Technical report “Remote sensing based Information and Insurance for Crops in emerging economies” (RIICE) in Kenya (2023 & 2024)” Sushree Satapathy1, Jeny Raviz1, Renaud Mathieu1, Vincent Koskei2, Faith Mwende2, David Aleri2, Peterson Gichobi2, Ryans Mwangi2, Alessandro Cattaneo3, Luca Gatti3 and Francesco Holecz3 1International Rice Research Institute (IRRI), 2National Irrigation Authority (NIA), 3Sarmap Background : The CGIAR Initiative on Digital Innovation and Transformation (WP4) aims to improve real time monitoring of food system dynamics, improve the flow of information for a better understanding of the food systems, and enhance digital data streams for improving decision making. To achieve these objectives, the International Rice Research Institute, in partnership with sarmap, Switzerland, and the National Irrigation Authority (NIA), Kenya, is piloting a digital rice production monitoring platform. RIICE is a remote sensing-based digital platform that delivers near real time, geolocalized, accurate, and updated seasonal information on rice area, yield, planting dates, and climate-driven yield losses. The system has been tested and the rice production was monitored during two main rain seasons and one off season in the Mwea irrigation scheme (2023-24). This monitoring has been extended to irrigation schemes in Western Kenya in 2024. The NIA team supported the acquisition of field data for each season and site. An inventory of rice areas was initiated for all irrigation schemes in Kenya. Stakeholders meetings were held in Nairobi to discuss the customization of the RIICE platform and review initial RIICE results. A stakeholder engagement event “Engagement for South-South Collaboration on Improved Rice Cultivation Using Geospatial Sciences” was organized to discuss opportunities for an R&D program focused on geospatial applications. This report covers activities conducted and outputs generated during 2023-2024. Date: 12/12/2024 // Work Package: 4. Real-time Monitoring // Partners: IRRI, Sarmap, NIA This report has been prepared as an output of CGIAR Research Initiative on Digital Innovation, which researches pathways to accelerate the transformation towards sustainable and inclusive agrifood systems by generating research-based evidence and innovative digital solutions. We thank all funders who supported this research through their contributions to CGIAR Trust Fund. 1 http://www.riice.org https://www.cgiar.org/initiative/25-harnessing-digital-technologies-for-timely-decision-making-across-food-land-and-water-systems/ https://www.cgiar.org/initiative/25-harnessing-digital-technologies-for-timely-decision-making-across-food-land-and-water-systems/ https://www.cgiar.org/funders https://www.cgiar.org/funders https://www.cgiar.org/funders 2 1. Introduction Rice is one of Kenya’s most important grains and has become an important staple food in urban areas. In 2008, the rate of increase in annual rice consumption was 12%, compared to 4% for wheat and 1% for maize, the nationwide staples (MoA, Kenya, 2020; Ndirangu and Oyange, 2019; Atera et al, 2018). Although the crop acreage, unit yield, and total volume of rice production have remained relatively stable since 2012, the volume of rice imports increased by approximately 60% between 2012 and 2019, aligning with the rise in consumption. The quantity of rice produced in 2019 was 80,000 tons, while consumption reached 710,000 tons (USDA, 2020). This indicates that rice consumption in Kenya is heavily reliant on the international market, which has significant implications for national food security (Atera et al, 2018). To address the growing demand for rice, the Government of Kenya developed a National Rice Development Strategy (NRDS II), aiming to increase local rice production sevenfold from 128,000 metric tons (MT) in 2019 to 846,000 MT by 2030. This is expected to reduce dependence on imports by 30% by 2027 (NRDS II, 2020). As part of the CGIAR Research Initiative on Digital Innovation, IRRI and sarmap aim to adapt, calibrate, and implement rice production monitoring technology for Kenya in collaboration with the National Irrigation Authority Mwea Irrigation Agricultural Development Center (NIA-MIAD). The goal is to develop and transfer a remote sensing-based rice production monitoring platform to the Government of Kenya. The methodology used in this study builds on geospatial technology developed since 2012 through the Remote Sensing-based Information and Insurance for Crops in Emerging Economies (RIICE, www.riice.org), which generates validated rice information such as rice area, start of season (SoS), and yield, - addressing where, when, and how much rice is produced in a given location. RIICE was developed and implemented by IRRI, sarmap, and national partners, with institutional support from the Swiss Development Cooperation; its initial years focused on development in Southeast and South Asia. Currently, RIICE is being piloted in Ivory Coast, Mali, Nigeria, Kenya, and Senegal. RIICE integrates Synthetic Aperture Radar-derived spatial products such as rice area, rice start of season and Leaf Area Index (LAI) with a process-based rice crop growth model (ORYZA) to forecast and estimate yield at mid-season and harvest time, respectively. The model can be scaled using local rice agronomic knowledge integrated into a rice ecosystem model alongside geographical and temporal information extracted from remote sensing data. In 2023, the system was tested and implemented in the Mwea Irrigation Scheme. The NIA team was trained to support the field data collection while monitoring rice production during the main season of 2023 in Mwea. In 2024, monitoring was extended to include the 2024 off season, and repeated for the main season in 2024. In addition, it was extended to the irrigation schemes located in Western Kenya, specifically Ahero and South West Kano. An inventory of rice areas was also initiated for all irrigation schemes across Kenya. This report 3 http://www.riice.org covers all activities conducted to generate rice production data for 2023 and 2024, as well as training and coordination activities with national stakeholders, particularly focusing on demonstrating the system’s effectiveness, adapting the system to local conditions and needs, and discussing a roadmap for its implementation and institutionalization. 2. Data and Methods 2.1. Study area The government of Kenya, via the National Irrigation Authority (NIA), has been actively improving, rehabilitating, and developing irrigation schemes throughout the country. Kenya’s irrigation schemes are crucial for enhancing agricultural productivity, particularly in rice cultivation. The goal is to ensure that farmers benefit from these irrigation schemes and ultimately improve food production in Kenya. There are eight main irrigation schemes in the country ( Figure 1). RIICE was piloted in the Mwea Irrigation Scheme (MIS) in the 2023-24 main season, and the piloting was extended to the Ahero Irrigation Scheme (AIS) and West Kano Irrigation Scheme (WKIS) located in Kisumu County, in 2024. Figure 1. Location map of the eight main government managed irrigation schemes in Kenya. MIS (Figure 2) is located in Kirinyaga County. It is the oldest and largest among the four major gravity-based irrigation schemes in Kenya and produces 80% of the paddy rice produced in the country. The development of MIS started in 1954 from the Tebere section with about 65 acres in irrigation farming, since then the scheme has grown to the current official area of 30,050 acres. Out of these, 73% (22,000 acres) have been developed for paddy rice production while the remaining area (8,050 acres) is utilized for settlement, public utilities and growing of subsistence 4 crops. The scheme has five sections: Mwea, Tebere, Thiba, Karaba, and Wamumu. Expansion also concerned areas surrounding the scheme, referred to as out-growers, by a total of 8,600 acres; increasing the total area for paddy rice production in MIS to 30,600 acres. Over the years NIA has improved, rehabilitated and developed irrigation infrastructure in the scheme. The scheme lies along the drainage basins of Rivers Nyamindi and Thiba which supply the irrigation water through gravity. There is a potential of up to 10,000 acres for further expansion within the surrounding areas, but this is currently constrained by lack of sufficient irrigation water. Currently, the scheme practices two and half production seasons in a year: (1) the main season (July to December), and a ratoon crop (or regrowth) from December to February; and (2) the second season (March to July). Depending on weather conditions, the area being cultivated from March to July varies from 5,000 to 10,000 acres. The main varieties produced in MIS are Basmati 370 ( 80% ) and Komboka ( 13%) and newly introduced hybrid varieties i.e. Arize Tej Gold & Arize 6444 Gold (7%). AIS and WKIS are both located in Kisumu County (Figure 3), in the Kano plains between the Nandi Escarpment and Nyabondo Plateau, while WKIS is on the shores of Lake Victoria. AIS was started as a pilot project to explore the feasibility of irrigation in the Kano Plains. Construction of the scheme started in 1966 and operations started in 1969. WKIS was constructed in 1974 and became operational in 1976. The main crop grown in the Ahero and West kano irrigation schemes is rice. The irrigation is based on water pumped from the Nyando River, which limits water provision, and forces a strict water management over the complete season (differential planned planting). Both schemes produce 22,000 tonnes of rice annually. The area under irrigation is 10,810 acres. The main rice varieties are 90% Sindano (IR-2793 & ITA 310), 5% aromatic (Basmati-370) and 5% hybrid (Arize Tej Gold & Arize 6444 Gold). Rice production was piloted in Ahero and West Kano irrigation scheme for main season 2024. 5 Figure 2. Map of Mwea Irrigation scheme (MIS). Figure 3. Map of Ahero and West Kano irrigation schemes. 2.2. Data 2.2.1. Satellite data Multi-temporal C-band Synthetic Aperture Radar (SAR) Sentinel-1A (S-1A) images were obtained at no cost from the Copernicus Open Access Hub. Additionally, optical imagery, including monthly composites from Planet, as well as Landsat 8,9, and Sentinel-2 data, were used. The characteristics of each satellite image are summarized in Table 1. In Mwea, a total of 47 S-1A images acquired between June 12, 2023 and November 9, 2024 were used to detect rice areas and identify start of season (SoS) dates (Table 2). The acquisition dates cover the rice growing season from land preparation and flooding, planting up to harvest. Landsat and Sentinel-2 images were obtained from USGS Earth Explorer and Copernicus Open Access Hub, respectively. NDVI (Normalized Difference Vegetation Index) images derived from Landsat 8 and 9 and Sentinel-2 data, were used as input for rice classification and for creating the rice-non rice masks. Monthly NDVI images derived from Planet data were mainly used for rice and non-rice masks. 2.2.2 Weather data Required weather data, specifically for the field modelling, include: daily values of solar radiation, minimum and maximum temperature, wind speed, vapor pressure, and rainfall. Local weather data were obtained from MIAD for one weather station. As only one station was available for the scheme, and some of the weather parameters were not available, weather data from the AgERA5 dataset (Copernicus) were used instead. Copernicus Climate Data Store (CDS) provides daily surface meteorological data for the period from 1979 to present, based on 6 https://scihub.copernicus.eu/ https://cds.climate.copernicus.eu/datasets/sis-agrometeorological-indicators?tab=overview hourly ECMWF ERA5 data at surface level. AgERA5 has a global coverage and a native resolution of 0.1 x 0.1 degrees, which is roughly 11 x 11 kilometers. Table 1. Characteristics of the satellite data used in rice mapping for 2023 and 2024 in Kenya. Satellite , Band Sentinel-1A (SAR-C) Landsat 8/9 Sentinel-2 Type Radar Optical Optical Mode Interferometric wide swath (IW) - - Product type Ground Range Detected (GRD) Level 1 Level 1 Swath width (km) 200x250 185x180 290 Spatial resolution (m) 20 30 10 to 60 Pass direction Ascending - - Polarization, incidence angle VV VH, 40° at scene center - - Repeat cycle (days) 12 16 10 Orbit number, Path/row Mwea 57 168/60 - Ahero, West Kano 130 Table 2. Satellite images used in rice mapping for each irrigation scheme by season. Irrigation Scheme Mwea Ahero and West Kano Year, season 2023 Main 2024 Off 2024 Main 2024 Main Number of images Sentinel-1A 17 16 14 24 Landsat 8/9 6 0 5 0 Sentinel-2 2 0 0 0 Planet 26 26 6 0 Start and end dates Sentinel-1A 12-Jun-23 21-Dec-23 19-Feb-24 17-Aug-24 06-Jun-24 09-Nov-24 12-Feb-24 14-Nov-24 Landsat 8/9 28-Jun-23 10-Oct-23 - 08-Jul-24 04-Oct-24 - 7 Sentinel-2 26-Aug-23 25-Sep-23 - - - 2.2.3 Other data Other data used include the 30m Digital Elevation Model (DEM) from the Shuttle Radar Topography Mission (SRTM). DEM was used as input in satellite image pre-processing and in the rice map classification. 2.2.4 Field data Three types of field data required for mapping the rice area and yield were collected for the monitored seasons. ● Rice and non rice fields for calibration and validation of rice area maps. Transplanting/sowing date for the validation of start of season map ● Leaf Area Index temporal dynamic for selected fields for calibrating crop growth ● Crop Cut Experiment to validate the yield estimates The number of samples for each activity for different seasons and different irrigation schemes are shown in Table 3. Table 3. Distribution of each activities in different season and different irrigation scheme Irrigation Scheme Calibration Validation LAI CCE Year, season Rice Non-rice Rice Non-rice Mwea 2023 Main 40 40 77 32 15 60 2024 Off 40 NA NA NA NA NA 2024 Main 30 29 39 39 8 60 Ahero 2024 Main 20 NA NA NA 5 NA 8 Calibration and validation data Two sets of field data for rice mapping were collected: calibration and validation (Figure 4, Table 3). Calibration data were used to extract rice and non-rice temporal backscatter values and calibrating the parameters used for discriminating rice and non-rice classes, while validation data were used for assessing the accuracy of the rice area maps. Field data for calibration were collected two to three weeks after SoS. Data collected include: geolocalized field boundary, photographs, crop establishment method, variety and maturity duration, and field conditions. Rice crop management information was also collected during the field campaign when the farmer was available for an interview. Non-rice areas planted with other crops were prioritized, however these were very limited and only very small fields were found within the scheme. Most non-rice were found on the outskirts of the scheme, and consequently surveyed non-rice fields were reduced. The non-rice fields collected include other crops such as maize, sorghum, tomato, sweet potato, beans, cabbage, watermelon, banana and chili and other land cover types (shrubland, grassland, fallow). Other obvious land cover classes such as forest, settlements and water bodies were not sampled because these types of land cover have stable signals, and their signature can be easily differentiated from that of rice. Field data for validation were collected, one to two weeks before the crops were harvested. In addition, during the field data collection for the 2024 main season, transplanting and sowing dates were gathered. For fields that are transplanted, the age of seedling was also recorded. A total of 23 fields provided this information, which was used to validate the SoS map. 9 Figure 4. Spatial distribution of calibration/validation field data, MIS, 2023-2024 main seasons. Leaf Area Index (LAI) data collection LAI measurements were taken starting 20 days after transplanting, every 12 days, corresponding to Sentinel-1 image acquisition dates, until flowering, to capture and calibrate crop growth temporal dynamics. A total of 15 fields were sampled during the main season of 2023 and 8 fields for the main season of 2024 for Mwea, and 5 fields for Ahero during the main season of 2024 (Figure 5). Fields were selected to best represent variability according to ecosystems and/or productivity classes. A total of 5 sets (dates) of LAI data were collected during the growing season in 2023 (from September 16 to November 3) and 2024 (from September 10 to October 25) for Mwea, while 4 sets of LAI data were collected during the main season from September 16 to November 4 for Ahero. The LAI measurements were conducted using the PocketLAI app, an Android-based application developed by the Cassandra Lab at the University of Milan, Italy (Confalonieri, et al., 2013). PocketLAI estimates LAI based on the gap fraction, which is the fraction of sky visible from below the canopy. The protocol used to measure LAI across a single field is shown in Figure 6. Across fields and throughout the season, recorded LAI values varied from 1.52 to 4.38 during the main season of 2023 and from 0.55 to 4.59 during the main season of 2024 for Mwea (see Annex Table 1). For Ahero, LAI values ranged from 0.13 to 3.26 during the main season ( See Annex Table 1). Figure 5. Fields selected for LAI data collection in Mwea Irrigation Scheme and Ahero Irrigation Scheme. Crop Cut Experiments (CCE) CCEs, or field yield measurements at harvest, were conducted using stratified random sampling techniques, with irrigation sections serving as sampling units to obtain representative field yield estimates. Three plots (2.5 x 2.5 m) were selected within each field, which had a minimum size of 1 ha (see Figure 6). The rice was harvested, threshed, and the yield was calculated at 14 % 10 moisture content (MC).The different procedures of yield data collection are shown in figure 7. CCE yields varied from 4 to 7 t/ha across the field during the main season of 2023 and from 4.3 to 8.6 t/ha during the main season of 2024, with average yields of 5.2 t/ha and 5.9 t/ha respectively. During 2023, yield values generated through the CCEs may have been affected by the exceptionally rainy conditions and crop lodging which occurred during the harvest period. The CCE selected fields are shown in Figure 8. Figure 6. Protocol to measure LAI for single homogeneous ricefields. 11 Figure 7. Yield data collection, Mwea Irrigation Scheme, main seasons 2023 and 2024 Figure 8. Fields selected for the yield measurements in MIS, 2023 and 2024 main seasons 2.3. Field experiment Calibration of the dominant varieties for a particular region is required for optimal performance of the crop growth model used for yield estimation, ORYZA. The variety/crop file used in ORYZA contains phenological, growth, and yield characteristics of the variety being modelled for a given geographical area. These data were not available for the varieties planted in Mwea, mostly Basmati 370 and Komboka. Consequently, field experiments were implemented during the main season of 2023 at the NIA-MIAD experimental station. A randomized complete block design was used, with fertilizer (five levels) as the main plot and variety (two rice varieties ; Basmati 370 (V1) and Komboka (V2)) as subplots, with four replicates. Basmati 370 is largely dominant in MIS, occupying 80% of the area, while some farmers/fields use Komboka (the remaining 20%). The five levels of fertilizer (F) treatments have increasing doses of Nitrogen (N), with phosphorus (P) and potassium (K) fixed. ● F0 – No NPK (0-0-0) ● F1 – Low NPK (40 kg N/ha-60 kg P2O5/ha-50 kg K2O/ha) ● F2 – Medium NPK (80 kg N/ha -60 kg P2O5/ha-50 kg K2O/ha) ● F3 – High NPK (120 kg N/ha -60 kg P2O5/ha-50 kg K2O/ha) ● F4 – High NPK (160 kg N/ha -60 kg P2O5/ha-50 kg K2O/ha) 12 Dates of sowing, panicle initiation, flowing, and maturity, along with the biomass and yield information from the field experiment, are used as a starting point to calibrate local varieties in the Mwea irrigated ecosystem. The field layout and experimental fields are shown in Figures 9 and 10. Field experiments for the main rice varieties planted in the western irrigation schemes (IR-2793 & ITA 310) could not be done due to budget constraints. Yield estimation will be done with a calibrated variety that will proxy the dominant rice varieties grown there. As soil data are also required for the ORYZA model, soil samples were collected before the start of the experiment at different depths (i.e. 0-25 cm, 25-50 cm, 50-75 cm, and 75-90 cm) at the experimental station. Basic soil physical and chemical properties were analyzed for input into the model (Table 4). Figure 9. Field layout of the experimental field, main season 2023 13 Figure 10. Field experiment site, NIA-MIAD research field. Table 4: Soil physical and chemical properties Soil layers (cm) Sand (%) Silt (%) Clay (%) pH N (%) OM (%) 0-25 12.6 6.00 81.4 5.13 0.099 3.04 25-50 10.6 7.71 81.7 6.01 0.073 2.58 50-75 12.6 3.86 83.6 6.00 0.074 2.61 75-90 14.6 5.86 79.6 6.35 0.062 2.24 2.4. Methods 2.4.1 Rice area and SoS mapping Multi-temporal Sentinel-1A imagery was used to detect and map rice areas and SoS. The algorithm analyzed the SAR multitemporal series and rice temporal signatures. The rice fields and SoS are detected using predefined rules. The overview of the methodology used in this study is presented in Figure 11. Each step is discussed in more detail in the following subsections. 14 Figure 11. Flowchart of rice and SoS mapping. Pre-processing of satellite images MAPscape-RICE, an image processing software developed by sarmap specifically for rice, was used to process the satellite images and delineate the rice and non-rice fields. The pre-processing phase sequentially performs a series of steps to convert the multi-temporal SAR GRD data into terrain geocoded backscatter δo values (Holecz et. al, 2013; Nelson et al. 2014), including briefly: (1) strip mosaicking of single frames in slant range geometry; (2) image coregistration in slant range geometry; (3) multi-temporal speckle filtering; (4) terrain geocoding, radiometric calibration and normalization with Digital Elevation Model (DEM); (5) anisotropic non-linear diffusion (ANLD) filtering; and (6) removal of atmospheric signal attenuation. Rice mask generation First, a rice and non-rice mask was generated using multi-temporal features extracted from Sentinel-1A intensity images, as well as Landsat 8-9, Sentinel-2, and Planet NDVI images. The mask is a proxy for a potential rice area and was used as input in the rice classification to discard pixels that were surely not rice. Rice and start of season detection The multitemporal stack of terrain geocoded δo images were analyzed using a rule-based detection method (Nelson et al. 2014). Rice areas were detected based on the SAR temporal signature following steps such as rice exclusion, detection of agronomic flooding and evidence of rice growth. Classification parameters are set based on ground reference and adapted to specific local conditions and production systems. 15 ● Exclusion of non rice pixels for: (1) areas with consistently low backscatter values, e.g. stable water bodies; (2) areas with consistently high backscatter values, e.g. built up areas, infrastructures; (3) areas where the backscatter remains below the minimum value longer than a given time, e.g. fishponds or wetlands; and (4) areas with unusually high variation in backscatter. Rice shows variation in backscatter values over time but there is a maximum variation that can be expected in biomass. ● Evidence of agronomic flooding at the start of season. Flooded rice fields exhibit low backscatter value at SoS. ● Evidence of crop growth based on: (1) an observed increase in backscatter between SoS to minimum δo value at maximum peak and (2) the variation in the δo signature reached a minimum that is consistent with what is observed for a rice crop. Pixels that meet these two conditions were retained for further processing as a potential rice pixel. ● The last rule looked for any unexpected drops in backscatter that would indicate flood or a new cropping season. If this condition was satisfied, the pixel was labeled as rice. Likewise, rules were applied to detect possible start of season dates using VV and VH polarizations. This method was tested by Raviz et al. (2018) in the Philippines. The parameters used for detecting potential SoS dates include the maximum and minimum backscatter value at SoS. The maximum δo value was used to identify possible rice pixels; if the backscatter is lower than the threshold in certain acquisition, the pixel represents a potential SoS date. The minimum δo value was used to discard possible non-rice pixels; if the backscatter is lower than the threshold in certain acquisition, the pixel cannot correspond to an SoS date. Pixels that meet these conditions were retained for the identification of potential SoS dates. The SoS dates were assigned based on the weights used to compute the reliability coefficient (RC). For each potential SoS date, the parameters used in the computation of RC include: (1) backscatter value detected as SoS from SAR input (ascending and descending, VV and VH polarizations); (2) trend of the SAR temporal signature, showing backscatter increase from SoS; (3) correspondence of SoS date from the SAR input polarizations; (4) consistency of backscatter with NDVI value, if NDVI is used; and (5) value of the local incidence angle (LIA) in areas covered by more than one relative orbit. Higher RC was assigned to the relative orbit with a large viewing angle. These parameters were considered on the basis of its own weight factor, the RC is obtained by summing up the contribution of each input multiplied by its own weight factor. The SoS date was assigned to the pixel with the highest RC. Accuracy assessment The accuracy of the rice area map was computed using standard confusion matrices (validation points are detailed in Table 3). The information (i.e. rice and non-rice land cover) collected on the ground was mapped and compared to the class detected on the map. The overall accuracy of the rice and non-rice classification and the kappa index were computed. Kappa index 16 is an assessment of the accuracy of the map compared to the accuracy that could have been achieved purely by chance. The accuracy of the start of season map was assessed using Root Mean Square Error (RMSE) and Normalized RMSE (NRMSE). To evaluate this accuracy, the planting dates reported by farmers were compared with the start-of-season dates identified on the map. Additionally, the deviation in days between the reported planting dates and the detected start of season dates was calculated. 2.4.2 Yield estimation The rice mapping and yield estimation processes are interconnected, as the products of the rice mapping process- particularly the SoS map and Leaf Area Index- are used to generate inputs for the crop yield model (CYM). The CYM inputs are crucial components within the Rice-Yield Estimation System (Rice-YES), for generating yield estimates. The Rice-YES software serves as an interface that integrates the SAR-derived spatial information (LAI and SoS date) with the ORYZA crop growth model. ORYZA is a weather-driven and process-based rice crop growth model that captures complex and dynamic interactions among weather, soil, varietal, and agronomic management practices affecting rice yield. Rice-YES captures the spatial variability of rice yield based on the inference of rice crop growth indicators in the form of LAI derived from SAR temporal signatures during the early expansion stage of the rice crop. The algorithm correlating SAR signatures to LAI is embedded in MAPscape-RICE software. The combined MAPscape-Rice and Rice-YES platform introduces an innovative spatial allocation approach that allows effective yield estimate mapping without requiring yield simulations for every single SAR data input pixel (Setiyono et al., 2019). Figure 12 illustrates the operational diagram of the yield estimation process. The assimilation of SAR-derived LAI products begins after the rice crop reaches the early expansion stage (roughly 40 days after SoS for a 120-day cropping duration). During this early part of the rice-growing cycle, leaf expansion parameters in ORYZA can be effectively calibrated against real ground conditions inferred from satellite observations. LAI values derived from SAR were used to recalibrate the early leaf growth parameters in ORYZA, particularly the relative leaf growth rate (RGRL) variable. The RGRL variable is an ideal target for calibration due to its known sensitivity of biomass accumulation and yield output of ORYZA to changes in this coefficient (Tan et al. 2016). In addition, the processing of rice yield estimates also requires non remote sensing information, including: ● Weather data: weather data for 2023 and 2024 from the Copernicus AGERA5 dataset were used. 17 ● Varietal characteristics: maturity duration and other growth parameters such as leaf death rate, fraction of stem reserves, and maximum grain weight were calibrated for Basmati 370 and Komboka varieties in Mwea using the field experimental data. Since field experiments were not conducted in the western sites, a calibrated variety, IR 72, is used as a proxy for the dominant variety (IR 2793-08-01) grown in Ahero/West Kano. The proxy variety was chosen based on the crop duration and varietal characteristics. ● Soil data: soil physical and hydrological properties): soil samples were extracted from the NIA-MIAD experimental station and analyzed in the lab (Table 4). ● Crop management information: These information were obtained from farmers during the field data collection and used to define a single crop management profile. The same parameters were used for Mwea across both main seasons and the off season; for example, the crop establishment method is “Transplanted” recommended fertilizer application is 80 kg N/ha (in three doses: planting, tillering, and before flowering); and irrigation water level is maintained at a depth of 2 cm until maturity. In Ahero/West Kano, fertilization was set at the recommended application of 142 kgN/ha (also in three doses: planting, tillering, and before flowering), although it is not always followed on the ground. Figure 12. Operational diagram of RICE Yield Estimation System (RiceYES). 18 3. Results 3.1. Field experiment Biomass sampling was done at each crop growth stage i.e. 35 DAS/14DAT, panicle initiation, flowering and grain filling for both the varieties (Basmati (V1) and Komboka (V2)) to be used in the ORYZAvariety calibration module. Table 5 shows the biomass partitioning data at each crop stage. Both Basmati 370 and Komboka have 120 days crop duration with plant height up to 120 cm, but Komboka has higher tillering capacity and more grains per panicle than Basmati 370. Table 5. Biomass sampling data at different crop growth stages for Basmati 370 (V1) and Komboka (V2), irrigated condition, main season 2023, Mwea Irrigation Scheme. Fertilizer Variety Sampling date Gleaves grams* Dleaves grams* Stem grams* Panicle grams* F0 V1 35DAS/14DAT 0.11 0.00 1.13 0.00 Panicle initiation 12.88 0.00 25.00 0.00 Flowering 46.83 12.00 249.88 22.00 V2 35DAS/14DAT 0.81 0.05 1.22 0.00 Panicle initiation 17.00 0.00 51.88 0.00 Flowering 30.00 12.25 234.75 53.13 F1 V1 35DAS/14DAT 0.45 0.17 0.72 0.00 Panicle initiation 22.38 0.00 45.38 0.00 Flowering 61.33 14.83 319.00 38.00 V2 35DAS/14DAT 1.20 0.04 1.90 0.00 Panicle Initiation 26.38 0.00 75.88 0.00 Flowering 46.75 11.75 311.63 55.75 F2 V1 35DAS/14DAT 0.61 0.14 0.95 0.00 Panicle initiation 22.38 0.33 47.75 0.00 Flowering 62.17 11.33 300.25 40.50 V2 35DAS/14DAT 1.07 0.04 1.97 0.00 Panicle initiation 21.25 0.00 58.13 0.00 Flowering 56.63 12.75 333.13 67.38 F3 V1 35DAS/14DAT 0.55 0.23 1.08 0.00 19 Panicle initiation 17.63 0.00 39.25 0.00 Flowering 83.33 9.50 302.88 30.13 V2 35DAS/14DAT 1.01 0.06 1.85 0.00 Panicle initiation 20.88 0.00 55.50 0.00 Flowering 71.63 10.88 337.13 54.50 F4 V1 35DAS/14DAT 0.84 0.10 1.20 0.00 Panicle initiation 17.88 0.00 39.50 0.00 Flowering 79.67 12.50 246.88 26.75 V2 35DAS/14DAT 0.83 0.05 1.41 0.00 Panicle initiation 23.25 0.00 62.88 0.00 Flowering 89.75 10.63 334.63 55.38 *DWGleaves : Green Leaves; DW Dleaves: Dead Leaves; DW stem; Stem; DW panicle; panicle (all oven dry weight) 3.2. Rice area and SoS Mwea Irrigation Scheme The validated rice area maps for Mwea for 2023 and 2024 main seasons are presented in Figure 13. In the 2023 main season, the rice map achieved a very good accuracy of 95% and a kappa index of 0.91 (Table 6). Some actual rice areas were misclassified as non-rice, likely due to small patches of rice areas being surrounded by other land covers, causing mixed signals. A total of 10,614 hectares (approximately 26,000 acres) were cultivated with rice, of which 72% (7,690 ha) was planted in the main irrigation sections (Table 7). The remaining 22% was planted in the outgrower areas, while 6% was outside the scheme. The outgrower areas, although not officially part of the scheme, are integrated in the scheme management, in particular for the water provision. The overall accuracy for the 2024 main season was 92%, with a kappa of 0.85 (Table 6). Similar to the main season of the previous year, 73% of the rice cultivation occurred in the main irrigation sections. However, there was a decline of over 900 hectares in rice planting from 2023 to 2024 during the main season, which affected mostly Karaba (200 ha) and Terebe (300 ha). The top rice growing areas during the main season across sections were (1) Tebere, (2) Mwea, and (3) Thiba (Table 7). The rice map for the 2024 off season (Figure 13) was not validated due to the lack of validation data. In total, 4,569 ha was detected as rice during the 2024 off season, of which 66% (3,052 ha) was planted in the main irrigation sections (Table 7). The remaining 26% was planted in the outgrower areas, while 8% was outside the scheme. At the scale of the entire scheme the area cultivated during the off season is only about 45% of the area cultivated during the main season. The percentages are 42, 48, and 54, for the main section, outgrowers, and outside the scheme sections, respectively. Note that Karaba is the only section which maintains a very high 20 off season cultivation percentage with about 88% compared to the main season. Overall during the off season of 2023, the top rice growing areas were (1) Karaba, (2) Tebere, and (3) Curukia in the outgrower sections. Figure 13. Rice area map in the Mwea Irrigation Scheme, 2023 to 2024. The main season maps were validated. Validation points are represented in dots (black - rice, yellow - not-rice). 21 Table 6. Accuracy of the 2023 and 2024 main season rice maps, Mwea Irrigated Scheme. Table 7. Rice area estimates in MIS, 2023 - 2024 cropping seasons. Year, season Rice area, ha 2023 Main 2024 Off* 2024 Main MIS 10,614 4,569 9,690 Main section 7,690 3,052 7,072 Karaba 1,340 1,103 1,177 Mwea 1,696 446 1,604 Tebere 1,953 722 1,656 Thiba 1,446 260 1,436 Wamumu 1,255 521 1,200 Outgrower 2,286 1,170 2,108 Curukia/Mutithi Expansion 1,390 670 1,240 Kiamanyeki 65 70 161 Kianugu 175 122 115 Marurumo 30 48 17 Ndekia 586 236 546 Ngucwi 41 24 29 22 Others (outside scheme) 638 347 510 * The 2024 off season rice map was not validated as validation points were not collected. The start of season (SoS) maps for the main seasons are presented in Figure 14. In both years, July (especially mid to end of July) was the predominant month for planting, accounting for 61% of total planting in 2023 and 60% in 2024. Most planting in July was concentrated in the main irrigation sections and Curukia within the outgrower area. Delayed planting was observed in Ndekia and parts of Karaba, occurring in mid August for 2024 and early September for 2023. For the 2024 off season, the majority of planting across the scheme was in March, averaging 69% of the total. The monthly distribution of planting by season is presented in Figures 15 through 17. The estimated start of season date for the 2024 main season has an overall accuracy of 96%. The average deviation between the SoS from RIICE maps and the planting dates reported by farmers is only 8 days. Given that the maximum revisit period of the Sentinel-1A satellite is 12 days, this 8-day deviation falls within the acceptable range. 23 Figure 14. Start of season map in Mwea Irrigation Scheme, 2023 to 2024 cropping seasons. Figure 15. Distribution of planting by month, 2023 main season, Mwea Irrigation Scheme. 24 Figure 16. Distribution of planting by month, 2024 off season, Mwea Irrigation Scheme. Figure 17. Distribution of planting by month, 2024 main season, Mwea Irrigation Scheme. Ahero & West Kano Irrigation Schemes Preliminary rice and SoS maps for Ahero and West Kano Irrigation Schemes are presented in Figure 18 and 19, respectively. Due to the late data collection and submission, the map has not yet been validated. Preliminary rice area estimates in Ahero and West Kano Irrigation Schemes are summarized in Table 8 and 9, respectively. 25 In the Ahero Irrigation Scheme, the total area planted to rice was 1,063 hectares. The start of the season varied significantly across different blocks. Of the total area, 32% was planted in February, primarily in blocks H, D, F, G, M, B, and L. In March, the start of the season was notable in Block O and some parts of Block H. Planting in April was observed in Block N, while May planting occurred in Block C. In September, 30% of the area was planted, particularly in blocks A, B, C, K, and L. The remaining 25% of the area was spread across the months of April to August, indicating a diverse planting schedule across the scheme. In the West Kano Irrigation Scheme, the total area planted to rice was 753 hectares. Approximately 54% of the area was planted in May, primarily in blocks E2, D1, C1, D2, H, F2, F1, G, B2, C4, B1, and C2. In April, 21% of the area was planted, with Block E1 being notable. Early planting in February, mainly observed in blocks A and C3, accounted for 13% of the total area. Planting in March accounts for 10%, while the accumulated planting from June to September accounts for only 2% of the total area. In comparison to Ahero, where planting is more distributed across the season with peaks in February and September, West Kano has a more concentrated planting period in May. As the rice and SoS maps were not validated in the western sites due to time and funding constraints, it will be important to confirm whether RIICE findings align with what is actually happening on the ground. Field data are essential to assess the accuracy of the products. Figure 18. Rice area map in Ahero and West Kano Irrigation Schemes, 2024 main season (not validated). 26 Figure 19. Start of season map in Ahero and West Kano, 2024 main season (not validated). Table 8. Preliminary estimates of rice area and start of season by month and blocks, Ahero Irrigation Scheme, 2024 main season. Blocks Area estimates (hectare) Rice Start of season Feb Mar Apr May Jun Jul Aug Sep Block H 30 14 14 - - - - - 3 Block A 116 0 17 23 6 1 11 5 53 Block B 96 32 15 - 0 3 8 1 36 Block C 31 1 - - 10 7 - - 13 Block D 93 65 6 - - - - - 21 Block F 155 78 3 1 6 5 40 - 23 Block G 24 24 - - - - - - 0 27 Block O 35 - 24 5 0 - 0 - 5 Block N 111 - 34 47 13 5 0 0 13 Block K 100 6 0 - 3 0 33 12 46 Block M 71 53 10 - 5 - - - 3 Block L 82 34 3 - 1 1 5 1 37 Outside blocks 118 34 7 2 3 1 3 - 69 Grand total 1,063 341 133 78 47 23 100 18 323 Table 9. Preliminary estimates of rice area and start of season by month and blocks, West Kano Irrigation Scheme, 2024 main season. Blocks Area estimates (hectare) Rice Start of season Feb Mar Apr May Jun Jul Aug Sep Block A 59 35 8 5 11 0 0 0 0 Block D1 55 0 0 25 30 0 0 0 0 Block C1 33 3 7 0 18 3 1 0 0 Block D2 42 0 0 24 17 0 0 0 0 Block J 25 3 21 0 0 0 0 0 0 Block H 114 14 16 25 56 1 1 0 0 Block E1 46 0 0 27 19 0 0 0 0 Block E2 27 0 0 0 27 0 0 0 0 Block F2 52 15 4 0 32 0 0 0 0 Block F1 47 0 0 20 27 0 0 0 0 Block G 54 5 0 1 47 0 1 0 0 28 No name 35 0 0 14 21 0 0 0 0 Block B2 42 14 9 0 18 0 0 0 0 Block C4 15 2 6 0 6 0 0 0 0 Block B1 43 3 0 6 31 1 1 0 0 Block C3 17 5 0 5 2 0 3 0 2 Block C2 50 0 0 4 46 0 0 0 0 Grand Total 753 100 73 155 409 6 8 0 2 3.3. End of season yield estimates Figure 20 presents the spatial distribution of estimated Leaf Area Index (LAI) at the early expansion stage in the irrigated ecosystem of MIS in the Mwea region for 2023 and 2024. Given that the whole area has an ample supply of water, rice vegetation is showing a high and mostly uniform LAI value at early expansion stage. The typical spread of LAI during the early expansion stage is around 1-1.5 in rainfed lowland ecosystems, whereas in Mwea the LAI maps showed high values around 2-2.5 in 2023, and 2.5-3 in 2024. The LAI maps were calibrated using the 2023 main season ground data and the 2023 calibrated parameters were used for generating the 2024 LAI map which is validated with ground LAI data. LAI is sensitive to a variety of factors. At a global scale it is related to climate and plant functional type. At a local scale it is affected by weather and site factors such as fertility, crop age and management treatment, disturbance history. So more validation from ground is required to calibrate the LAI map. The yield estimate map illustrates the spatial distribution of end-of-season yield estimates for Mwea during the main season of 2023 and 2024 (Figure 21). The term "end-of-season yield estimates" typically refers to the anticipated agricultural output in t/ha per hectare of the land at the conclusion of the growing season (harvest). The pixel yield values ranged from 5.0 to 6.9 t/ha with average yield of 6.4 t/ha for 2023 main season and 4.0 to 7.6 t/ ha with average of 6.2 t/ha for 2024 main season. The slightly higher yield in 2023 may be linked to favorable weather conditions such as solar radiation and temperature compared to the 2024 main season, however the difference is not significant. During off season 2024, yield ranged from 3.6 to 6.4 t/ha with a mean of 5.5 t/ha (Figure 22). The lower yield during off season may be due to lower temperature during flowering which affects the grain filling and rice blast is a common disease during off season which lower the grain yield. LAI calibration has not been done during off season and LAI maps were generated using the default parameters in the model. Yield field validation could not be implemented in these other sites due to lack of resources. The main reason for the high yield during the main season is the availability of good irrigation supply throughout the entire 29 cropping season. Irrigation plays a crucial role in agriculture by providing a controlled and consistent water supply to crops, especially in regions where rainfall may be insufficient or unreliable. The consistent irrigation in these areas has likely contributed to optimal growing conditions, resulting in higher yields compared to regions that may have faced challenges related to water availability. These yield products were generated using a calibrated variety file Basmati370, dominating in the scheme. In Ahero/ WestKano, ground LAI data and crop management, irrigation and variety information has been collected with farmers' surveys. The spatial yield will be generated after finalizing the processing of the rice area, start of season and LAI maps. Figure 20. Leaf area index at the early expansion stage for MIS, main season, 2023 and 2024. 30 Figure 21. End of season yield estimates for MIS, main season (a) 2023 and (b)2024. Figure 22. End of season yield estimates for MIS, off season 2024. 3.4 Yield validation The validation of the spatially-explicit yield generated by RIICE was carried out in 60 fields in the Mwea irrigation scheme for both years. During the 2023 main season, CCE yields varied from 4 to 7 t/ha across the field with an average of 5.2 t/ha and in 2024 main season, CCE yields varied from 4.3 to 8.6 with an average of 5.9 t/ha (see Figure 23). The RIICE estimated yield was validated with CCE yield for each field. The average deviation of estimated yield from 31 CCE yield is 25% during 2023 season and 10% during 2024 season. The higher variability during the 2023 season may be linked with crop damage due to lodging during harvest time. Figure 23. Comparison between field measured CCE yield and RIICE estimated yield, (top) main season 2023 and (bottom) main season 2024 in Mwea 4. Capacity building Online training on the collection of field data for use in the rice mapping was carried out by Jeny Raviz and Sushree Satapathy on August 22 and 25, 2023. The training was attended by 32 four partners from the National Irrigation Authority (NIA) and one IRRI consultant deployed on the project and based in NIA. The program of activities and the list of participants are shown in Annex Table 2 and 3, respectively. The training aimed to discuss the protocols for field data collection using appropriate tools and digital survey forms. The training covered basic GIS data loading, viewing and editing, data collection with GPS and training KoboToolBox, use of PocketLAI App for LAI data collection with a cellphone. During the main season 2024, the RIICE activities extended to the western irrigated schemes ( Ahero/ West Kano). A series of virtual training on field data collection was provided to the field team in Ahero/ West kano and as well as a new resource persons joined in Mwea irrigation scheme. The list of participants are shown in Annex 3. In addition, regular on-line meetings (2 weeks) were held throughout the project to ensure on-the-job training and quality field data collection. 5. Stakeholder engagement meeting for developing RIICE, Geospatial and Digital activities in Kenya IRRI in partnership with sarmap and the Ministry of Agriculture organized a Stakeholders’ Engagement meeting to introduce the RIICE technology to the Kenyan community. The meeting was hosted in Pride Inn Azure, in Nairobi on the 5th of September 2023. Around 23 participants representing different agencies from Kenya attended this workshop (Annex Table 4). The workshop mainly focused on introducing the RIICE technology to local partners, with the objectives of customizing the technology based on local requirements and identifying key national partners to engage in the RIICE technology. The proposed next steps and recommendations encompasses: (1) Stakeholders informed on the RIICE tool and applications. (2) Stakeholders' inputs in customizing RIICE for Kenya are documented. (3) Keys partners identified and engaged for the customization of RIICE for Kenya. (4) Collaborative roadmap for the customization of RIICE for Kenya discussed and initial elements documented. Figure 24. RIICE stakeholder’s meeting in Nairobi, September 5, 2023. 33 A follow up virtual stakeholder meeting was organized on 21st May, 2024 to present the 2023 achievements, activities for 2024 and the way forward. Participants were from NIA, KALRO, Ministry Agriculture - Rice Promotion Program, JKUAT, RCRMD, Kenya National Bureau of Statistics. The meeting mainly focused on RIICE achievements in 2023 in Mwea and the way forward were discussed, in particular in relation with the development of an institutional framework for rice (plan), identification of national champions to support implementation, and funding opportunities for a larger program. Figure 25. RIICE stakeholder’s virtual meeting, May 21, 2024. Recently, on November 14-15 2024 a workshop titled “Stakeholder Engagement for South-South Collaboration on Improved Rice Cultivation Using Geospatial Sciences” was organized in Mwea under the patronage of the CGIAR Digital Initiative, and as part of the RIICE activities. The workshop was organized by IRRI with NIA, KALRO, the Ministry of Agriculture of Kenya, and the ICAR-IARI’s Nanaji Deshmukh Plant Phenomics Centre, India. It aimed to explore in a wider context the implementation of a dedicated R&D program and strategies to improve rice cultivation practices using geospatial tools. Mwea and the region can serve as a flagship site to test, implement and disseminate geospatial solutions adapted to rice production. South-south collaboration in that regard is essential, in particular with India. On the one hand India has the largest area (44.6M ha) under rice in the world and is second only to China in terms of rice production, and on the other hand India is also a leader in geospatial technologies and space industry in the global south. See agenda here. 6. Collaboration with the TomorrowNow OSIRIS II project In 2024, and as part of these activities, IRRI initiated a collaboration with the TomorrowNow Osiris II project. We aim to test the use of advanced regional weather historical and forecast datasets produced for East Africa by the OSIRIS II project for improving the rice yield productions delivered by the IRRI/sarmap technology, in Mwea and more generally in Kenya. We are: ● Assessing the value and impact of TomorrowNow’s 4km spatial resolution historical reanalysis weather dataset for modeling rice yield (as compared to the Copernicus AgERA5 25 km dataset) 34 https://docs.google.com/document/d/1kqHwHV7SehbHBrBT8vcMkWPPzQhK2qR6bUTGDe4oF4M/edit?tab=t.0 ● Assessing TomorrowNow’s forecast weather datasets (seasonal 3-6 months) for near real time predicting rice yield. In addition, an Arable MARK III sensor was installed on the NIA perimeter for seed productions at the Mwea irrigation scheme site. The sensor will be used to validate TomorrowNow’s weather datasets (as part of a wider regional network) and to link with ground vegetation crop dynamic with satellite time series, with the spectroradiometric data collected by the Arable sensor. Figure 26. Setting up of the Arable MARK III sensor in Mwea, November, 2024. TomorrowNow.org is a climate tech nonprofit dedicated to empowering 100 million small-scale farmers across Africa with next-generation Agromet advisories by 2030. To support this mission, the Osiris II project, funded by the Bill & Melinda Gates Foundation, was launched as a three-year initiative (Dec 2023 - Dec 2026). The primary objective of Osiris II is to demonstrate the value and enhance the quality of innovative weather and climate data, which are crucial for small-scale farming and for facilitating inclusive, localized climate adaptation across the continent. 7. Conclusions IRRI, in collaboration with sarmap and NIA, conducted rice mapping and monitoring activities in the Mwea Irrigation Scheme as part of the RIICE pilot program in Kenya. The team effectively mapped rice, start-of-season and yield, and generated results in a timely manner. High-quality ground data collected by the field team provided valuable insights into rice cultivation management and practices within the Mwea irrigation scheme. A field experiment was successfully conducted at the NIA-MIAD research field to collect the biomass data for 35 https://www.arable.com/mark3/ http://tomorrownow.org calibrating two dominant rice varieties. The validation of the rice area maps achieved an average of 94% accuracy for 2023 and 2024 main seasons, which is very encouraging for a pilot program. A total of 10,600 and 9,700 ha were found to be cultivated in rice in Mwea during the main season 2023 and 2024, respectively, while the 2024 off season only had 4,500 ha (or 45% of the main season acreage). Planting time was detected to be in the 2nd part of August for the main season, and March for the off season, while RIICE detected well localized planting deviation, as well as differential and managed planting in the western sites, due to water pumping limitation. Yield estimation results were promising compared to the previous year’s yields. Crop cut experiments were conducted, and final yields were compared with CCE data. The RIICE system’s results for both rice mapping and yield estimation are optimistic. This success has led to the further implementation of the RIICE technology in Mwea for 2024 off and main seasons and extension to western irrigation schemes - Ahero and West Kano. References Atera, E.A.; Onyancha, F.N.; Majiwa, E.B.O. Production and marketing of rice in Kenya: Challenges and opportunities. J. Dev. Agric. Econ. 2018, 10, 64–70. Confalonieri, R., Foi, M., Casa, R. et al., 2013. Development of an app for estimating leaf area index using a smartphone. Trueness and precision determination and comparison with other indirect methods. Computers and Electronics in Agriculture, 96, 67-74. Ndirangu, S.N.; Oyange, W.A. Analysis of Millers in Kenya’s Rice Value Chain. IOSR J. Agric. Vet. Sci. (IOSR-JAVS) 2019, 12, 38–47. Bouman, B.A.M.; Kropff, M.J.; Tuong, T.P.; Wopereis, M.C.S.; Ten Berge, H.F.M.; van Laar, H.H. ORYZA2000: Modeling Lowland Rice; International Rice Research Institute, Los Baños, Philippines, and Wageningen University and Research Centre: Wageningen, The Netherlands, 2001. Holecz, F., Barbieri, M., Collivignarelli, F., Gatti, L., Nelson, A., Setiyono, T.D., Boschetti, M., Manfron, G., Brivio, P.A., Quilang, E.J., et al. 2013. An operational remote sensing-based service for rice production estimation at national scale. In: Proceedings of the Living Planet Symposium 2013, Edinburgh, UK, 9-11 September 2013; ESA: Edinburgh, UK. Ministry of Agriculture, Livestock and Fisheries, Government of Kenya. National Rice Development Strategy (2008–2013) Revised Edition 2014; Ministry of Agriculture, Livestock and Fisheries, Government of Kenya: Nairobi, Kenya, 2014; pp. 1–44. Muhunyu, J.G. Is Doubling Rice Production in Kenya by 2018 Achievable? J. Dev. Sustain. Agric. 2012, 7, 46–54 National Rice Development Strategy 2 (2019-2030). Ministry of Agriculture, Republic of Kenya, 2020. 36 Nelson A, Setiyono T, Rala AB, Quicho ED, Raviz JV, Abonete PJ, et al. 2014. Towards an operational SAR-based rice monitoring system in Asia: examples from 13 demonstration sites across Asia in the RIICE project. Remote Sens. 6:10773–10812. Raviz, J., Laborte, A., Gatti, L., Mabalay, M.R., Holecz, F. 2018. Detection of start of season dates of rice crop using SAR and optical imagery, Central Luzon, Philippines. In Proceedings 39th Asian Conference on Remote Sensing: Remote Sensing Enabling Prosperity, ACRS 2018 (Vol. 2, pp. 815-822). Asian Association on Remote Sensing. Setiyono, T.D., Quicho, E.D., Holecz, F.H., Khan, N.I., Romuga, G., Maunahan, A., Garcia C., et al. 2019. Rice yield estimation using synthetic aperture radar (SAR) and the ORYZA crop growth model: development and application of the system in South and South-east Asian countries. International Journal of Remote Sensing. 40 (21): 8093-8124 https://doi.org/10.1080/01431161.2018.1547457. Tan, J., and Y. Luo. 2016. “Global Sensitivity Analysis of Outputs over Rice-Growth Process in ORYZA Model.” Environmental Modelling and Software 83:36-46. doi:10.1016/j.envsoft.2016.05.001. USDA_PS&D Online, PSD Data Sets, Last Updated Aug/12/2020. Available online: https://apps.fas.usda.gov/psdonline/app/ index.html#/app/downloads (accessed on 13 August 2020). 5. 37 https://doi.org/10.1080/01431161.2018.1547457 Annexure Annex 1. LAI data from 15 selected fields during the main season 2023 (a) and 8 fields in 2024 (b) in Mwea Irrigation Scheme and 5 fields during main season 2024 in Ahero (c ) (a) Main season, 2023 Field ID LAI 16-Sep 28-Sep 10-Oct 23-Oct 03-Nov 354 2.25 2.79 3.04 3.82 4.07 361 2.05 4.02 4.05 4.19 4.13 368 2.78 4.11 4.17 4.38 4.27 375 2.41 3.15 3.55 3.75 4.11 382 1.52 2.96 3.19 3.54 3.93 389 2.10 2.97 3.34 3.78 3.92 396 1.64 2.65 2.84 3.42 3.69 403 1.69 3.12 3.48 3.80 4.01 410 2.19 3.04 3.38 3.64 4.12 417 1.98 2.94 3.66 3.68 3.97 424 2.17 1.66 2.65 3.37 3.86 431 2.07 2.72 3.23 3.69 3.98 438 1.76 2.25 2.74 3.48 3.92 445 1.90 2.84 3.34 3.57 4.03 452 2.02 2.21 2.92 3.63 3.87 (b) main season, 2024 38 Field ID LAI 10-Sep 23-Sep 04-Oct 17-Oct 25-Oct 361 1.09 2.95 2.79 3.87 4.59 403 1.86 2.18 3.44 4.38 4.45 417 0.78 2.32 2.82 3.41 3.89 445 0.55 2.05 3.13 3.63 3.42 2021 1.30 2.66 3.33 4.02 4.36 2028 2.52 3.31 3.38 4.05 4.30 2035 1.70 3.71 4.13 4.19 4.37 2045 1.15 2.26 3.13 3.66 4.13 (c ) Main season 2024, Ahero Field ID LAI 16-Sep 28-Sep 23-Oct 04-Nov 405 0.91 1.02 2.07 3.26 412 0.34 0.37 2.70 3.07 419 1.14 2.07 2.68 2.83 426 0.13 0.55 2.32 3.15 433 0.53 1.02 2.50 2.92 39 Annex 2. Program of activities during the online training on field data collection. Date Topic August 22, 2023 Guidelines for field data collection for calibration ● Aim ● Timing ● Equipment needed ● Number of fields to sample ● Criteria for field selection ● Data to be collected GPS device setting and transferring to BaseCamp ● Overview of Global Positioning System (GPS) ● Importing .gpx file ● Exporting data Data collection using Kobo Toolbox ● Introduction to Kobo Toolbox ● Setting up mobile device to use kobo forms ● Creating shortcut of Kobo forms on mobile device ● Collecting data and saving the form ● Submitting forms August 25, 2023 Mapping and viewing geotagged photos in QGIS ● Mapping geotagged photos ● Viewing photos in pop-up ● Showing ● direction of photos ● Displaying photos as markers Guidelines for field data collection for validation ● Aim ● Timing ● Equipment needed ● Number of fields to sample ● Criteria for field selection ● Data to be collected 40 August 25, 2023 Guidelines for LAI data collection for yield mapping ● Aim ● Timing ● Equipment needed ● Number of fields to sample ● Criteria for field selection ● Data to be collected 41 Annex 3. Participants and their affiliations during the online training on field data collection Name Institution Gender Simon Njau Kariuki NIA Male David Aleri NIA Male Faith Mwende NIA Female Kipngetich Vincent NIA Male Kelvin Wafula IRRI Male ryansmwangi NIA Male Saidi Mboya NIA Male Fjuma NIA Female Kevin Ochieng NIA Male Charles Odago NIA Male Labangwaro NIA Male Sisseybrenda NIA Female Jeny Raviz (Trainer) IRRI Female Sushree Satapathy (Trainer) IRRI Female Renaud Mathieu (Project lead) IRRI Male 42 Annex 4. Participants during the Stakeholders meeting, September 2023 Name Country Title Institution Dr Francesco Holecz Switzerland CEO sarmap Dr Renaud Mathieu India Scientist Geospatial Science CGIAR IRRI Dr Sushree Satapathy India Senior Specialist Crop Modelling CGIAR IRRI Dr Daniel Jimenez (online) France Scientist Big Data Science CGIAR CIAT Dr Daniel Menge Nairobi Scientist Seed System CGIAR IRRI Godwin Githinji Kuria Nairobi Senior Superintending Engineer Ministry of Agriculture Dr. Wilson Oyange Nairobi Project manager Mwea Irrigation Scheme Tom Dienya Nairobi Nat. Project Coordinator CountrySTAT & Crops Insurance Ministry of Agriculture Raphael Kitonyi Nairobi Rice promotion program - Agribusiness officer Ministry of Agriculture Vincent Koskei Mwea Senior Research Officer - Manager Irrigation Scheme NIA David Aleri Mwea Agronomist NIA Charlotte Ooro Kisumu Officer - Ahero Irrigation Scheme NIA Dr Kennedy Were Nairobi Chief Research Scientist Geospatial KALRO Dr Japheth Wanyama Kitale Institute Director KALRO Food Crops Research Institute Enock Sing'oei Nairobi Head Insurance Market Development & Digital Innovations Syngenta Foundation Wycliffe Kiplagat Nairobi Data and Actuarial Lead Syngenta Foundation Evans Ochieng Nairobi Actuarial and Digital Innovation Associate Syngenta Foundation Jackline Muthengi Nairobi Pricing and Product Manager ARC Ndege Muriuki Mwea Chairman Board Director Mwea Rice Growers Multipurpose Cooperative Joseph Makanga Nairobi Rep RCRMD Josphat Mutwiri Nairobi Rep Kenya Plant Health Inspectorate Service Simon M. Gachuiri Nairobi Rep Kenya Meteorology Department 43 John Mburu Nairobi Manager Agricultural Statistics Kenya National Bureau of Statistics Invited Name Country Title Institution Dr Mary Mutembei Nairobi Head Rice Promotion Program Ministry of Agriculture Rep Nairobi Ministry of Environment, CC, and Forestry Joel Tanui Nairobi Deputy general manager - Operations NIA Vincent Kabuti Nairobi Deputy general manager - Planning NIA Dr Ruth Musila Mwea Research Scientist, Center Director KALRO Mwea KALRO Dr Lusike Wasilwa Nairobi Deputy Director General, KALRO KALRO Crop Systems Ken Fujie Nairobi General Coordinator CARD Dr. John Gachara Mwea County Executive Committee Member Agriculture, Livestock & Fisheries, Kirinyaga County Dr Mercy Mwaniki Nairobi Dept Geomatic Eng & GIS JKUAT Prof Charles Gachene Nairobi Dept Soil Science UoN 44 1.​Introduction 2.​Data and Methods 2.1.​Study area 2.2.​Data 2.3.​Field experiment 2.4.​Methods 3.​Results 4.​Capacity building 5.​Stakeholder engagement meeting for developing RIICE, Geospatial and Digital activities in Kenya 7.​Conclusions Annexure