0701) (0.0757) (0.0757) Area in Ha 0.00236 0.0373 -0.00896 0.0549 0.0170 (0.0456) (0.0592) (0.0585) (0.0780) (0.0711) HPA index 0.151*** 0.124*** 0.159*** 0.145*** 0.113* (0.0338) (0.0401) (0.0386) (0.0462) (0.0607) Tropical livestock units 0.0935*** 0.0688* 0.105*** 0.0996** 0.0759 (0.0277) (0.0358) (0.0314) (0.0390) (0.0527) Access to extension (dummy) 0.00580 0.0462 0.0137 0.110 -0.0570 (0.0478) (0.0666) (0.0531) (0.0728) (0.0991) Distance to main road (km) -0.0360 -0.0945** -0.0347 -0.117** -0.0135 (0.0269) (0.0397) (0.0325) (0.0507) (0.0505) Use of improved seed (dummy) -0.0473 -0.0743 0.0227 -0.0209 -0.165* (0.0439) (0.0557) (0.0503) (0.0617) (0.0858) Use of fertilizer (dummy) 0.310*** 0.304*** 0.392*** 0.349*** 0.137* (0.0473) (0.0609) (0.0583) (0.0741) (0.0758) Use of manure (dummy) 0.208*** 0.152*** 0.177*** 0.118* 0.331*** (0.0414) (0.0552) (0.0508) (0.0670) (0.0686) Rainfall (mm) 1.331*** - 1.350*** - 3.920 (0.414) - (0.417) - (4.050) Temperature (◦c) 12.75** - 0.557 - -3.740 (5.529) - (0.813) - (7.459) Lagged rainfall (mm) - 1.233*** - 1.370*** - - (0.436) - (0.438) - Lagged temperature (◦c) - 6.243 - 0.713 - - (6.510) - (0.854) . - Altitude (m) 0.0347 -0.222 0.00619 -0.0738 0.0289 (0.0387) (0.220) (0.0417) (0.0455) (0.244) Slope (◦) -0.0616 -0.101 -0.129** -0.248*** 0.103 (0.0589) (0.158) (0.0639) (0.0722) (0.128) Soil organic carbon (g/kg) 0.102 -3.701 0.133 0.392** -3.648 (0.193) (5.098) (0.194) (0.196) (4.133) Observations 1,166 38 1,166 787 787 379 Time dummy Yes No Yes No Yes Dependent variable: Value of crop income in CFA per Ha for all models x: shows interaction term; All continuous covariates are expressed in their natural logs. Standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1 fourth row shows a similar trend, with the use of daily weather forecasts associated with 9.9% and 24.2% higher total crop values (non-lagged and lagged models, respectively) for farmers with access to an MWG. The fifth row shows that the combined use of seasonal and daily weather forecasts in the presence of the MWG led to 12.5% and 25% increase in total crop value (non-lagged and lagged models, respectively) compared to farmers that either did not use WCI or did not have access to an MWG. The gains were similar for farmers with access to an MWG, and the use of both daily and seasonal weather forecasts was associated with respective increases of 12.4% and 23.6% in the total value of crops. Interestingly, the use of seasonal and daily weather forecasts did not have a significant impact on the total crop value, as shown in all results in column 5. In this section, we use quantitative evidence to show that the use of weather and climate infor- mation generally leads to higher crop productivity. As noted in Section 3, when done properly, the advantage of using quantitative methods is that besides being used to establish causal rela- tions, the results can also be generalized to larger areas or to populations in similar contexts. However, explaining how the pathways occur that lead to these impacts is often a knotty prob- lem, particularly for complex interventions such as WCIs. To provide a more nuanced and fuller understanding of how these impacts occur, we complete these estimates with participa- tory impact pathways as described in section 3.2.2. 5.2 Participatory pathways linking the use of WCI, MWG and livelihoods Figures 6 and 7 show the participatory impact pathways for farmers that use WCI and have access to an MWG and those that do not use WCI whether they have access to the MWG or not, respectively. As can be seen, the trajectories of these impacts are very different. As 39 Table 5: Estimates of the impact of WCI use from MWG on crop incomes: abridged SLR results Dep. var: Value of crop income CFA per ha Pooled sample With MWG Without MWG SLR SLR (lag) SLR SLR (lag) SLR Use of seasonal forecasts x: MWG 0.106* 0.207*** - - - (0.0583) (0.0717) - - - Use of seasonal forecasts - - 0.104* 0.193*** -0.0312 - - (0.0588) (0.0705) (0.0772) Use of daily forecasts x: MWG 0.0929* 0.254*** - - - (0.0555) (0.0700) - - - Use of daily forecasts - - 0.0987* 0.242*** -0.0364 - - (0.0559) (0.0692) (0.0772) Use of seasonal x: daily x: MWG 0.125** 0.253*** - - - (0.0717) (0.0868) - - - Use of seasonal x: daily - - 0.124** 0.236*** -0.0433 - - (0.0506) (0.0629) (0.0819) Observations 1,166 1,166 787 787 379 Household characteristics Yes Yes Yes Yes Yes Farm characteristics Yes Yes Yes Yes Yes Institutional factors Yes Yes Yes Yes Yes Biophysical factors Yes Yes Yes Yes Yes Time dummy Yes No Yes No Yes Dependent variable: Value of crop income in CFA per Ha for all models x: shows interaction term; All continuous covariates are expressed in their natural logs. Standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1 shown in Figure 6, for WCI users, access to seasonal forecasts mediated by an MWG plays a key role in agronomic planning decisions (species, varieties, land allocation, inputs), but also in investment decisions including whether to source for credit. Farmer promoters play a key role in translating the information received from the MWG or through SMS/voice calls to farmers, interpreting in their own words what the WCI means and guiding them on the optimal farm management responses. The link between the farmer promoters and the MWGs also plays a role as the promoters relay the recommendations included in the bulletins. For instance, if seasonal forecasts announce a "bad" season, meaning late onset or a long break in the rainy season, and farmers have 40 Output Outcome Impact 1 Impact 2 Stakeholders use the The outcomes generate impacts The outcomes further generate impacts Products of the intervention products of the intervention for the actors directly or indirectly for other actors (scaling out, up, spillovers) to change practices and interacting with the intervention behaviors outputs Increased trust and recognition: women Change in mentality: promoters become focal people are looking for points in their communities weather information Workshops, Network of Promoters share climate Increased reputation and Multiplication of the Increased security of training of promoters information and recognition of the central role number of farmers using recommendations populations against climatic promoters of promoters in their forecasts to make decisions events communities (scaling out) Farmers understand climate information and see results of Demonstration taking into account forecasts in Reduction of seed and plots their agricultural practices crop losses Women promoters and farmers Trainings Women promoters use have increased capacity to adapt Cultivation on lower quality seeds and their decisions based on the climate areas with higher yieldsrecommendations information Increased yields Farmers can plan better the cultural calendar: weeding, Ten day forecast clearing, inputs ex. Dry spell Farmers change the harvesting date Increased or stable Ten day forecast: Farmers change the type of ex. end of rainy post-harvest storage income and Bulletins, SMS, consumptionseason radio, voice Farmers use the most Increased availability calls, promoters appropriate variety: e.g. Increased Possible decrease of of products for the cooperative buys short- production selling prices processing units, at cycle ones better prices Farmers change the variety Farmers sell the seeds or Reduced production of groundnuts or millet (eg process them to buy other costs short cycle or long cycle) varietes Farmers make different investment Potential risk increase decisions (inputs, labor) because of the use of short- cycle varieties Seasonal Farmers take different forecast indebtedness decisions Fewer people lose their Rationalization of assets to financial Herders mover the herd or change indebtedness institutions the date to start moving the herd Adaptation of hay rations Seasonal forecast: ex. Unfavorable Use of fields near the house, rainy season reduction of cultivated areas Legend Seasonal forecast: ex. Investment in fewer crops, Negative impacts Favorable rainy more profitable, on larger season surfaces Elements mentioned both by men and women Figure 6: Participatory pathways for users of WCIS the resources to do so, they can switch to shorter cycle varieties of maize and groundnuts or apply for credit in a more rational way. This leads to better indebtedness decisions and investments, higher yields linked to the use of more appropriate varieties, and ultimately to increased availability of harvest for their own consumption or for sale. If farmers’ resources are more limited, they might prefer long-cycle varieties or cultivate fields located closer to their homestead, reduce their planted area and increase crop diversity, favor sorghum and cowpea over commercial crops, improve food availability through self-consumption. A key element repeatedly stressed in the focus group discussions was that better planning of 41 productive activities also reduces the total workload. Moreover, an indirect benefit of WCI is that livestock owners used it in their decisions concerning animal movement and protec- tion, increasing animal security and protecting their sources of income. However, negative impacts are possible, as identified in particular by farmers with long term access to the MWGs (Kaffrine): when the harvest is good and many farmers bring their products to the market, the sales price falls. While this can have negative effects on their revenues, the farmers acknowl- edge that it can have a favorable effect on the prices charged by processing units, who can lower their prices. The farmers also identified potentially higher risks associated with the use of short cycle varieties: some less risk averse farmers may seek bigger loans to invest in short cycle varieties planted in larger areas, and use more chemical inputs. This can have negative consequences if the seasonal forecast proves to be wrong. Characteristics of the season Decisions Consequences The season is bad: ex. Floods after the dry spell dry spell at the can decimate livestock beginning of the season They can decide to use different soils (deck, dhior) but it depends Stable production on the availability of land They use more fertilizer Increased production costs If production fails, they look for to obtain good yields masonry work, petty commerce in urban areas, they go fishing Some decide to reseed several times If the harvest fails, women If production fails, those who are switch to petty trade, linked to a cooperative use the animals, garden vegetables agricultural insurance Some change crops if first sawn crops fail If peanuts fail, men switch to cowpea, watermelon, Changes in human diet: choice sorghum of less preferred foods They reduce the quantity sown and the areas Variable income and cultivated availability of food Changes in human diet: choice of foods of inferior quality Traditional indicators They do not change the date of show signs it will be a sowing because "the first sowing bad season is always the best" Some households decide Increased indebtedness to take credit to buy more fertilizer Livestock owners buy Increased costs of animal supplements for animals production The season is bad: ex. rainfall deficit Water needs to be found Increased drudgery further Legend They apply more fertilizer Increased yieldsThe season is good Negative consequences Figure 7: Participatory pathways for non-users of WCIS Overall, the participatory impact pathway support the interpretation that farmers that use WCI, in particular when mediated by an active and longstanding MWG like the one in Kaffrine, shift from reactive to proactive management, enabling better risk management and adaptation. 42 This shift contributes to the efficient use of agricultural inputs thereby reducing production costs and human and animal workloads. Daily weather forecasts and early warnings are also important as household members, in particular women, can adapt their chores and hence reduce their workload. Daily weather forecasts and early warnings are also important for the safety of fishermen and livestock herders in particular, who can decide not to go out to sea or to move their livestock. The security of children is improved, as they are not allowed to play too far away from their home when an extreme event is announced, and household members can better to decide whether or not to travel or go to market. At another level, both male and female farmer promoters benefit from increased visibility and leadership in their communities. Moreover, their acquired skills in interpreting forecasts and their leadership role in transmitting WCI allows them to share their results with other farmers in their communities and with local agricultural services, thereby increasing use by other farmers. The role of promoter farmers therefore appears particularly important in both areas with and without an MWG, farmers mentioned the need to increase the number of promoters. Indirectly, this reveals, the importance of the network of transmitters of information. The MWGs are built on networking, coordination, and transmission principles. The trust in farmer promoters, and in general in the sources of climate and weather information, was confirmed here as a key factor that drives use. Finally, we were also able to identify the causal logic of decisions taken by farmers who do not use WCIs either because they do not trust them or do not have access to the mediating role of an MWG. Figure 7 shows this causal logic. Farmers use traditional signs to plan their productive activities. If they consider the season will be "bad", with low rainfall or late onset, they will still plant at their usual planting date, but will reduce the area planted or ask for a loan if they can. During a "bad" season if their resources allow them to do so, they will sow more than once. Men said that if groundnuts fail, they might switch to other crops such as 43 sorghum, while women will focus on gardening and petty commerce. The strategies they use may support production stability even though its direct costs (e.g., chemical inputs, credit interests) and indirect (e.g., human labor) costs may be higher. Comparing WCI users and non-users, the key difference that WCI appears to make is that planning decisions are made rationally which reduces production costs, drudgery, and improves risk management. 6 Discussion and conclusions The provision of tailored climate information services is increasingly recognized as an im- portant component in helping decision makers adapt to and mitigate the effects of climate change and climate variability. However, there is a dearth of empirical evidence showing the contribution of contribution of WCIs to farmers’ livelihoods. This study helps fill this gap by analyzing the impact of weather and climate information co-produced under the MWG model on the livelihoods of smallholder farmers in Senegal. Most studies in extant literature use ‘snap-shot’ analysis, i.e, limited to one period, and focus on outcomes (awareness, access, uptake, and behavioral changes) but do not control for selection bias. The present study used data collected over two agricultural seasons from a balanced sample of 595 households in Senegal, and applies a panel data approach to estimate impacts on farmers’ livelihoods to account for selection bias. We used a unique mixed-methods research design that comple- mented the quantitative approach with the participatory characterization of the how WCI is used and how its use contributes to farmers’ livelihoods from the farmers’ perspective, en- abling a deeper understanding of the causal linkages between the use of WCI, the MWG and farmers’ livelihoods. 44 Quantitative analyses confirmed that when disseminated or mediated through the MWG model, the use of weather and climate information affects farmers’ production choices and is associated with improved livelihoods. More specifically, the use of seasonal and daily fore- casts increased the average income of farmers with access to the MWG by between 10-25%, depending on the functional forms used in the regression models. These findings are sup- ported by the results of qualitative data which show that the use of seasonal and 10-day fore- casts improves crop production activities, particularly step-by-step cropping plans: from land preparation, the choice of variety, planting dates, and harvest to farm management decisions. Beyond this, they contribute to more rational indebtedness decisions and more informed in- vestments. They also contribute to human and livestock security. While providing actionable information to farmers, promoter farmers gain visibility and a reputation within their com- munities. The link between the existence of an MWG and farmers’ use of WCI appears to be mediated by the action of the farmer promoters and the network of MWG members. Having access to farmer promoters and being able to interpret the information, either through training or through support from local farmer promoters, technical services, and others, appears to play a key role in building farmers’ confidence in and use of WCIs. Other qualitative studies have found similar uses and impacts of WCI in terms of increased security for populations, but also of building visibility for key figures connected to the transmission of the information in the communities (e.g., Chaudhuri and Kendall, 2021). Two broad lessons can be learned from the results of this study. First, coordinated platforms, like the MWG co-production model, bring diverse stakeholders together and gets them to work together towards achieving a shared vision, and context-specific solutions can be instru- mental in tackling climate change and local adaptation, directly contributing to SDG 13. The availability and dynamism of networks and farmers’ confidence in the sources of informa- tion, particularly through training and dissemination events, appear to be important factors 45 in CIS use. Second, the evidence showing that the provision of tailored WCI leads farmers to make better informed decisions is associated with improved crop productivity and income, which can consequently help lower poverty levels (SDG 1), reduce the incidence of hunger and improve nutrition (SDG 2). We used an innovative evaluation design for this study that improved the robustness of our findings. However, some limitations need to be mentioned. First, while we used panel data based on two agricultural seasons, this period might not be long enough to capture year to year weather variability. Vaughan et al., 2019 contends that a considerable number of years are required to sample the range of weather variability and these usually exceed a typical project cycle. Second, farmers in the control group may have been directly or indirectly ex- posed to downscaled WCI and agricultural advisories from the MWG program. As noted in Chiputwa et al., 2020, WCI and other climate-related knowledge exhibit characteristics of pub- lic goods and are thus more often disseminated through public means such as national radio, television and extension services, which make it difficult to have a pure control group that will not have access to the treatment. The resulting spill-over effects may reduce the estimated treatment effects. Third, we analyzed the effectiveness of WCIs co-produced by the MWG in Kaffrine, Senegal. However, differences in social, economic, and institutional factors across different contexts may mean our findings do not apply in a different setting. The findings of this study should thus be interpreted with caution and confirmed through further research. 46 7 Acknowledgements This work was funded under the 2018 call for ex-post impact assessment (ep-IA) of climate change related work in the CGIAR. The call was a competitive bid funded by the CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS) with the aim of providing robust impact evaluations of climate change related work. The work was carried out with support from CGIAR Fund Donors and through bilateral funding agreements. For de- tails see https://ccafs.cgiar.org. Part of this work was also supported by USAID through grant number USAID Contract No: AID-OAA-A-16-00072 under the Climate Information Services Research Initiative (CISRI), A Learning Agenda for Climate Information Services in sub- Saharan Africa. We are grateful to the Meteorological Department in Senegal (ANACIM) for their assistance, to the team of data collectors led by Adama Ba and Souleymane Jules Diol, and to all the respondents in our interviews. 47 References Abrams, M., Crippen, R., & Fujisada, H. (2020). ASTER Global Digital Elevation Model (GDEM) and ASTER Global Water Body Dataset (ASTWBD). Remote Sensing, 12(7), 1–12. https://doi.org/10.3390/rs12071156 Adegbola, P., & Gardebroek, C. (2007). The effect of information sources on technology adoption and modification decisions. Agricultural Economics, 37(1), 55–65. https : //doi.org/10.1111/j.1574-0862.2007.00222.x AGRA. (2014). African agriculture sta-tus report: Climate change and smallholder agriculture in sub-Sahara Africa. https://www-cif.climateinvestmentfunds.org/sites/default/files/ aasr-2014climate-change-and-smallholder-agriculture-in-ssa.pdf{\%}0A Alfani, F., Dabalen, A., Fisker, P., & Molini, V. (2015). Can we measure Resilience? A pro- posed method and Evidence from Countries in the Sahel.World Bank: Policy Re- search Working Paper, (January), 1–28 NULL. Amare, M., Jensen, N. D., Shiferaw, B., & Cissé, J. D. (2018). Rainfall shocks and agricul- tural productivity: Implication for rural household consumption. Agricultural Systems, 166(July), 79–89. https://doi.org/10.1016/j.agsy.2018.07.014 Amegnaglo, C. J., Anaman, K. A., Mensah-Bonsu, A., Onumah, E. E., & Amoussouga Gero, F. (2017). Contingent valuation study of the benefits of seasonal climate forecasts for maize farmers in the Republic of Benin, West Africa. Climate Services, 6, 1–11. https://doi.org/10.1016/j.cliser.2017.06.007 Anuga, S. W., & Gordon, C. (2016). Adoption of climate-smart weather practices among smallholder food crop farmers in the Techiman municipal : Implication for crop yield. 5(9), 279–286. Björkman-Nyqvist, M. (2013). Income shocks and gender gaps in education: Evidence from Uganda. Journal of Development Economics, 105, 237–253. https://doi.org/10.1016/J. JDEVECO.2013.07.013 Blundo-Canto, G., Andrieu, N., Soule-Adam, N., Ndiaye, O., & Chiputwa, B. (2021). Scaling Weather and Climate information Services for agriculture: evaluating systemic but overlooked effects. Climate Services, 22(100216), 100216. https://doi.org/10.1016/j. cliser.2021.100216 Bruno Soares, M., Daly, M., & Dessai, S. (2018). Assessing the value of seasonal climate forecasts for decision-making.Wiley Interdisciplinary Reviews: Climate Change, (February 2017), 1–19. https://doi.org/10.1002/wcc.523 Carr, E. R. (2013). Livelihoods as Intimate Government: Reframing the logic of livelihoods for development. Third World Quarterly, 34(1), 77–108. https://doi.org/10.1080/ 01436597.2012.755012 Carr, E. R. (2014). From description to explanation: Using the Livelihoods as Intimate Gov- ernment (LIG) approach. Applied Geography, 52, 110–122. https://doi.org/10.1016/j. apgeog.2014.04.012 48 Cash, D. W., Clark, W. C., Alcock, F., Dickson, N. M., Eckley, N., Guston, D. H., Jäger, J., & Mitchell, R. B. (2003). Knowledge systems for sustainable development. Proceedings of the National Academy of Sciences of the United States of America, 100(14), 8086– 8091. https://doi.org/10.1073/pnas.1231332100 CCAFS. (2015). The impact of climate information services in Senegal, CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS). www.ccafs. cgiar.org Chaudhuri, B., & Kendall, L. (2021). Collaboration without consensus: Building resilience in sustainable agriculture through ICTs. Information Society, 37(1), 1–19. https://doi.org/ 10.1080/01972243.2020.1844828 Chiputwa, B., Wainaina, P., Makui, P., Nakelse, T., Zougmoré, R., & Ndiaye, O. (2019). Eval- uating the Impact of the Multidisciplinary Working Group Model on Farmers’ Use of Climate Information Services in Senegal. (May). https : / / www. climatelinks . org / sites / default / files / asset / document / 2019{\ _ }USAID{\ _ }MercyCorps{\ _ }EvaluatingtheImpactoftheMultidisciplinaryWorkingGroupModelonFarmers{\% }27UseofCISinSenegal.pdf Chiputwa, B., Wainaina, P., Nakelse, T., Makui, P., Zougmore, R. B., Ndiaye, O., & Minang, P. A. (2020). Transforming climate science into usable services : The effectiveness of co-production in promoting uptake of climate information by smallholder farmers in Senegal. Climate Services, 20(100203). D’Alessandro, S., Fall, A. A., Grey, G., Simpkin, S., & Wane, A. (2015). Senegal: Agricul- tural Sector Risk Assessment, World Bank Group. www.worldbank.org Dilling, L., & Lemos, M. C. (2011). Creating usable science: Opportunities and constraints for climate knowledge use and their implications for science policy. Global Environ- mental Change, 21(2), 680–689. https://doi.org/10.1016/j.gloenvcha.2010.11.006 European Commission. (2015). A European research and innovation roadmap for climate services. https://op.europa.eu/en/publication-detail/-/publication/73d73b26-4a3c- 4c55-bd50-54fd22752a39 FAO. (2015). Global Forest Resources Assessment 2015 - Desk reference. Fischer, G., Shah, M., Tubiello, F., & Van Velhuizen, H. (2005). Socio-economic and climate change impacts on agriculture: an integrated assessment, 1990–2080. Philosophical Transactions of the Royal Society of London B: Biological Sciences, 360(1463), 2067– 2083. Funk, C., Verdin, A., Michaelsen, J., Peterson, P., Pedreros, D., & Husak, G. (2015). A global satellite-assisted precipitation climatology. Earth System Science Data. https://doi.org/ 10.5194/essd-7-275-2015 Gelaro, R., McCarty, W., Suárez, M. J., Todling, R., Molod, A., Takacs, L., Randles, C. A., Darmenov, A., Bosilovich, M. G., Reichle, R., Wargan, K., Coy, L., Cullather, R., Draper, C., Akella, S., Buchard, V., Conaty, A., da Silva, A. M., Gu, W., . . . Zhao, B. (2017). The modern-era retrospective analysis for research and applications, version 2 (MERRA-2). Journal of Climate. https://doi.org/10.1175/JCLI-D-16-0758.1 Griggs, D., Stafford-smith, M., Warrilow, D., Street, R., Vera, C., Scobie, M., & Sokona, Y. (2021). Use of weather and climate information essential for SDG implementation. 49 Nature Reviews Earth & Environment, 2(January), 2–4. https: / /doi .org/10.1038/ s43017-020-00126-8 Hansen, J. W. (2005). Integrating seasonal climate prediction and agricultural models for insights into agricultural practice. Philosophical Transactions of the Royal Society B: Biological Sciences, 360(1463), 2037–2047. https://doi.org/10.1098/rstb.2005.1747 Hansen, J. W., Mason, S. J., Sun, L., & Tall, A. (2011). Review of seasonal climate forecast- ing for agriculture in sub-Saharan Africa. https://doi.org/10.1017/S0014479710000876 Hengl, T., Jesus, J. M. D., Heuvelink, G. B. M., Ruiperez, M., Kilibarda, M., Blagoti, A., Shangguan, W., Wright, M. N., Geng, X., Bauer-marschallinger, B., Guevara, M. A., Vargas, R., Macmillan, R. A., Batjes, N. H., Leenaars, J. G. B., Ribeiro, E., Wheeler, I., Mantel, S., & Kempen, B. (2017). SoilGrids250m : Global gridded soil information based on machine learning. PLoS ONE, 12(2), 1–40. https://doi.org/10.1371/journal. pone.0169748 Imbens, G. W., & Wooldridge, J. M. (2009). Recent Developments in the Econometrics of Program Evaluation. Journal of Economic Literature, 47(1), 5–86. https://doi.org/10. 1257/jel.47.1.5 Imbens, G. (2000). The role of the propensity score in estimating dose-response functions. Biometrika, 87(3), 706–710. http://biomet.oxfordjournals.org/content/87/3/706.short IPCC. (2014). Climate Change 2014: Mitigation of Climate Change. https://doi.org/10.1017/ CBO9781107415416 ISRIC. (2018). SoilGrids: an automated system for global soil mapping. Kripfganz, S., & Schwarz, C. (2019). Estimation of linear dynamic panel data models with time-invariant regressors. Journal of Applied Econometrics. https://doi.org/10.1002/ jae.2681 Lemos, M. C., & Morehouse, B. J. (2005). The co-production of science and policy in in- tegrated climate assessments. Global Environmental Change, 15(1), 57–68. https : //doi.org/10.1016/j.gloenvcha.2004.09.004 Lo, H. M., & Dieng, M. (2015). Impact assessment of communicating seasonal climate fore- casts in Kaffrine, Diourbel, Louga, Thies and Fatick (Niakhar) regions in Senegal (March). Loayza, J. R., Toure, E., & Niane, A. (2018). Senegal Economic Update: Recent Growth Drivers in Senegal , and the Role of Agriculture in Developing a Resilient and Inclu- sive Economy, World Bank Group. Machingura, F., Nyamwanza, A., Hulme, D., & Stuart, E. (2018). Climate information ser- vices, integrated knowledge systems and the 2030 agenda for sustainable development. Sustainable Earth, 1(1), 1–7. https://doi.org/10.1186/s42055-018-0006-1 Mudombi, S., & Nhamo, G. (2017). Access to Weather Forecasting and Early Warning Infor- mation by Communal Farmers in Seke and Murewa Districts, Zimbabwe. Journal of Human Ecology, 48(3), 357–366. Ndiaye, O., Moussa, A., Seck, M., Zougmore, R., & Hansen, J. (2013). Communicating sea- sonal forecasts to farmers in Kaffrine, Senegal for better agricultural management. Hunger, Nutrition and Climate justice, 4. http://cgspace.cgiar.org/handle/10568/27888 50 Njuki, J., Kaaria, S., Chamunorwa, A., & Chiuri, W. (2011). Linking smallholder farmers to markets, gender and intra-household dynamics: Does the choice of commodity matter? European Journal of Development Research, 23(3), 426–443. http://www.palgrave- journals.com/ejdr/journal/v23/n3/abs/ejdr20118a.html O’Brien, K., Sygna, L., Næss, L. O., Kingamkono, R., & Hochobeb, B. (2000). Is Information Enough ? User Responses to Seasonal Climate Forecasts in Southern Africa, Center for International Climate and Environmental Research. OECD. (2016). OECD Policy Instruments for the Environment (PINE) database, 1–6. http: //oe.cd/pine Ouédraogo, M., Barry, S., Zougmoré, R. B., Partey, S. T., Somé, L., & Baki, G. (2018). Farm- ers’ willingness to pay for climate information services: Evidence from Cowpea and Sesame producers in Northern Burkina Faso. Sustainability (Switzerland), 10(3). https: //doi.org/10.3390/su10030611 Ouédraogo, M., Zougmoré, R., Barry, S., Somé, L., Grégoire, B., & 2015, S. (2015). The value and benefits of using seasonal climate forecasts in agriculture : evidence from cowpea and sesame sectors in climate-smart villages of Burkina Faso. CCAFS Info Note, (September), 1–4. https://doi.org/10.13140/RG.2.1.4958.5682 Patt, A., Suarez, P., & Gwata, C. (2005). Effects of seasonal climate forecasts and partici- patory workshops among subsistence farmers in Zimbabwe. Proceedings of the Na- tional Academy of Sciences, 102(35), 12623–12628. https://doi.org/10.1073/pnas. 0506125102 Rao, K. P. C., Hansen, J, Njiru, E, Githungo, W. N., & Oyoo, A. (2015). Impacts of seasonal climate communication strategies on farm management and livelihoods in Wote, Kenya. CCAFS Working Paper, (No.137), 42 pp. Rasmussen, J., Andersen, A., Fisker, A. B., Ravn, H., Sodemann, M., Rodrigues, A., Benn, C. S., & Aaby, P. (2012). Mid-upper-arm-circumference and mid-upper-arm circumfer- ence z-score: The best predictor of mortality. European Journal of Clinical Nutrition, 66(9), 998–1003. https://doi.org/10.1038/ejcn.2012.95 Roncoli, C. (2006). Ethnographic and participatory approaches to research on farmers' responses to climate predictions. Climate Research, 33, 81–99. https://doi.org/10. 3354/cr033081 Roncoli, C., Jost, C., Kirshen, P., Sanon, M., Ingram, K. T., Woodin, M., Somé, L., Ouat- tara, F., Sanfo, B. J., Sia, C., Yaka, P., & Hoogenboom, G. (2009). From accessing to assessing forecasts: An end-to-end study of participatory climate forecast dissem- ination in Burkina Faso (West Africa). Climatic Change, 92(3-4), 433–460. https : //doi.org/10.1007/s10584-008-9445-6 Roudier, P., Muller, B., D’Aquino, P., Roncoli, C., Soumaré, M. A., Batté, L., & Sultan, B. (2014). The role of climate forecasts in smallholder agriculture: Lessons from partici- patory research in two communities in Senegal. Climate Risk Management, 2, 42–55. https://doi.org/10.1016/j.crm.2014.02.001 Rubin, D. (1974). Estimating causal effects of treatments in randomized and nonrandomized studies. Journal of educational Psychology, 66(5), 688. http : / / psycnet . apa . org / journals/edu/66/5/688/ 51 Shiferaw, B., Kebede, T., Kassie, M., & Fisher, M. (2015). Market imperfections, access to information and technology adoption in Uganda: Challenges of overcoming multiple constraints. Agricultural Economics (United Kingdom), 46(4), 475–488. https://doi. org/10.1111/agec.12175 Singh, C., Daron, J., Bazaz, A., Ziervogel, G., Spear, D., Krishnaswamy, J., Zaroug, M., & Kituyi, E. (2017). The utility of weather and climate information for adaptation decision-making: current uses and future prospects in Africa and India. Climate and Development, 0(0), 1–17. https://doi.org/10.1080/17565529.2017.1318744 Stats4SD. (2017). Evaluation of Climate Services Interventions in the GFCS Adaptation Programme for Africa (October), World Food Programme, CGIAR Research Program on Climate Change, Agriculture, and Food Security (CCAFS). Tall, A., Coulibaly, J. Y., & Diop, M. (2018). Do climate services make a difference? A review of evaluation methodologies and practices to assess the value of climate information services for farmers: Implications for Africa. Climate Services, (July 2017), 0–1. https: //doi.org/10.1016/j.cliser.2018.06.001 Tarhule, A., & Lamb, P. J. (2003). Climate research and seasonal forecasting for West Africans. Bulletin of the American Meteorological Society, 84(12), 1741–1759. https://doi.org/ 10.1175/BAMS-84-12-1741 Vaughan, C., & Dessai, S. (2014). Climate services for society: Origins, institutional arrange- ments, and design elements for an evaluation framework.Wiley Interdisciplinary Reviews: Climate Change, 5(5), 587–603. https://doi.org/10.1002/wcc.290 Vaughan, C., Hansen, J., Roudier, P., Watkiss, P., & Carr, E. (2019). Evaluating agricultural weather and climate services in Africa: Evidence, methods, and a learning agenda. Wiley Interdisciplinary Reviews: Climate Change, 586. https://doi.org/10.1002/wcc. 586 WMO. (2017). White Paper on the Contribution of the Global Framework for Climate Ser- vices to Transforming our World : the 2030 Agenda for Sustainable Development ( Agenda 2030 ). Wooldridge, J. M. (2010). Econometric Analysis of Cross Section and Panel Data (Second Edi). The MIT Press. https://doi.org/10.1515/humr.2003.021 52 8 Appendix Output Outcome Impact 1 Impact 2 Products of the intervention Stakeholders use the products The outcomes generate impacts The outcomes further generate impacts of the intervention to change for the actors directly or indirectly for other actors (scaling out, up, spillovers) practices and behaviors interacting with the intervention outputs Capacity building Network of farmer Farmer promoters share climate Increased reputation and events for farmer Multiplication of farmers relais information and recommendations recognition of the role of farmer promoters using forecasts to make promoters in their communities decisions (scaling out) Capacity building Farmers understand climate information and events for observe the results of using the forecasts to farmers take management decisions Demonstration Farmers apply proactive adaptation plots strategies, Farmers make better informed Reduced credit decisions Seasonal forecast indebtedness Livestock producers change the transhumance date Improved animal Farmers program their survival crop pattern Eventual reduction Better prices in marketing prices for processing Communication Farmers change the type of units channels: SMS, variety: ex. short cycle radio, calls from Improved farmer promoters Ex. Seasonal Selling or exchange of long cycle Reduced investment yields forecast is negative varieties or processing costs (inputs, Increased Increased Farmers privilege fields closer to the weeding) consumption and food house reduce cultivated areas, Potential increase in marketing availability favoring sorghum, niebé crop diversity Ex. Seasonal Potential decrease in forecast is Investment in fewer crops, more crop diversity favorable profitable, on larger surfaces Farmers adapt their crop calendar: Three day or ten clearing, weeding, use of inputs Reduced workload day forecast: ex. Dry spell Livestock producers adapt the fodder stock and/or the daily Three or ten day rations Farmers choose the most forecast: ex. Rain after the end of appropriate harvesting date Reduced post- harvest losses Legend the season Farmers take post-harvest storage Negative impacts decision according to rain forecasts Daily forecasts, early warnings Women change their daily chores Reduced household Potential negative decisions workload for women impacts Herdsman (especially children) and Increased security Increased human Potential positive fishermen better plan their impacts of family members securitymovements Figure 8: Participatory pathways for non-users of WCIS 53 Table A1: Estimates of the impact of access to an MWG and use of WCI on agricultural productivity using different models Dep. var: Value of crop income CFA per ha) Pooled sample Exposure to MWG Pooled OLS BE FE RE RE (With MWG) RE (No MWG) Use of both seasonal x: daily x: MWG 0.130** 0.161** 0.0453 0.114** - - (0.0515) (0.0784) (0.0692) (0.0513) - - Use of seasonal x: daily - - - - 0.125** -0.0284 - - - - (0.0494) (0.0850) Male household head (dummy) 0.237*** 0.112 0.582*** 0.270*** 0.184 0.362** (0.0895) (0.112) (0.180) (0.0945) (0.113) (0.165) Education of household head (years) -0.0416* -0.0308 -0.111 -0.0421 -0.0316 -0.0805 (0.0236) (0.0273) (0.0827) (0.0257) (0.0281) (0.0539) Age of household head (years) 0.0310 0.00569 -0.292 0.0203 0.0991 -0.179 (0.0768) (0.0912) (0.217) (0.0830) (0.0921) (0.171) Full time farming (dummy) 0.114** 0.117 0.136* 0.118** 0.142** 0.0858 (0.0533) (0.0800) (0.0718) (0.0531) (0.0659) (0.0913) Area (Ha) 0.00357 0.0272 -0.162** -0.00791 -0.0146 -0.00476 (0.0415) (0.0569) (0.0681) (0.0425) (0.0522) (0.0729) HPA index 0.151*** 0.169*** 0.0844* 0.144*** 0.153*** 0.110** (0.0292) (0.0410) (0.0432) (0.0295) (0.0352) (0.0528) Tropical Livestock unit 0.0929*** 0.0972** 0.0397 0.0958*** 0.110*** 0.0695 (0.0303) (0.0393) (0.0567) (0.0316) (0.0372) (0.0579) Access to extension (dummy) -0.00835 -0.0204 0.0338 -0.00177 -0.00201 -0.0216 (0.0492) (0.0750) (0.0661) (0.0491) (0.0532) (0.113) Distance to road -0.0353 -0.0625 -0.00923 -0.0305 -0.0339 -0.0154 (0.0280) (0.0420) (0.0378) (0.0279) (0.0324) (0.0548) Use of improved seed (dummy) -0.0540 -0.00143 -0.0482 -0.0581 0.0198 -0.173* (0.0465) (0.0706) (0.0623) (0.0462) (0.0530) (0.0933) Use of chemical fertilizers (dummy) 0.305*** 0.336*** 0.254*** 0.302*** 0.381*** 0.174* (0.0482) (0.0681) (0.0714) (0.0489) (0.0561) (0.0949) Use of manure (dummy) 0.210*** 0.259*** 0.162*** 0.200*** 0.176*** 0.292*** (0.0432) (0.0628) (0.0611) (0.0435) (0.0511) (0.0801) Rainfall (mm) 0.919* 1.177** 4.185*** 0.820 1.061** 3.884 (0.500) (0.578) (1.320) (0.544) (0.515) (4.360) Temperature (◦c) 7.079 12.04 -34.96*** 4.983 -31.28*** -18.69 (6.392) (7.580) (6.782) (6.924) (4.231) (32.08) Altitude (m) 0.0679 0.0610 - 0.0769 0.0281 0.0484 (0.0564) (0.0640) - (0.0622) (0.0603) (0.353) Slope (◦) -0.0538 -0.0425 - -0.0467 -0.118* 0.115 (0.0601) (0.0673) - (0.0668) (0.0704) (0.155) Soil organic carbon (g/kg) 0.184 0.184 - 0.201 0.206 -3.135 (0.238) (0.269) - (0.264) (0.243) (6.299) Constant -19.69 -38.04 102.9*** -12.10 107.1*** 64.79 (23.86) (28.16) (16.38) (25.87) (13.19) (82.40) Observations 1,166 1,166 1,166 1,166 787 379 Time FE Yes Yes Yes Yes Yes Yes Individual FE Yes Yes Yes Yes Yes Yes Number of households 595 595 595 595 402 193 Dependent variable: Value of crop income in CFA per Ha; RE: Random effects; BE: Between effects; FE: Fixed effects x: shows interaction term; All continuous covariates are expressed in their natural logs. Standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1 54 A1. The MWG model CCAFS has worked closely with the National Meteorological Agency (ANACIM) to develop locally-relevant climate information services and enhance the capacity of partners to commu- nicate this information to end users. The national MWG mainly comprises the Department of Agriculture (DA), the Institute of Agricultural Research of Senegal (ISRA), the Ecological Monitoring Center (CSE), the National Agricultural and Rural Council Agency (ANCAR), the National Agricultural Insurance Company of Senegal (CNAAS) and ANACIM (Ndiaye et al., 2013). In 2011, CCAFS scientists partnered with ANACIM with the aim of 1) developing WCIS that are tailor-made for users; 2) building the capacity of partners who were tasked to communi- cate climate information to farmers; and 3) enhancing the transmission of WCI and agricul- tural advisories for farmers. Under this initiative, the MWGs were set up both at the national and local levels. MWGs are decisive and inclusive bodies that facilitate the development of CIS, its interpretation to actionable decisions, diffusion and subsequent use by users at the district level. Local MWGs are set up to closely monitor climatic events and phenomena and to translate climate forecasts into timely advisory services that help guide farmers in making informed decisions (Ouédraogo et al., 2018). The Kaffrine region has four departments – Kaffrine, Koungueul, Birkilane and Malem Hodar. However, climate information activities began and focused more extensively on the depart- ment of Kaffrine. The Kaffrine climate services project was one of the first local MWGs to be set up. It was implemented in 2011 under the CCAFS flagship 2 program. The objective of the initiative was to provide smallholder farmers with relevant climate information to man- age the risk posed by climate and weather variability through informed decision making. In line with this objective, the goal of the Kaffrine project was to provide tailored, down-scaled climate information and advisory services to support climate risk management and enhance resilience. Activities included strengthening the capacity of ANACIM to produce downscaled climate information and agricultural advisories. Several types of climate information have been produced by ANACIM, including seasonal forecasts of the onset of the rainy season, to- tal amount of rainfall, end of the rainy season, daily weather forecasts, 10-day weather outlook and early warnings. These climate products were designed to be useful for specific types of agricultural activities and have been disseminated right down to the district level. For exam- ple, information on the onset of rains is used to schedule planting (buying seeds, preparing the land, hiring labor, planting seeds); information on the total rainfall amount will help farmers choose the appropriate crops and which varieties to plant; and daily weather forecasts help plan daily activities such as weeding and application of chemicals and fertilizers. The dissemination chain of climate information involves several stakeholders including ANACIM, members of the MWG, community radios, farmers and some selected leader or promoter farmers in their villages. Within this setup, ANACIM is the main provider of scien- tific climate information and works in close collaboration with members of the MWGs. The technocrats from ANACIM interact with farmers and exchange ideas on how to integrate farm- 55 ers’ indigenous knowledge and forecasting methods with scientific weather forecasts. Farmers are also given the opportunity to request specific climate information as well as say how they would like to receive climate information. Such tailoring of climate information to meet the needs of end users can increase access to and use of the information. Once produced, information is disseminated directly through short message services (SMS) to a number of farmers in the ANACIM’s Short Message Service (SMS) database, the MWGs, community radios, the Rural Department for Development Services (SDDR), and local ad- ministrative authorities. In the department of Kaffrine, the MWG includes representatives of the decentralized administrative services (Ministry of Agriculture, Livestock, Environment, etc.), NGOs, and Union des Radios Associatives et Communautaires du Senegal (URAC). This group meets every 10 days to discuss how climate information related to agronomic advice can be translated into actionable information for farmers. The outcomes of these dis- cussions are delivered to promoter farmers through radio, cell phone calls, SMS, or word of mouth. Promoter farmers are progressive farmers, leaders of farmers’ organizations, or farmers with strong influence (e.g., religious and community leaders) who are in charge of delivering information to other farmers. They are selected by the district SDDR to convey climate information in their villages. Farmer promoters share the information with fellow farmers through SMS, phone calls and by word of mouth. Farmers can also receive the WCI directly by listening to the community radios or from the SDDR agent. The existence of an MWG at the department level and leader farmers in villages to relay information appeared to be instrumental in the peer farmers’ ac- cess to climate information. For example, in other departments of the Kaffrine region, such as the department of Birkilane, where an MWG has only recently been created, access to and use of climate information seem to be limited. This multidisciplinary partnership in the co-production and dissemination of climate information is summarized in Figure 9 below. Local MWGs also run an early warning system (EWS) based on climate information received from ANACIM. They meet every 10 days and produce agricultural advice that is shared with policymakers and farmers through a special program broadcast on community radio. The interactive radio program allows listeners to give their feedback, including additional informa- tion, views, and requests for clarification. By 2015, the project had partnered with an association of 84 community radios that reached out to a population of 7.4 million rural households in Senegal (CCAFS, 2015; Lo & Dieng, 2015; Ouédraogo et al., 2018). In 2011, CCAFS scientists started collaborating with the Na- tional Civil Aviation and Meteorology Agency of Senegal (ANACIM) which aimed to: 1) develop downscaled weather and climate information services; 2) enhance the capacity of partners who were tasked with communicating weather and climate information to farmers; and 3) enhance the transmission of WCI and agricultural advice to farmers. A total 84 jour- nalists from community radio stations were trained to understand and communicate climate information (CCAFS, 2015; Lo & Dieng, 2015). ANACIM organizes a seminar at the beginning of each rainy season with all local partners to inform farmers of the major trends expected. The seminar also provides farmers with an opportunity to share their own forecasts based on traditional knowledge, with other stakehold- ers. In all program areas, the major channels for disseminating information are e-mail), SMS, 56 Climate information production, transla- tion and dissemination under the MWG Seasonal Daily weather forecasts forecasts Instant EWS ANACIM Farmers Local authority NGOs Local MWG (customize climate information) Community Extension radio services Government ministries Local MWG Text Social Rural radio Bulletin messaging gatherings Figure 9: Conceptual schematization of the MWG model Adapted from Chiputwa et al., 2020 radio, television and by word of mouth. To reduce the costs associated with CI dissemina- tion via SMS, in each intervention area, the project targeted farmer leaders whose contacts could be obtained from the SDDR. The information is sent to the SDDR by ANACIM, and dispatched to the MWG and contact farmers and extension agents who then disseminate it to the farmers. 57 Climate Stakeholders: Community information Experts and (dissemination) products decision makers The CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS) brings together some of the world’s best researchers in agricultural science, development research, climate science and Earth system science, to identify and address the most important interactions, synergies and tradeoffs between climate change, agriculture and food security. For more information, visit us at https://ccafs.cgiar.org/. Titles in this series aim to disseminate interim climate change, agriculture and food security research and practices and stimulate feedback from the scientific community. CCAFS is led by: CCAFS research is supported by: Science for a food-secure future Science for a food-secure future