Research Paper Decision support for managing an invasive pathogen through efficient clean seed systems: Cassava mosaic disease in Southeast Asia☆ Kelsey F. Andersen Onofre a,b,c,d,*, Erik Delaquis e, Jonathan C. Newby e, Stef de Haan f,k, Cu Thi Le Thuy g, Nami Minato h, James P. Legg i, Wilmer J. Cuellar j, Ricardo I. Alcalá Briseño a,b,c,1, Karen A. Garrett a,b,c,* a Plant Pathology Department, Institute of Food and Agricultural Sciences, University of Florida, Gainesville, FL, USA b Global Food Systems Institute, University of Florida, Gainesville, FL, USA c Emerging Pathogens Institute, University of Florida, Gainesville, FL, USA d Department of Plant Pathology, Kansas State University, Manhattan, KS, USA e International Center for Tropical Agriculture (CIAT), Vientiane, Laos f International Potato Center (CIP), Lima, Peru g International Center for Tropical Agriculture (CIAT), Hanoi, Viet Nam h Institute of Science and Technology, Niigata University, Niigata, Japan i International Institute of Tropical Agriculture (IITA), Dar es Salaam, Tanzania j International Center for Tropical Agriculture (CIAT), Cali, Colombia k Biosystematics Group, Wageningen University and Research (WUR), Wageningen, The Netherlands H I G H L I G H T S G R A P H I C A L A B S T R A C T • Seed systems must distribute improved varieties and slow the spread of pathogens. • Cassava mosaic disease threatens crop production in Southeast Asia. • We developed a meta-population network model of disease spread in seed systems. • We evaluated scenarios for seed system management to slow disease spread. • Consistent management of regions initially highly affected was often most effective. A R T I C L E I N F O Editor: Val Snow Guest Editor: Tjeerd Jan Stomph Keywords: Clean seed A B S T R A C T Context: Effective seed systems must distribute high-performing varieties efficiently, and slow or stop the spread of pathogens and pests. Epidemics increasingly threaten crops around the world, endangering the livelihoods of smallholder farmers. Responding to these challenges to food and economic security requires stakeholders to act ☆ This article is part of a Special issue entitled: ‘Seed Systems’ published in Agricultural Systems. * Corresponding authors. E-mail addresses: andersenk@ksu.edu (K.F. Andersen Onofre), karengarrett@ufl.edu (K.A. Garrett). 1 Current address of R. I. Alcalá Briseño: Department of Plant Pathology and Environmental Microbiology, Pennsylvania State University, State College, PA, USA. Contents lists available at ScienceDirect Agricultural Systems journal homepage: www.elsevier.com/locate/agsy https://doi.org/10.1016/j.agsy.2025.104435 Received 13 August 2024; Received in revised form 6 May 2025; Accepted 17 June 2025 Agricultural Systems 229 (2025) 104435 Available online 15 July 2025 0308-521X/© 2025 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ ). mailto:andersenk@ksu.edu mailto:karengarrett@ufl.edu www.sciencedirect.com/science/journal/0308521X https://www.elsevier.com/locate/agsy https://doi.org/10.1016/j.agsy.2025.104435 https://doi.org/10.1016/j.agsy.2025.104435 http://crossmark.crossref.org/dialog/?doi=10.1016/j.agsy.2025.104435&domain=pdf http://creativecommons.org/licenses/by/4.0/ Plant disease epidemics Emerging pathogens Machine learning Participatory modeling Cassava mosaic disease quickly and decisively during the early stages of pathogen invasions, typically with limited resources. A current threat is the introduction of cassava mosaic virus in Southeast Asia. Objectives: Our goal in this study is to provide a decision-support framework for efficient management of healthy seed systems, applied to cassava mosaic disease. The specific objectives are to (1) evaluate disease risk in disease- free parts of Cambodia, Lao PDR, Thailand, and Vietnam; (2) incorporate estimated risk of disease establishment with seed exchange survey data and whitefly spread in the landscape to model epidemic spread; and (3) identify candidate regions to be prioritized in seed system management. Methods: We used machine learning to integrate disease occurrence, climate, topology, and land use, and network meta-population models of epidemic spread. We used scenario analyses to identify candidate priority regions for management. Results and conclusions: The analyses allow stakeholders to evaluate strategic options for allocating their resources in the field, guiding the implementation of seed system programs and responses. Consistently targeting initially high priority locations with clean seed produced more favorable outcomes in this model, as did prioritization of a higher number of districts for the deployment of smaller volumes of clean seed. Significance: The decision-support framework presented here can be applied widely to seed systems challenged by the dual goals of distributing seed efficiently and reducing disease risk. Data-driven approaches support evidence-based identification of optimized surveillance and mitigation areas in an iterative fashion, providing guidance early in an epidemic, and revising recommendations as data accrue over time. 1. Introduction Effective seed systems should both distribute good crop varieties and limit the dispersal of crop pathogens and pests. From a pragmatic perspective, given the strongly informal nature of many seed systems globally, effective intervention often requires integrated collaboration in the seed system to build on the strengths of existing structures. A major destabilizing force in seed systems is the introduction of invasive plant pathogens, which are likely to inflict ever-greater damage as planted area and density of crops increase globally (Ristaino et al., 2021). Fueled by globalized trade, growing cropland connectivity, and climate change, crop pest and disease outbreaks have increased in important crop production areas of Asia since the mid-20th century (Oerke, 2006; Wang et al., 2022). In the last decade, Southeast Asia’s Greater Mekong Subregion (Cambodia, Lao PDR, Myanmar, Thailand, Vietnam, and the Southern Chinese provinces of Guanxi and Yunnan) has faced transboundary pest and pathogen invasions in numerous major economic crops, including fall armyworm in maize (Nagoshi et al., 2020), Fusarium oxysporum f. sp. cubense Tropical Race 4 in banana (Zheng et al., 2018), and cassava mosaic viruses (Siriwan et al., 2020). These crops represent the livelihoods of tens of millions of smallholders across the region, and the bulk of agricultural export balances that drive rural economies. The latter two diseases in particular require effective, integrated seed systems to minimize seed-borne disease spread and disseminate resilient or resistant varieties, safeguarding critical agri cultural systems in the region. In the Greater Mekong Subregion, smallholder farmers often produce cassava as an industrial cash crop on marginal land (Delaquis et al., 2018; Graziosi et al., 2016), making it important for the livelihood of rural populations. Production is threatened by cassava mosaic viruses (Geminiviridae, Begomovirus), causal agents of cassava mosaic disease (CMD). Cassava mosaic viruses are dispersed via infected cassava planting stems, and the widely distributed whitefly Bemisia tabaci (Homoptera, Aleyrodidae). Frequently recognized as one of the most important plant diseases in the tropics, the first report of CMD in Southeast Asia originated from a single cassava plantation in Cambodia in 2015 (Wang et al., 2016). Subsequently, the disease was reported in Vietnam (Uke et al., 2018), China (Wang et al., 2020; Wang et al., 2019), Thailand (Leiva et al., 2020), and Lao PDR (Siriwan et al., 2020). The species of cassava mosaic virus found in the Greater Mekong subregion, Sri Lankan cassava mosaic virus (SLCMV), was previously only reported on the Indian subcontinent, and is reported to be more virulent than Indian cassava mosaic virus (ICMV) (Saunders et al., 2002). Although only recently reported in Southeast Asia, cassava mosaic vi ruses have been a major constraint to cassava production for many years in Africa (Legg, 1999; Legg and Fauquet, 2004; Rey and Vanderschuren, 2017) and the Indian subcontinent (Saunders et al., 2002). In Africa, losses have been estimated to be greater than 1 billion USD per year (Rojas et al., 2018), with the most severe pandemic outbreak occurring in the 1990s (Fargette et al., 2006; Legg and Fauquet, 2004). Unlike in sub-Saharan Africa, where CMD was problematic for many years prior to reaching pandemic levels in the 1990s (Thresh and Cooter, 2005), introduction in Southeast Asia has been sudden and spread has been rapid. The arrival of an emerging pathogen in a new region forces national programs tasked with production of clean planting material and plant disease surveillance to make complex decisions. For CMD, one of the most important management strategies is the deployment of disease- free, ‘clean’ planting material. Due to low clonal field multiplication rates, limited volumes of vegetative planting stems can be produced by national or donor-funded programs. This raises the question: which strategies maximize the impact of the relatively small amounts of available clean planting material disseminated through integration with the existing seed systems? To maximize epidemic mitigation in the landscape, it is necessary to calibrate volumes of planting material distributed, and the number and geographic location of sites to which they should be distributed. Current strategies are largely informed by local or national disease incidence, with decision making largely undertaken in the absence of compre hensive efforts to evaluate potential strategies or to visualize likely impacts of those strategies embedded in a larger regional context. There is a need for practical approaches to integrate multiple elements of the disease spread cycle to provide evidence-based recommendations for seed system interventions. Seed system management can benefit from adapting concepts from crop epidemiology, such as risk-based surveillance and management (Cameron, 2012; Carvajal-Yepes et al., 2019; Stark et al., 2006). Like wise, seed system management is a key component of ‘disaster plant pathology’, since the introduction of new pathogens through seed after disasters such as droughts or floods is an important risk factor (Etherton et al., 2024). Risk-based approaches are especially critical early in a pathogen invasion, when data are limited but effective mitigation or eradication remains possible (Parnell et al., 2014). Optimizing data- driven management strategies early in an emerging outbreak is an important challenge (Epanchin-Niell et al., 2012). Early intervention is often necessary for effective eradication of an invading pathogen (Cunniffe et al., 2016), and heuristic approaches are being developed for the best strategies for managing and surveying invasive species spreading in networks such as seed systems (Chadès et al., 2011). Models have been developed for disease control in other cassava systems that are largely driven by informal planting material exchange (McQuaid et al., 2017b; McQuaid et al., 2017a) and there is the potential to K.F. Andersen Onofre et al. Agricultural Systems 229 (2025) 104435 2 integrate both seed networks and socio-economic networks of growers and institutions in management scenario analysis (Etherton et al., 2023; Garrett, 2021). Here we use a modeling framework incorporating machine learning to provide decision support for seed system management. We model environmental risk in an epidemic metapopulation network model of pathogen spread through the landscape to evaluate the likely impact of seed system management and disease surveillance campaigns. We draw on tools previously described in Andersen et al. (2019), to integrate both environmental risk for disease establishment – defined here as risk of dis ease establishment in the landscape due to weather, climate, and land use factors favorable for disease – and risk of CMD spread due to cassava seed system patterns and whitefly movement (dispersal risk). Network analysis has frequently been used to model the risk of pathogen spread and to identify candidate locations for sampling (e.g., Andersen Onofre et al., 2021; Buddenhagen et al., 2017; Garrett, 2021; Garrett et al., 2018; Harwood et al., 2009; Martinetti and Soubeyrand, 2019; Moslonka-Lefebvre et al., 2011; Pautasso, 2015; Pautasso and Jeger, 2008; Sanatkar et al., 2015; Shaw and Pautasso, 2014; Sutrave et al., 2012). Network analysis is particularly suited to dispersal in seed sys tems where nodes are geographic land units and links are the movement of planting material (and potentially vectors) between them. We draw on data from an empirical survey of cassava planting stem exchange in the Greater Mekong Subregion (Delaquis et al., 2018) to estimate network trade patterns. We present a decision support framework for seed system manage ment that integrates environmental geospatial data layers in a multi layer network model, to evaluate and map seed system outcomes in terms of CMD risk in Cambodia and Vietnam, and in neighboring Thailand and Lao PDR. The objectives of this study are to (1) evaluate disease risk in non-infected parts of Cambodia, Lao PDR, Thailand, and Vietnam by integrating disease occurrence, climate, topology, and land use, using machine learning; (2) incorporate estimated environmental risk for disease establishment with seed exchange survey data and whitefly spread in the landscape to model epidemic spread in a network meta-population model; and (3) use scenario analysis to identify candidate regions to be prioritized for the deployment of clean seed material for management of CMD in Southeast Asia. 2. Methods Our analysis had three stages. First, we estimated disease risk as a function of environmental variables, comparing a set of machine learning methods. Second, we estimated a network of seed exchange through the system, which could spread pathogens. Third, we combined these components in scenario analyses to evaluate the likely influence of management options on pathogen spread. In the first component of our analysis, we used environmental predictors, coupled with disease occurrence and absence data points from surveys conducted in Cambodia and Southern Vietnam, to evaluate the relative likelihood of occurrence based on environmental correlates. We drew on methods developed for species distribution modeling and invasive species dis tribution modeling. We extrapolated to neighboring Thailand and Lao PDR. This portion of our analysis assumed that there are environmental differences between locations where CMD establishes and locations where it does not, and that these differences are based on the ecological niche preferences of the pathogen and vector, not on the lack of dispersal to these regions. The assumption that dispersal limitations do not contribute to absence may not be met completely, as we discuss below. We compared several machine learning methods (described in detail below) to identify the best fitting model. We then incorporated this estimated environmental risk raster layer (as an estimated probability of establishment), described by the best fitting model, into simulation analyses of CMD spread. The resulting simulation model incorporates not only dispersal risk via trade and whitefly dispersal, but also the likelihood that the environment will be favorable for establishment. 2.1. Invasive species distribution model of CMD establishment 2.1.1. Geographic distribution of disease (response variables) SLCMV occurrence and absence data were collected from a survey conducted in 2016 in Cambodia and Vietnam (Minato et al., 2019). In brief, the virus survey was conducted in parallel with a baseline survey designed to understand farmer demographics and the regional move ment of cassava planting material (Delaquis et al., 2018). In total, 419 fields were sampled from 15 districts (8 provinces) of Vietnam and 16 districts (11 provinces) of Cambodia. Districts were selected by priori tizing areas under intensive cassava cultivation. In addition to survey data from Minato et al. (2019), incidence data collected from 2016 to 2018 were also incorporated from the online repository https://pestdisp lace.org/ associated with various CMD surveillance projects (Cuellar et al., 2018). Observations were also obtained from government sur veillance campaigns in Vietnam in 2018. In total, 1022 PCR-confirmed presence/absence data points were obtained from Cambodia, Vietnam, and Thailand in 2016–2018. 2.1.2. Predictor selection and dimension reduction To model the influence of environmental predictors on CMD estab lishment and spread in the landscape, and to evaluate the risk of path ogen establishment in regions not yet surveyed, a series of bioclimatic, topographic, and land use predictors were assembled. Predictors were obtained from open-source repositories whenever possible. A set of 19 spatially interpolated gridded climate predictors, or climate surfaces, were obtained from WorldClim Version 2.0 (Fick and Hijmans, 2017) in addition to monthly climate averages (temperature maximum, temper ature minimum, solar radiation, average precipitation, and windspeed). Elevation data were obtained from a digital elevation map (DEM) (Reuter et al., 2007). Cassava planting area data were obtained at the province or district level for Cambodia, Vietnam, Lao PDR, and Thailand from the national agricultural ministries of each country. In addition, environmental land cover estimates were obtained for 2015 from the European Space Agency (ESA ESA, 2017) Climate Change Initiative (CCI) land cover project, from which percentage of area falling into land use categories (cropland, tree cover, shrubland, urban, etc.) were calculated. The predictors used in this analysis are described in Sup plement 1. All data were obtained as raster grids and managed in R using the packages raster (Hijmans, 2019), sf (Pebesma, 2018), rgdal (Bivand et al., 2019), and rgeos (Bivand and Rundel, 2019). All predictors were extracted from these raster layers for each of the CMD presence/absence geolocations. Models incorporating environmental variables typically encounter multicollinearity, which can lead to model over-fitting and poor model projection to new regions. We used the variance inflation factor (VIF) to reduce our predictor set to those that had a VIF below our defined threshold (20). This threshold was selected to avoid discarding too many predictors that may be biologically relevant, while still reducing the dimension of the dataset. This multicollinearity analysis was implemented using the vifstep function in the R package usdm (Naimi et al., 2014). 2.1.3. Classification models Machine learning has been used for species distribution models in ecology for many years (e.g., Drake et al., 2006; Elith and Leathwick, 2009; Lorena et al., 2011; Olden, 2008), and more recently in plant pathology to predict disease occurrence as a function of weather or climate predictors (e.g., Harteveld et al., 2017; Martinetti and Sou beyrand, 2019; Shah et al., 2014). There is great potential to predict the future geographic spread of emerging pathogens prior to the availability of complete information about epidemiology in the region (Jiménez- Valverde et al., 2011; Meentemeyer et al., 2008). Classic species distri bution models are often used to model the global (or fundamental) niche of a species under the assumption that there is equilibrium in the landscape, meaning that the species has occupied all potential niches, K.F. Andersen Onofre et al. Agricultural Systems 229 (2025) 104435 3 https://pestdisplace.org/ https://pestdisplace.org/ and those that are unoccupied are not environmentally conducive for establishment. The principal problem with invasive species distribution modeling is that it is often difficult to differentiate regions where the pathogen is absent due to lack of niche suitability from those where the pathogen is absent due to lack of introduction or dispersal (Gallien et al., 2012; Jarnevich et al., 2015; Jiménez-Valverde et al., 2011; Mainali et al., 2015). Researchers have proposed solutions for this by limiting the geographic range for absence datapoints (Mainali et al., 2015; Narouei-Khandan et al., 2016), or by incorporating dispersal compo nents (Meentemeyer et al., 2008). Using the restricted set of predictors identified in our analysis of multicollinearity (Table 1), we compared machine learning algorithms – random forest (RF) (Breiman, 2001) and support vector machines (SVMs) (Guo et al., 2005) – for identifying the best fitting model to correctly classify the disease occurrence and absence data points. We also compared logistic regression (LR). 2.1.4. Machine learning RF, SVMs, and LR were implemented in the mlr package in R (Bischl et al., 2016). All models were evaluated through k-fold cross validation (10-fold), repeated 50 times. The model was evaluated using average AUC, TSS, and Cohen’s Kappa statistics, common metrics for evaluating ecological species distribution model accuracy (Allouche et al., 2006) (Table 2). 2.1.5. CMD risk Once the best model was established (based on AUC, TSS, and Cohen’s Kappa), it was used to project predicted probability of estab lishment to the remainder of the region. Predictions were made based on a random gridded sample of the uninfected region, 3000 geolocations in total, throughout unsampled regions of Cambodia and Vietnam, as well as Thailand and Lao PDR, two important neighboring regions where cassava is produced. The average probability per district was evaluated and denoted as risk of establishment, ε. A scaling parameter, α, was added to simulation models (described below) to modify the environ mental influence on establishment. Districts where no planted cassava area was reported in the previous three years were not included in this analysis. 2.2. Network model of pathogen spread Environmental conduciveness for establishment (described above) is only one component of the risk of pathogen spread. Dispersal, through the movement of virus-infected planting material in seed systems and via the movement of viruliferous whiteflies, also plays an important role in determining where CMD is present in the landscape. We estimated the relative likelihood of both modes of dispersal, as described below, and combined them for use in our simulation experiments. Once the simu lation model was constructed, scenarios for clean seed deployment were tested. The epidemic model we implemented builds on Andersen et al. (2019), using a discrete time network SI (susceptible-infected) Markov chain, with the following changes to adapt the model to this system. First, we incorporated within-node disease spread dynamics as a func tion of transmission rate (β) between healthy and infected planted area (hectares) within a district (node) and between neighboring districts, representing local spread via whiteflies (Fig. 1). Second, the probability of infection was not based solely on the probability of transmission, but also on the probability of establishment, ε (or influence of environment, described above). Finally, for management/intervention scenario anal ysis we introduced incomplete recovery as a function of the proportion of healthy cassava planting material introduced into select nodes at the start of the season. 2.3. Model components 2.3.1. Seed system network We fit a network model of planting stem exchange in the region (Cambodia, Vietnam, Lao PDR, and Thailand) where nodes are aggre gated to the scale of districts (sub-provincial administrative units) and links are directed and calculated based on the probability of stem ex change between nodes (districts). Existing data on farmer behavior related to planting material exchange (Delaquis et al., 2018) were used to estimate link probabilities. The probability of a link forming between any two districts i and j (λij) was estimated based on these household survey results. The frequency with which a link was reported between two districts for each of a set of distance ranges (0–100, 100–300, and 300–500 km) was calculated, and then the product of this frequency and the total number of districts at each range was taken as the probability that a link would form in each season. This incorporates a low likelihood Table 1 Predictors retained after variable inflation factor analysis to reduce multicollinearity. Variable Type Name Reference VIFa Climate Minimum temperature in May (◦C) WorldClim 2.0 (Fick and Hijmans, 2017) 10.6 Maximum temperature in June (◦C) 7.8 Solar radiation in June 10.3 Solar radiation in September 8.9 Solar radiation in October 7.5 Precipitation in March 11.0 Precipitation in April 14.2 Precipitation in May 13.8 Precipitation in June 19.0 Precipitation in August 19.8 Precipitation in September 17.1 Precipitation in October 10.7 Precipitation in December 11.7 Isothermality (BIO2/BIO7) (* 100) (Bio3) 10.9 Precipitation of Driest Month (Bio14) 12.8 Precipitation Seasonality (Coefficient of Variation) (Bio15) 14.3 Precipitation of Driest Quarter (Bio17) 17.8 Precipitation of Warmest Quarter (Bio18) 5.1 Precipitation of Coldest Quarter (Bio19) 5.2 Land Use Percentage water bodies (Agland5) ESA, 2017 1.2 Percentage tree cover (Agland6) 1.7 Percentage shrub or herbaceous cover (Agland7) 1.9 Percentage mosaic cropland (Agland8) 1.4 Percentage urban areas (Agland10) 1.0 a Variance inflation factor – a metric to detect collinearity through a stepwise procedure. A large VIF indicates a collinearity issue for that particular variable. For variable selection in this study, 20 was set as a cutoff and variables with a VIF > 20 were removed. Table 2 Classification model performance. Method AUCa TSSb Cohen’s Kappac Random Forest 0.94 0.52 0.62 Logistic regression 0.80 0.22 0.30 SVMs 0.83 0.38 0.45 a AUC- area under the curve. b TSS – true skill statistic (sensitivity + specificity – 1). c Cohen’s Kappa – Agreement between observed and expected, accounting for agreement expected by chance. K.F. Andersen Onofre et al. Agricultural Systems 229 (2025) 104435 4 that a district would trade with each neighbor, but a district would tend to trade with some neighbors in each season. No trade events were observed at >500 km distance, so probabilities of exchange between nodes >500 km apart were set to zero. 2.3.2. Short-distance whitefly dispersal CMD spread in the landscape occurs not only via seed systems, but also via the local movement of viruliferous whiteflies. The bulk of whiteflies disperse anisotropically in the landscape up to a maximum of 2 km from their source (Byrne et al., 1996), with a majority not migrating from their source. Because the nodes in our analysis represent districts, we consider whitefly movement only between adjacent dis tricts within the month-long timesteps of the model. In the next time step, new generations of whiteflies may emerge and infect adjacent districts. Movement of whiteflies between districts occurs in the model during 10 timesteps during the season corresponding to a roughly 10- month cassava growing season. This process continues until the end of the season, at which time the final amount of infected area is maintained at the beginning of the next season. 2.4. Simulation model In the simulation model, each node (district in Vietnam, Cambodia, Lao PDR, and Thailand which reported cassava planted area in 2018, 1255 in total) begins with a fixed area of cassava (ha), corresponding to the area of planted cassava reported for the district in 2018 (CIAT 2018). Additional starting conditions for the simulation include CMD infection status of each district, area (ha) infected with CMD (Dit), and corre sponding area uninfected (Hit). All districts reported to have CMD infection (based on previous survey results, Ministry of Agriculture re ports, and the opinion of experts from the region) were designated as infected. In areas where CMD was only reported at the scale of the province, three districts with the highest number of planted hectares were designated as CMD positive. In total, 125 districts were CMD positive at the start of the simulations, reflecting the estimated CMD prevalence at the close of the 2019 season. Districts were each assigned a conservative estimate of 1 % incidence at the start of the simulation (1 % of area (ha) infected). The infection process in node i by which a portion of previously healthy hectares of cassava becomes infected during a month-long timestep results in a new value of area infected by CMD for the new timestep (Di,(t+1)) is described as Di,(t+1) = Dit + ((β*ε)(Dit*Hit) + (β*ε*α) (Nit*Hit) ) (3-1) where Dit is the number of infected hectares in a district in district i in month t, Hit is the number of healthy hectares in district i in month t, Nit is the sum of infected hectares in all districts adjacent to district i (external inoculum sources), β is the transmission rate of infection from infected to uninfected hectares via whitefly spread, ε is a parameter describing the environmental conduciveness to disease (based on the species distribution modeling), and α is the relative influence of neigh boring inoculum compared to within-district inoculum. The healthy area in the next timestep (month) was Hi,(t+1) = max ( 0,Hit–Di,(t+1) ) . (3–2) At the close of each timestep, Nit is the sum of infected hectares in districts adjacent to district i (external inoculum sources), calculated as Nit = ∑ j∋Vj∈Mi Dj,(t− 1) , (3–3) where Vj is node j and Mi is the set of nodes linked to node i. This simulation is carried out over the course of 10 timesteps, corresponding to the 10-month typical cassava growing season in this region. At the end of each season t, a set of ending conditions for each district was obtained. At this point a network model of the seed system was implemented, simulating the seed exchange that takes place prior to planting at the start of a new season (t + 1). Exchange of planting ma terial occurs across network trade links (where λij, the probability of trade between districts i and j, is used to stochastically generate trade events as described above). If a trade event occurs from node i (the source node) to node j, the relative likelihood that the material moved is infected with CMD was set to 10 if node i was infected. If the source node i was uninfected, there was no chance of transmission from node i to node j. This newly infected area (hectares) at sink node j (set to 5 ha) is then summed with the infected hectares of node j (Dit) at the conclusion of the last season to generate a new starting value for the subsequent season (and the corresponding Hjt is also updated). Five hectares was Fig. 1. Network metapopulation model schematic for cassava mosaic virus spread through a dynamic seed system. The network illustration on the left side of the figure represents cassava seed trade links at the start of the season. If cassava planting material is moved from an infected district, districts to which infected material is moved can become infected (the ‘sink’ district). The box on the right side of the figure illustrates within- and between-node spread dynamics during the course of the season. Infected planting material can infect healthy cassava areas within a district (leading to within-season, within-district spread), but can also infect healthy cassava areas in adjacent districts. Within-season spread represents local whitefly transmission in the model. Between-node links (representing cassava seed trade) only occur at the initiation of a season. Within-node spread, depicted in the breakout box, occurs over the course of 10 monthly timesteps within a season. K.F. Andersen Onofre et al. Agricultural Systems 229 (2025) 104435 5 chosen as the trade volume based on previous survey results where the average number of stems per transaction was 11,000 (Delaquis et al., 2018). On average it takes approximately 2000 stems to plant a hectare of cassava in this region, so five hectares was determined to be a reasonable area to be introduced by trade. For this subsequent season, after seed exchange, a new round of within-season simulations is started, as described above. At the end of season t, a new Di,(t+1) and Hi,(t+1) are stored for each node, and are used as the starting conditions for the next season, in which the process is repeated. Simulations over 10 seasons with 1000 realizations per parameter combination were evaluated in terms of the total number of hectares infected in the region and the total number of districts infected. 2.4.1. Uncertainty A sensitivity analysis (uncertainty quantification) was conducted for model parameters α and β. The transmission parameter, β, represents the rate at which uninfected hectares become infected by within-district infected hectares and neighboring infected hectares, reflecting the spread of CMD via infected whiteflies. Because insufficient data were available to estimate this parameter, a range of values was evaluated (0.0002, 0.00002, 0.000002, 0.0000002). Additionally, a range of values were evaluated for the parameter α (0.1, 0.3, 0.5, 0.7, 0.9), which modifies the effect of the inoculum of neighboring districts in the model. Model output was evaluated for each pairwise combination of these parameter values. 2.5. Simulation experiments Using the above-described network model, we carried out simulation experiments with corresponding sensitivity analyses to address the following key questions: 1) how does completely restricting stem trade influence the spread of CMD in the region, 2) how does incorporating clean seed into the system each year impact spread, as a function of clean seed volume, and 3) how should locations for clean seed in terventions be selected to maximize reduction in infected area? 2.5.1. Experiment one: seed trade restriction A scenario with no stem exchange was evaluated. In this “trade restricted” simulation, planting material that was infected at the end of the previous season (t - 1) was used as the starting amount of planting material for the current season (t), over the course of ten seasons. Like all scenarios in this study, we assumed no positive selection or reversion and thus no reduction in infected area from year to year. This set of simulations was deterministic. The percent reduction in infected area across all districts and total number of infected districts was calculated by comparing the outcomes from the simulations described above for “full trade” with these “trade restricted” scenarios. 2.5.2. Experiment two: clean seed intervention strategies Because regional capacity for clean stem production is limited, an urgent and practical question is: What are the optimal locations, and corresponding volumes, for clean cassava planting material deploy ment? In this region, it typically takes 10,000 cuttings to replant one hectare of cassava (where five cuttings are obtained per stem for most varieties). In 2019, Vietnam’s government reported over 31,000 ha infected by CMD with varying levels of severity, requiring >300,000,000 cuttings to replace this area with clean planting material. Producing this volume of clean seed material is prohibitively costly and logistically challenging, so prioritizing regional deployment strategies is necessary to maximize the impact of available resources. We introduce “management” into the cassava system model in terms of stems certified disease free, from stock originating in tissue culture and bulked in the field where there was no risk of infection. After the exchange event in the model, as part of the start of the season (t), a set volume of added clean planting material is summed with the clean material already present in a district to be managed, and removed from the proportion of infected hectares Di,(t+1) for the district from the end of the previous season. The assumptions of the model are that: 1) the proportion of infected planting material from the previous year is otherwise retained, except for what is added as clean planting material, 2) the hectares with infection at the end of the previous season is equal to the proportion of infected hectares at the start of the next season (farmers do not change their cassava area from season to season, and farmers do not employ positive or negative selection on saved planting material), 3) clean stems will completely replace a proportion of infec ted hectares at the start of the season, and 4) cassava plants from clean stems can become infected during the season at the same rate as all other cassava plants. We tested three methods for selecting locations to manage: 1) “random selection in infection zones” where cassava-producing districts that fall in infection zones (infected themselves and/or within a 50 km buffer of infected districts) are randomly selected in each season, 2) "consistent management" across seasons of districts selected within the same infection zones as described in scenario one based on their starting area of diseased material (those with highest area selected first), and 3) "reactive management" of locations selected each season, prioritized based on their infected area in the previous season, and again with areas with the most infection managed first. We not only wanted to under stand the best strategy for managing locations, but also the optimal number of locations and volume to distribute to each location. We tested managing 20, 40, 60, and 80 districts, with each combination of 20, 40, 60 and 80 ha of clean planting stems. For example, for one treatment combination, 20 locations would be managed with 20 ha each of clean planting stems, and so on. Each management scenario (combination of (a) method of selecting locations, (b) number of locations, and (c) vol ume of disease-free stems) was evaluated, and scenarios were compared for their utility in slowing the spread of the pathogen through the region over time. We implemented these management scenarios for the two most conservative parameter estimates: β = 0.000002 and α = 0.1. The code for examples of these analyses and a simplified version of within season spread dynamics is available at https://github.com/kel seyonofre. Key aspects of the analyses are being incorporated as part of the R2M Plant Health Toolbox (www.garrettlab.com/r2m) for rapid risk assessment to support mitigation of plant pathogens and pests (e.g., Andersen et al., 2019; Andersen Onofre et al., 2021; Buddenhagen et al., 2017; Buddenhagen et al., 2022; Etherton et al., 2023; Etherton et al., 2025; Garrett, 2021; Garrett et al., 2022; Margosian et al., 2009; Ndu wimana et al., 2022; Xing et al., 2020). 2.6. Model feedback and stakeholder participation Initial model assumptions, underlying data, and preliminary model outputs were described to model stakeholders during a session of a workshop held in Vientiane, Lao PDR, September 12, 2019. In addition to CIAT researchers and the study authors, the group included partici pants from local universities, private cassava product companies, and national plant protection programs. The workshop feedback survey is provided in Supplement 2. 3. Results 3.1. Environmental predictor selection and model performance The environmental variables remaining after assessment of multi collinearity were included in downstream analyses (Table 1). Among the models tested, random forest had the highest AUC, TSS and Kappa values and was used for calculating district-wide mean establishment risk (e) estimates (Table 2). The best-fitting random forest model was used to project risk to the remainder of the region. Average district-level establishment risk estimates varied from 0.05 to 0.94 (Fig. 2). K.F. Andersen Onofre et al. Agricultural Systems 229 (2025) 104435 6 https://github.com/kelseyonofre https://github.com/kelseyonofre http://www.garrettlab.com/r2m 3.2. Epidemic spread and model parameter sensitivity Simulations of pathogen spread, with 125 districts starting as infected, were conducted for all pairwise combinations of β and α (Fig. 3). Lower values of β give lower rates of pathogen spread. For example, in the scenario for the lowest α value, 0.1, and lowest β value, 0.000002, after the completion of the first season a mean (across sim ulations) of 419 districts had a non-zero level of CMD incidence (min = 417 districts, max = 431). By season 5, the mean was 489 districts infected (min = 482, max = 512). The total area infected for the same parameter combination was a mean of 7679 ha (min = 7674 ha, max = 7695 ha) with CMD, representing approximately 0.2 % of total cassava area in these four countries. By season 5, the mean number of hectares infected was 10,253 (min = 10,255, max = 10,304). In contrast, for the scenarios with the highest values of β and α (0.002 and 0.9, respec tively), the mean number of districts infected by the close of the first season was 922 (min = 920, max = 975), or 73 % of all districts across the four countries. At the end of this first season, over 2 million hectares of cassava in the region had some level of infection, or 82 % of cassava producing area, and this number jumped to 2.4 million by the end of the 5th season. 3.3. Effect of halting seed trade In an initial scenario analysis to assess the upper limits of seed system management efficacy in the system, we addressed the question of how much a complete trade restriction would limit pathogen spread. In this scenario, the only potential for spread was via spread of viruliferous whiteflies in the landscape. This scenario slowed the epidemic for all parameter combinations, when compared to scenarios of free trade Fig. 2. Estimated risk of cassava mosaic disease (CMD) establishment in SE Asia based on environmental predictors. These district-level estimates are the mean of approximately five estimates for points selected randomly from within a district. Low scores (light colors) represent low CMD risk, and high scores (dark colors) represent high risk. Fig. 3. Cassava mosaic disease (CMD) in the Greater Mekong Subregion in simulations of pathogen spread over 10 seasons for each combination of four values of a dispersal parameter (β) and five values of a parameter (α) modifying the effect of the environment and external inoculum. Points indicate the number of districts with CMD present in 100 realizations. K.F. Andersen Onofre et al. Agricultural Systems 229 (2025) 104435 7 (Fig. 4). Across all the parameter values, the percent reduction in infected hectares ranged from 4 % to 92 % with an average reduction of 46 %, across all parameter combinations and timesteps. 3.4. Clean seed provisioning strategies We tested three scenarios for selecting locations for clean seed deployment, as well as all combinations of managing 20, 40, 60, and 80 districts and 20, 40, 60, and 80 ha. As expected, the least effective method for selecting locations to manage was scenario one, randomly selecting districts within 50 km of infected areas without consideration of infection status (Fig. 5). Although in some realizations this method did perform well, the results were highly variable across realizations. This high variability was likely due in part to new locations being selected in each realization, which may or may not have been districts with previous CMD infection. The best strategy for reducing CMD was scenario two, the "consistent location campaign”, selecting locations at the start of the season with the highest number of infected hectares to manage and managing them each season with the same volume of clean planting material (Fig. 5b). This scenario also has important practical benefits, as it is easier to manage a fixed number of infected areas in a concentrated campaign compared to constantly seeking out new regions to manage. Consistent locations are also more logistically realistic given the stationary nature of most planting material multiplication infrastructure, paired with the rela tively high transport cost of bulky cassava stems. Interestingly, the consistent location campaign was more effective than the “reactive management campaign”, where locations for management were selected at the start of each season based on the number of locations that had the highest number of infected hectares in the previous season. The reactive strategy results in shifting locations to manage, allowing for more hectares of infected area in some cases. For the “consistent location campaign” and the “reactive management campaign” the variation from simulation to simulation was relatively low (Fig. 5 and Fig. 5b). This is in part due to the relatively small infected area and low rate of increase due to the choice of the conservative parameter estimates α = 0.1 and β = 0.000002. An important component of this analysis was identifying the best strategy for prioritizing locations to manage, and total area to manage, given a limited amount of available planting material. For the “consis tent location campaign” managing only 20 locations, even 20 ha led to a decline in CMD prevalence season after season (Fig. 5b), with a more rapid decline when 40, 60, or 80 ha were managed for each of the 20 locations. As the number of districts managed increased, the decline in CMD prevalence was more pronounced. Interestingly, in all cases, managing 40, 60, and 80 ha gave similar results, suggesting that an intermediate volume of clean planting material may approach the maximum benefit from this type of management. 3.4.1. Stakeholder workshop feedback Generally, stakeholders were positive about the utility of the model results (Fig. 6). Among survey respondents, 75–80 % indicated that model results would be useful for planning surveillance and mitigation campaigns. Most agreed that the model results were easy to interpret. Several useful suggestions were left in open-ended comments on the worksheet. This stakeholder feedback was carefully considered in sub sequent rounds of analysis and revision of this work. 4. Discussion Regional deployment of clean seed is an important management strategy for improving the performance of integrated seed systems. Our findings promote evidence-based decision-making to prioritize locations for integrating clean seed deployment in an existing pathosystem. Consistently managing even a small percentage of the 1255 cassava- producing districts in the Greater Mekong subregion, when selected Fig. 4. Cassava mosaic disease in the Greater Mekong Subregion in simulations with seed trade halted so that spread is solely due to whitefly vectors. Path ogen spread is simulated over 10 seasons for each combination of four values of a dispersal parameter (β) and five values of a parameter (α) modifying the effect of the environment and external inoculum. Points indicate the mean number of districts with CMD present in 100 realizations. K.F. Andersen Onofre et al. Agricultural Systems 229 (2025) 104435 8 based on being among those with the highest prevalence of CMD, could reduce disease year after year and result in an overall slowing of spread in the sub-region. Even minor reductions in the rate of spread can pro vide additional time to prepare control measures and conduct research and plant breeding. The consistent management of infected districts (“consistent location scenario”) greatly outperformed management of randomly selected districts near infected areas (Fig. 6). Consistent management of highly infected areas also outperformed the “reactive management” model in which locations changed based on seasonal disease levels. The underperformance of the reactive management regime for deploying clean seed may be due to the insufficient disease eradication resulting from limited resources, allowing target districts to become re- infected in subsequent seasons. Additionally, a small fraction of districts were managed at any given time, with the maximum of 80 representing ~6 % of the total number of districts in the system. Once the epidemic escapes from the initially infected area, management becomes less effective (accounting for the shift from a decrease in total area infected to an increase). It should be noted that districts to manage were selected solely based on the area infected, not accounting for spatial autocorre lation of districts. Selecting districts to manage by considering their spatial proximity (for example, managing adjacent districts or ‘foci’) might improve the method, and would be a good topic for future analysis. Fig. 5. a-c. Simulations of cassava mosaic disease in the Greater Mekong Subregion across time under three scenarios of clean seed provisioning. A) In the “random management scenario”, clean seed is deployed at locations randomly selected each year within a 50 km radius of infected districts. B) In the “consistent location scenario”, clean seed deployment was targeted to the initial area with CMD, and selected locations were managed each season. C) In the “reactive man agement scenario”, locations with the greatest area with CMD were selected independently each year. Subplots represent simulation outcomes for sub-scenarios in which 20, 40, 60 or 80 districts are managed, with the area managed per district indicated by color. K.F. Andersen Onofre et al. Agricultural Systems 229 (2025) 104435 9 Informal seed systems for crops that are vegetatively propagated, like cassava, can move viruses rapidly. In Southeast Asia, cassava stakes commonly move through informal seed systems, without regard for in ternational boundaries, and often over distances >400 km (Delaquis et al., 2018). The model scenarios of complete trade restriction showed a dramatic reduction in pathogen spread when movement was limited to local whitefly dispersal and transmission in the landscape. Trade re strictions could dramatically reduce spread, particularly to regions not adjacent to infected areas. Although informal trade is legally restricted in several Greater Mekong Subregion countries, laws are difficult or impossible to enforce completely for clonally propagated crops, and informal seed systems remain the backbone of the cassava sector. Thus, seed trade policies alone are not sufficient to control CMD in the region. The invasive species distribution model of CMD found climate and land use to be suitable predictors for estimating the potential distribu tion range in the region. In this study we compared the efficacy of ma chine learning techniques that are generally well suited for classification of invasive species distribution data, with the ability to handle a large Fig. 5. (continued). Fig. 6. Survey responses from key regional stakeholders from Vietnam, Cambodia, Lao PDR, Thailand, and China (35 participants total in the workshop, 17 agreed to participate in the survey) about the perceived usefulness of model results to inform decision-making for clean seed deployment to limit pathogen spread. K.F. Andersen Onofre et al. Agricultural Systems 229 (2025) 104435 10 suite of explanatory predictors: random forest (RF) and support vector machines (SVMs). The RF ensemble classification algorithm (Breiman, 2001), is a machine learning method that has also been widely adopted in ecological species distribution modeling due to its high classification accuracy (Bradter et al., 2013; Cutler et al., 2007; Fernández-Delgado et al., 2014; Kampichler et al., 2010). RF is particularly well suited to plant disease incidence datasets which are often non-normal, non-linear, and unbalanced. This method allows for predictors of multiple types (continuous and categorial, for example) and is insensitive to differences in units between predictors (precipitation and solar radiation, for example). Support vector machines (SVMs) also have previously demonstrated utility for mapping invasive pathogens based on envi ronmental and weather predictors, particularly when data about occurrence and absence are limited (Guo et al., 2005). Although a plethora of machine learning classification algorithms have been developed in recent years, across a wide range of disciplines, RF and SVMs performed best in an evaluation of 179 different classifiers when tested across a range of data sets (Fernández-Delgado et al., 2014). We found that RF was best for classifying our data and it was thus used to estimate environmental risk of establishment in our simulation models. As is often the case for pathogens emerging in new regions, data for this pathosystem remain sparse and disconnected. This is the crux of the problem in most emerging epidemics: a trade-off between the amount of data to parameterize models for intervention and the ability to act early enough to mitigate spread. This study integrated disparate sources of data to construct a model of pathogen spread and scenario analysis of clean seed provisioning. Disease occurrence and absence data used to train the invasive species distribution model were compiled from survey sources with different objectives, and often do not represent unbiased systematic observations. Iterative modeling can improve as more data become available, while guiding early decision making, a goal of the R2M Plant Health Toolbox for rapid risk assessment to support mitiga tion of pathogens and pests (www.garrettlab.com/r2m). Iterative data incorporation can be combined with expert and stakeholder participa tory feedback to make the most of available resources. Decision support for seed system management can potentially be improved by considering better economic models. For cash crops, regional farmer behaviors are often driven by return on investment. For example, stem procurement networks are not static and would rewire based on the economics of stem and starch prices. These trends may drive the locations where planting material is needed and sourced, in addition to the number of years a farmer may wait before obtaining off- farm planting material. Decision support can also be improved with greater epidemiological understanding of Sri Lankan cassava mosaic virus. Most current knowledge has been drawn from African cassava mosaic viruses (Holt et al., 1997). High rates of CMD reinfection for clean stems planted under field conditions have been reported in central Cambodia, sug gesting that positive and negative selection strategies by farmers may have significant potential in disease management (Malik et al., 2022). Work is ongoing to understand the biotypes of whitefly in the region and their seasonal abundance and dispersal (Götz and Winter, 2016; Ram Kumar et al., 2016; Chi et al., 2020; Leiva et al., 2022). Research is also needed to detect the introduction or recombination of new cassava mosaic virus strains, which may vary in virulence. Stakeholder feedback during model development can help increase the practical utility of models for specific local decisions. This early feedback can be used to further calibrate models, and allow for data gaps to be identified based on stakeholder opinion. This is particularly important where key data are not systematically published. In this study we implemented a simple stakeholder feedback session. Participatory modeling has been sparsely used in plant disease epidemiology (Gaydos et al., 2019), and more frequently in human epidemiology. There is high value in interactions between modelers and stakeholders, as often the most interesting “modeling questions” do not pragmatically intersect with the more practical “boots on the ground” issues that demand immediate action. Stakeholder participation provides a meeting ground where these two interests can converge, increasing the potential for real- world use of modeling output. The development of decision support approaches could also be expanded to include new data about cassava seed systems across the adjacent producing regions in Myanmar and Southern China, which also maintain considerable cassava production. Countries in the insular re gion of Southeast Asia, including Indonesia and the Philippines, are also at high risk for the arrival of CMD, and could benefit from early pre ventative research on stem exchange patterns and whitefly suitability mapping to identify optimal surveillance and management nodes. The impact of effective seed systems is also multiplied when combined in an integrated seed health strategy with use of resistant varieties and pro grams to support on-farm management (Hareesh et al., 2023; Legg et al., 2022; Thomas-Sharma et al., 2017). CRediT authorship contribution statement Kelsey F. Andersen Onofre: Writing – review & editing, Writing – original draft, Validation, Software, Methodology, Investigation, Formal analysis, Conceptualization. Erik Delaquis: Writing – review & editing, Writing – original draft, Resources, Methodology, Investigation, Data curation, Conceptualization. Jonathan C. Newby: Writing – review & editing, Investigation, Data curation, Conceptualization. Stef de Haan: Writing – review & editing, Investigation, Data curation, Conceptuali zation. Cu Thi Le Thuy: Writing – review & editing, Investigation, Data curation, Conceptualization. Nami Minato: Writing – review & editing, Investigation, Data curation, Conceptualization. James P. Legg: Writing – review & editing, Investigation, Conceptualization. Wilmer J. Cuel lar: Writing – review & editing, Investigation, Conceptualization. Ricardo I. Alcalá Briseño: Writing – review & editing, Investigation, Formal analysis, Conceptualization. Karen A. Garrett: Writing – review & editing, Writing – original draft, Project administration, Methodology, Investigation, Formal analysis, Conceptualization. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgements We appreciate the support of the CGIAR Seed Equal Initiative, CGIAR Research Program on Roots, Tubers and Bananas (RTB), and CGIAR Plant Health Initiative, supported by CGIAR Trust Fund contributors (https://www.cgiar.org/funders/). We also appreciate support by USDA NIFA grant 2020-51181-32198 and USDA Animal and Plant Health In spection Service (APHIS) Cooperative Agreements AP21PPQS&T00C195 and AP22PPQS&T00C133. This work does not necessarily represent the views of the USDA. We appreciate the support of the Australian Centre for International Agricultural Research (ACIAR), project numbers AGB/2016/032 and AGB/2018/172. We appreciate helpful comments from E. Goss, R. Muneepeerakul, A. I. Plex Sulá, J. Robledo, I. Small, and Agricultural Systems reviewers. Appendix A. Supplementary data Supplementary data to this article can be found online at https://doi. org/10.1016/j.agsy.2025.104435. Data availability Data will be made available on request. K.F. Andersen Onofre et al. Agricultural Systems 229 (2025) 104435 11 https://www.garrettlab.com/r2m https://www.cgiar.org/funders/ https://doi.org/10.1016/j.agsy.2025.104435 https://doi.org/10.1016/j.agsy.2025.104435 References Allouche, O., Tsoar, A., Kadmon, R., 2006. Assessing the accuracy of species distribution models: prevalence, kappa and the true skill statistic (TSS). J. Appl. Ecol. 43, 1223–1232. Andersen, K.F., Buddenhagen, C.E., Rachkara, P., Gibson, R., Kalule, S., Phillips, D., Garrett, K.A., 2019. Modeling epidemics in seed systems and landscapes to guide management strategies: the case of sweetpotato in Northern Uganda. Phytopathology 109, 1519–1532. Andersen Onofre, K.F., Forbes, G.A., Andrade-Piedra, J.L., Buddenhagen, C.E., Fulton, J., Gatto, M., Khidesheli, Z., Mdivani, R., Xing, Y., Garrett, K.A., 2021. An integrated seed health strategy and phytosanitary risk assessment: potato in the Republic of Georgia. Agric. Syst. 191, 103144. Bischl, B., Lang, M., Kotthoff, L., Schiffner, J., Richter, J., Studerus, E., Casalicchio, G., Jones, Z.M., 2016. mlr: Machine Learning in R. J. Mach. Learn. Res. 17, 1–5. Bivand, R., Keitt, T., Rowlingson, B., 2019. rgdal: Bindings for The ’Geospatial’ Data Abstraction Library, R Package Version, 1.4–4 ed. Bivand, R., Rundel, C., 2019. rgeos: Interface to Geometry Engine - Open Source (‘GEOS’). R Package Version 0.4–3 ed. Bradter, U., Kunin, W.E., Altringham, J.D., Thom, T.J., Benton, T.G., Peres-Neto, P., 2013. Identifying appropriate spatial scales of predictors in species distribution models with the random forest algorithm. Methods Ecol. Evol. 4, 167–174. Breiman, L., 2001. Random forests. Mach. Learn. 45, 5–32. Buddenhagen, C.E., Hernandez Nopsa, J.F., Andersen, K.F., Andrade-Piedra, J., Forbes, G.A., Kromann, P., Thomas-Sharma, S., Useche, P., Garrett, K.A., 2017. Epidemic network analysis for mitigation of invasive pathogens in seed systems: potato in Ecuador. Phytopathology 107, 1209–1218. Buddenhagen, C.E., Xing, Y., Andrade Piedra, J., Forbes, G.A., Kromann, P., Navarrete, I., Thomas-Sharma, S., Choudhury, R.A., Andersen, K.F., Schulte-Geldermann, E., Etherton, B.A., Plex Sulá, A.I., Garrett, K.A., 2022. Where to invest project efforts for greater benefit: a framework for management performance mapping with examples for potato seed health. Phytopathology 112, 1431–1443. Byrne, D.N., Rathman, R.J., Orum, T.V., Palumbo, J.C., 1996. Localized migration and dispersal by the sweet potato whitefly, Bemisia tabaci. Oecologia 105, 320–328. Cameron, A.R., 2012. The consequences of risk-based surveillance: developing output- based standards for surveillance to demonstrate freedom from disease. Prev. Vet. Med. 105, 280–286. Carvajal-Yepes, M., Cardwell, K., Nelson, A., Garrett, K.A., Giovani, B., Saunders, D.G.O., Kamoun, S., Legg, J.P., Verdier, V., Lessel, J., Neher, R.A., Day, R., Pardey, P., Gullino, M.L., Records, A.R., Bextine, B., Leach, J.E., Staiger, S., Tohme, J., 2019. A global surveillance system for crop diseases. Science 28, 1237–1239. Chadès, I., Martin, T.G., Nicol, S., Burgman, M.A., Possingham, H.P., Buckley, Y.M., 2011. General rules for managing and surveying networks of pests, diseases, and endangered species. Proc. Natl. Acad. Sci. USA 108, 8323–8328. Chi, Y., Pan, L-L., Bouvaine, S., Fan, Y-Y., Liu, Y-Q, Liu, S-S., Seal, S., Wang, X-W., 2020. Differential transmission of Sri Lankan cassava mosaic virus by three cryptic species of the whitefly Bemisia tabaci complex. Virology 540, 141–149. Cuellar, W., Mwanzia, L., Lourido, D., Garcia, C., Martínez, A., Cruz, P., Pino, L., Tohme, J., 2018. PestDisPlace: Monitoring the Distribution of Pests and Diseases, Version 2.0. International Center for Tropical Agriculture (CIAT). http://pestdisp lace.org. Cunniffe, N.J., Cobb, R.C., Meentemeyer, R.K., Rizzo, D.M., Gilligan, C.A., 2016. Modeling when, where, and how to manage a forest epidemic, motivated by sudden oak death in California. Proc. Natl. Acad. Sci. 113, 5640–5645. Cutler, R., Edwards, T.C., Beard, K.H., Cutler, A., Hess, K.T., Gibson, J., Lawler, J.J., 2007. Random forests for classification in ecology. Ecology 88, 2783–2792. Delaquis, E., Andersen, K.F., Minato, N., Thi Le Cu, T., Karssenberg, M., Sok, S., Wyckhuys, K., Newby, J., Burra, D., Srean, P.I.P., Le, N., Thi Pham, N., Garrett, K.A., de Haan, S., 2018. Raising the stakes: cassava seed networks at multiple scales in Cambodia and Vietnam. Front. Sustain. Food Syst. 2, 73. Drake, J.M., Randin, C., Guisan, A., 2006. Modelling ecological niches with support vector machines. J. Appl. Ecol. 43, 424–432. Elith, J., Leathwick, J.R., 2009. Species distribution models: ecological explanation and prediction across space and time. Annu. Rev. Ecol. Evol. Syst. 40, 677–697. Epanchin-Niell, R.S., Haight, R.G., Berec, L., Kean, J.M., Liebhold, A.M., 2012. Optimal surveillance and eradication of invasive species in heterogeneous landscapes. Ecol. Lett. 15, 803–812. Etherton, B.A., Choudhury, R.A., Alcalá-Briseño, R.I., Xing, Y., Plex Sulá, A.I., Carrillo, D., Wasielewski, J., Stelinski, L., Grogan, K.A., Ballen, F., Blare, T., Crane, J., Garrett, K.A., 2023. Are avocados toast? A framework to analyze decision- making for emerging epidemics, applied to laurel wilt. Agric. Syst. 206, 103615. ESA, 2017. Land Cover CCI Product User Guide Version 2. Tech. Rep.. Available at: maps. elie.ucl.ac.be/CCI/viewer/download/ESACCI-LC-Ph2-PUGv2_2.0.pdf. Etherton, B.A., Choudhury, R.A., Alcalá Briseño, R.I., Mouafo-Tchinda, R.A., Plex Sulá, A. I., Choudhury, M., Adhikari, A., Lei, S.L., Kraisitudomsook, N., Robledo Buritica, J., Cerbaro, V.A., Ogero, K., Cox, C.M., Walsh, S.P., Andrade-Piedra, J., Omondi, B.A., Navarrete, I., McEwan, M.A., Garrett, K.A., 2024. Disaster plant pathology: smart solutions for threats to global plant health from natural and human-driven disasters. Phytopathology 114, 855–868. Etherton, B.A., Plex Sulá, A.I., Mouafo-Tchinda, R.A., Kakuhenzire, R., Kassaye, H.A., Asfaw, F., Kosmakos, V.S., McCoy, R.W., Xing, Y., Yao, J., Sharma, K., Garrett, K.A., 2025. Translating Ethiopian potato seed networks: identifying strategic intervention points for managing bacterial wilt and other diseases. Agric. Syst. 222, 104167. Fargette, D., Konaté, G., Fauquet, C., Muller, E., Peterschmitt, M., Thresh, J.M., 2006. Molecular ecology and emergence of tropical plant viruses. Annu. Rev. Phytopathol. 44, 235–260. Fernández-Delgado, M., Cernadas, E., Barro, S., 2014. Do we need hundreds of classifiers to solve real world classification problems? J. Mach. Learn. Res. 15, 31333181. Fick, S.E., Hijmans, R.J., 2017. WorldClim 2: new 1-km spatial resolution climate surfaces for global land areas. Int. J. Climatol. 37, 4302–4315. Gallien, L., Douzet, R., Pratte, S., Zimmermann, N.E., Thuiller, W., 2012. Invasive species distribution models - how violating the equilibrium assumption can create new insights. Glob. Ecol. Biogeogr. 21, 1126–1136. Garrett, K.A., 2021. Impact network analysis and the INA R package: decision support for regional management interventions. Methods Ecol. Evol. 12, 1634–1647. Garrett, K.A., Alcalá-Briseño, R.I., Andersen, K.F., Buddenhagen, C.E., Choudhury, R.A., Fulton, J.C., Hernandez Nopsa, J.F., Poudel, R., Xing, Y., 2018. Network analysis: a systems framework to address grand challenges in plant pathology. Annu. Rev. Phytopathol. 56, 559–580. Garrett, K.A., Bebber, D.P., Etherton, B.A., Gold, K.M., Plex Sulá, A.I., Selvaraj, M.G., 2022. Climate change effects on pathogen emergence: artificial intelligence to translate big data for mitigation. Annu. Rev. Phytopathol. 60, 357–378. Gaydos, D.A., Petrasova, A., Cobb, R.C., Meentemeyer, R.K., 2019. Forecasting and control of emerging infectious forest disease through participatory modelling. Philos. Trans. R. Soc. Lond. Ser. B Biol. Sci. 374, 20180283. Götz, M., Winter, S., 2016. Diversity of Bemisia tabaci in Thailand and Vietnam and indications of species replacement. J. Asia-Pac. Entomol. 19, 537–543. Graziosi, I., Minato, N., Alvarez, E., Ngo, D.T., Hoat, T.X., Aye, T.M., Pardo, J.M., Wongtiem, P., Wyckhuys, K.A., 2016. Emerging pests and diseases of south-east Asian cassava: a comprehensive evaluation of geographic priorities, management options and research needs. Pest Manag. Sci. 72, 1071–1089. Guo, Q., Kelly, M., Graham, C.H., 2005. Support vector machines for predicting distribution of sudden oak death in California. Ecol. Model. 182, 75–90. Hareesh, P.S., Resmi, T.R., Sheela, M.N., Makeshkumar, T., 2023. Cassava mosaic disease in south and Southeast Asia: current status and prospects. Front. Sustain. Food Syst. 7, 1086660. Harteveld, D.O.C., Grant, M.R., Pscheidt, J.W., Peever, T.L., 2017. Predicting ascospore release of Monilinia vaccinii-corymbosi of blueberry with machine learning. Phytopathology 107, 1364–1371. Harwood, T.D., Xu, X., Pautasso, M., Jeger, M.J., Shaw, M.W., 2009. Epidemiological risk assessment using linked network and grid based modelling: Phytophthora ramorum and Phytophthora kernoviae in the UK. Ecol. Model. 220, 3353–3361. Hijmans, R.J., 2019. raster: Geographic Data Analysis and Modeling, R Package Version 2.9–5 ed. Holt, J., Jeger, M.J., Thresh, J.M., 1997. An epidemilogical model incorporating vector population dynamics applied to African cassava mosaic virus disease. J. Appl. Ecol. 34, 793–806. Jarnevich, C.S., Stohlgren, T.J., Kumar, S., Morisette, J.T., Holcombe, T.R., 2015. Caveats for correlative species distribution modeling. Eco. Inform. 29, 6–15. Jiménez-Valverde, A., Peterson, A.T., Soberón, J., Overton, J.M., Aragón, P., Lobo, J.M., 2011. Use of niche models in invasive species risk assessments. Biol. Invasions 13, 27852797. Kampichler, C., Wieland, R., Calmé, S., Weissenberger, H., Arriaga-Weiss, S., 2010. Classification in conservation biology: a comparison of five machine-learning methods. Eco. Inform. 5, 441–450. Legg, J.P., 1999. Emergence, spread and strategies for controlling the pandemic of cassava modsaic virus disease in east and Central Africa. Crop Prot. 18, 627–637. Legg, J.P., Fauquet, C., 2004. Cassava mosaic geminiviruses in Africa. Plant Mol. Biol. 56, 585–599. Legg, J.P., Diebiru-Ojo, E., Eagle, D., Friedmann, M., Kanju, E., Kapinga, R., Kumar, P.L., Lateef, S., Magige, S., Mtunda, K., 2022. Commercially sustainable cassava seed systems in Africa. In: Root, Tuber and Banana Food System Innovations: Value Creation for Inclusive Outcomes. Springer International Publishing Cham, pp. 453–482. Leiva, A.M., Chittarath, K., Lopez-Alvarez, D., Vongphachanh, P., Gomez, M.I., Sengsay, S., Wang, X.-W., Rodriguez, R., Newby, J., Cuellar, W.J., 2022. Mitochondrial genetic diversity of Bemisia tabaci (Gennadius) (Hemiptera: Aleyrodidae) associated with cassava in Lao PDR. Insects 13, 861. Leiva, A.M., Siriwan, W., Lopez-Alvarez, D., Barrantes, I., Hemniam, N., Saokham, K., Cuellar, W.J., 2020. Nanopore-based complete genome sequence of a Sri Lankan cassava mosaic virus (Geminivirus) strain from Thailand. Microbiol. Resour. Announc 9, e01274–19. Lorena, A.C., Jacintho, L.F.O., Siqueira, M.F., Giovanni, R.D., Lohmann, L.G., de Carvalho, A.C.P.L.F., Yamamoto, M., 2011. Comparing machine learning classifiers in potential distribution modelling. Expert Syst. Appl. 38, 5268–5275. Mainali, K.P., Warren, D.L., Dhileepan, K., McConnachie, A., Strathie, L., Hassan, G., Karki, D., Shrestha, B.B., Parmesan, C., 2015. Projecting future expansion of invasive species: comparing and improving methodologies for species distribution modeling. Glob. Chang. Biol. 21, 4464–4480. Malik, A.I., Sophearith, S., Delaquis, E., Cuellar, W.J., Jimenez, J., Newby, J.C., 2022. Susceptibility of cassava varieties to disease caused by Sri Lankan cassava mosaic virus and impacts on yield by use of asymptomatic and virus-free planting material. Agronomy 12, 1658. Margosian, M.L., Garrett, K.A., Hutchinson, J.M.S., With, K.A., 2009. Connectivity of the American agricultural landscape: assessing the national risk of crop pest and disease spread. BioScience 59, 141–151. Martinetti, D., Soubeyrand, S., 2019. Identifying lookouts for epidemio-surveillance: application to the emergence of Xylella fastidiosa in France. Phytopathology 109, 265–276. McQuaid, C.F., Gilligan, C.A., van den Bosch, F., 2017b. Considering behaviour to ensure the success of a disease control strategy. R. Soc. Open Sci. 4, 170721. K.F. Andersen Onofre et al. Agricultural Systems 229 (2025) 104435 12 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0005 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0005 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0005 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0015 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0015 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0015 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0015 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0010 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0010 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0010 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0010 http://refhub.elsevier.com/S0308-521X(25)00175-1/optUWABBpolM3 http://refhub.elsevier.com/S0308-521X(25)00175-1/optUWABBpolM3 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0020 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0020 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0025 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0025 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0025 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0030 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0035 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0035 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0035 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0035 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0040 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0040 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0040 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0040 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0040 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0045 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0045 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0050 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0050 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0050 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0055 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0055 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0055 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0055 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0065 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0065 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0065 http://refhub.elsevier.com/S0308-521X(25)00175-1/opt12804ICJvn http://refhub.elsevier.com/S0308-521X(25)00175-1/opt12804ICJvn http://refhub.elsevier.com/S0308-521X(25)00175-1/opt12804ICJvn http://pestdisplace.org http://pestdisplace.org http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0075 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0075 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0075 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0080 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0080 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0085 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0085 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0085 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0085 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0090 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0090 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0095 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0095 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0100 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0100 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0100 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0105 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0105 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0105 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0105 http://maps.elie.ucl.ac.be/CCI/viewer/download/ESACCI-LC-Ph2-PUGv2_2.0.pdf http://maps.elie.ucl.ac.be/CCI/viewer/download/ESACCI-LC-Ph2-PUGv2_2.0.pdf http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0110 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0110 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0110 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0110 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0110 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0110 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0115 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0115 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0115 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0115 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0120 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0120 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0120 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0130 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0130 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0135 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0135 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0145 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0145 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0145 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0150 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0150 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0155 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0155 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0155 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0155 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0160 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0160 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0160 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0165 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0165 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0165 http://refhub.elsevier.com/S0308-521X(25)00175-1/optDUGFEIMJHt http://refhub.elsevier.com/S0308-521X(25)00175-1/optDUGFEIMJHt http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0170 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0170 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0170 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0170 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0175 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0175 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0180 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0180 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0180 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0185 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0185 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0185 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0190 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0190 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0190 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0195 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0195 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0200 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0200 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0200 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0205 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0205 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0210 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0210 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0210 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0215 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0215 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0215 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0220 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0220 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0225 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0225 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0235 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0235 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0235 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0235 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0235 http://refhub.elsevier.com/S0308-521X(25)00175-1/optZn3cHrLcoM http://refhub.elsevier.com/S0308-521X(25)00175-1/optZn3cHrLcoM http://refhub.elsevier.com/S0308-521X(25)00175-1/optZn3cHrLcoM http://refhub.elsevier.com/S0308-521X(25)00175-1/optZn3cHrLcoM http://refhub.elsevier.com/S0308-521X(25)00175-1/optzfz19rloeT http://refhub.elsevier.com/S0308-521X(25)00175-1/optzfz19rloeT http://refhub.elsevier.com/S0308-521X(25)00175-1/optzfz19rloeT http://refhub.elsevier.com/S0308-521X(25)00175-1/optzfz19rloeT http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0240 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0240 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0240 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0245 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0245 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0245 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0245 http://refhub.elsevier.com/S0308-521X(25)00175-1/optI4HxJsu5Ww http://refhub.elsevier.com/S0308-521X(25)00175-1/optI4HxJsu5Ww http://refhub.elsevier.com/S0308-521X(25)00175-1/optI4HxJsu5Ww http://refhub.elsevier.com/S0308-521X(25)00175-1/optI4HxJsu5Ww http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0250 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0250 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0250 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0255 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0255 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0255 http://refhub.elsevier.com/S0308-521X(25)00175-1/optvExJNDUWM3 http://refhub.elsevier.com/S0308-521X(25)00175-1/optvExJNDUWM3 McQuaid, C.F., van den Bosch, F., Szyniszewska, A., Alicai, T., Pariyo, A., Chikoti, P.C., Gilligan, C.A., 2017a. Spatial dynamics and control of a crop pathogen with mixed- mode transmission. PLoS Comput Biol 13, e1005654. Meentemeyer, R.K., Anacker, B.L., Mark, W., Rizzo, D.M., 2008. Early detection of emerging forest disease using dispersal estimation and ecological niche modeling. Ecol. Appl. 18, 377390. Minato, N., Sok, S., Chen, S., Delaquis, E., Phirun, I., Le, V.X., Burra, D.D., Newby, J.C., Wyckhuys, K.A.G., de Haan, S., 2019. Surveillance for Sri Lankan cassava mosaic virus (SLCMV) in Cambodia and Vietnam one year after its initial detection in a single plantation in 2015. PLoS ONE 14, e0212780. Moslonka-Lefebvre, M., Finley, A., Dorigatti, I., Dehnen-Schmutz, K., Harwood, T., Jeger, M.J., Xu, X.M., Holdenrieder, O., Pautasso, M., 2011. Networks in plant epidemiology: from genes to landscapes, countries, and continents. Phytopathology 101, 392–403. Nagoshi, R.N., Htain, N.N., Boughton, D., Zhang, L., Xiao, Y., Nagoshi, B.Y., Mota- Sanchez, D., 2020. Southeastern Asia fall armyworms are closely related to populations in Africa and India, consistent with common origin and recent migration. Sci. Rep. 10 (1), 1421. Naimi, B., Hamm, N.A.S., Groen, T.A., Skidmore, A.K., Toxopeus, A.G., 2014. Where is positional uncertainty a problem for species distribution modelling. Ecography 37, 191–203. Narouei-Khandan, H.A., Halbert, S.E., Worner, S.P., van Bruggen, A.H.C., 2016. Global climate suitability of citrus huanglongbing and its vector, the Asian citrus psyllid, using two correlative species distribution modeling approaches, with emphasis on the USA. Eur. J. Plant Pathol. 144, 655–670. Nduwimana, I., Sylla, S., Xing, Y., Simbare, A., Niyongere, C., Garrett, K.A., Omondi, B. A., 2022. Banana seed exchange networks in Burundi–Linking formal and informal systems. Outlook Agric. 51, 334–348. Oerke, E.-C., 2006. Crop losses to pests. J. Agric. Sci. 144, 31–43. Olden, J.D., 2008. Machine learning methods without tears: a primer for ecologists. Q. Rev. Biol. 83, 171–193. Parnell, S., Gottwald, T.R., Riley, T., van den Bosch, F., 2014. A generic risk-based surveying method for invading plant pathogens. Ecol. Appl. 24, 779–790. Pautasso, M., 2015. Network simulations to study seed exchange for agrobiodiversity conservation. Agron. Sustain. Dev. 35, 145–150. Pautasso, M., Jeger, M.J., 2008. Epidemic threshold and network structure: the interplay of probability of transmission and of persistence in small-size directed networks. Ecol. Complex. 5, 1–8. Pebesma, E., 2018. Simple features for R: standardized support for spatial vector data. R J. 10, 439–446. Ram Kumar, N., Chang, J.C., Narayanan, M.B., Ramasamy, S., 2016. Phylogeographical structure in mitochondrial DNA of whitefly, Bemisia tabaci Gennadius (Hemiptera: Aleyrodidae) in southern India and Southeast Asia. Mitochondrial DNA Part. A 28, 621–631. Reuter, H.I., Nelson, A., Jarvis, A., 2007. An evaluation of void-filling interpolation methods for SRTM data. Int. J. Geogr. Inf. Sci. 21, 983–1008. Rey, C., Vanderschuren, H., 2017. Cassava mosaic and brown streak diseases: current perspectives and beyond. Annu. Rev. Phytopathol. 4, 429–452. Ristaino, J.B., Anderson, P.K., Bebber, D.P., Brauman, K.A., Cunniffe, N.J., Fedoroff, N. V., Finegold, C., Garrett, K.A., Gilligan, C.A., Jones, C.M., Martin, M.D., MacDonald, G.K., Neenan, P., Records, A., Schmale, D.G., Tateosian, L., Wei, Q., 2021. The persistent threat of emerging plant disease pandemics to global food security. Proc. Natl. Acad. Sci. 118, e2022239118. Rojas, M.R., Macedo, M.A., Maliano, M.R., Soto-Aguilar, M., Souza, J.O., Briddon, R.W., Kenyon, L., Bustamante, R.F.R., Zerbini, M., Adkins, S., Legg, J.P., Kvarnheden, A., Wintermantel, W.M., Sudarshana, M.R., Peterschmitt, M., Lapidot, M., Martin, D.P., Moriones, E., Inoue-Nagata, A.K., Gilbertson, R.L., 2018. World management of geminiviruses. Annu. Rev. Phytopathol. 56, 637–677. Sanatkar, M.R., Scoglio, C., Natarajan, B., Isard, S., Garrett, K.A., 2015. History, epidemic evolution, and model burn-in for a network of annual invasion: soybean rust. Phytopathology 105, 947–955. Saunders, K., Salim, N., Mali, V.R., Malathi, V.G., Briddon, R., Markham, P.G., Stanley, J., 2002. Characterisation of Sri Lankan cassava mosaic virus and Indian cassava mosaic virus: evidence for acquisition of a DNA B component by a monopartite begomovirus. Virology 293, 63–74. Shah, D.A., De Wolf, E.D., Paul, P.A., Madden, L.V., 2014. Predicting Fusarium head blight epidemics with boosted regression trees. Phytopathology 104, 702–714. Shaw, M.W., Pautasso, M., 2014. Networks and plant disease management: concepts and applications. Annu. Rev. Phytopathol. 52, 477–493. Siriwan, W., Jimenez, J., Hemniam, N., Saokham, K., Lopez-Alvarez, D., Leiva, A.M., Martinez, A., Mwanzia, L., Becerra Lopez-Lavalle, Cuellar, W.J., 2020. Surveillance and diagnostics of the emergent Sri Lankan cassava mosaic virus (Fam. Geminiviridae) in Southeast Asia. Virus Res. 285, 197959. Stark, K.D., Regula, G., Hernandez, J., Knopf, L., Fuchs, K., Morris, R.S., Davies, P., 2006. Concepts for risk-based surveillance in the field of veterinary medicine and veterinary public health: review of current approaches. BMC Health Serv. Res. 6, 20. Sutrave, S., Scoglio, C., Isard, S.A., Hutchinson, J.M., Garrett, K.A., 2012. Identifying highly connected counties compensates for resource limitations when evaluating national spread of an invasive pathogen. PLoS ONE 7, e37793. Thomas-Sharma, S., Andrade-Piedra, J., Carvajal Yepes, M., Hernandez Nopsa, J., Jeger, M., Jones, R., Kromann, P., Legg, J., Yuen, J., Forbes, G., Garrett, K.A., 2017. A risk assessment framework for seed degeneration: informing an integrated seed health strategy for vegetatively-propagated crops. Phytopathology 107, 1123–1135. Thresh, J.M., Cooter, R.J., 2005. Strategies for controlling cassava mosaic virus disease in Africa. Plant Pathol. 54, 587–614. Uke, A., Hoat, T.X., Quan, M.V., Leim, N.V., Ugaki, M., Natsuaki, K.T., 2018. First report of Sri Lankan cassava mosaic virus infecting cassava in Vietnam. Plant Dis. 102, 2669. Wang, C., Wang, X., Jin, Z., Müller, C., Pugh, T.A., Chen, A., Wang, T., Huang, L., Zhang, Y., Li, L.X., Piao, S., 2022. Occurrence of crop pests and diseases has largely increased in China since 1970. Nature Food 3 (1), 57–65. Wang, D., Yao, X., Huang, G., Shi, T., Wang, G., Ye, J., 2019. First report of Sri Lankan cassava mosaic virus infected cassava in China. Plant Dis. 103, 10–1094. Wang, D., Huang, G., Shi, T., Wang, G., Fang, R., Zhang, X., Ye, J., 2020. Surveillance and distribution of the emergent Sri Lankan cassava mosaic virus in China. Phytopathol. Res. 2, 18. Wang, H., Cui, X., Wang, X., Liu, S., Zhou, X., 2016. First Report of Sri Lankan cassava mosaic virus infecting cassava in Cambodia. Plant Dis. 100, 1029. Xing, Y., Hernandez Nopsa, J.F., Andersen, K.F., Andrade-Piedra, J.L., Beed, F.D., Blomme, G., Carvajal Yepes, M., Coyne, D.L., Cuellar, W.J., Forbes, G.A., Kreuze, J. F., Kroschel, J., Kumar, P.L., Legg, J.P., Parker, M., Schulte-Geldermann, E., Sharma, K., Garrett, K.A., 2020. Global cropland connectivity: a risk factor for invasion and saturation by emerging pathogens and pests. BioScience 70, 744–758. Zheng, S.J., García-Bastidas, F.A., Li, X., Zeng, L., Bai, T., Xu, S., Yin, K., Li, H., Fu, G., Yu, Y, Yang, L, Nguyen, HC, Douangboupha, B, Khaing, AA, Drenth, A, Seidl, M.F., Meijer, H.J.G., Kema, G.H.J., 2018. New geographical insights of the latest expansion of Fusarium oxysporum f.sp. cubense Tropical Race 4 Into the Greater Mekong Subregion. Front. Plant Sci. 9, 457. K.F. Andersen Onofre et al. Agricultural Systems 229 (2025) 104435 13 http://refhub.elsevier.com/S0308-521X(25)00175-1/optXoQlOfImE5 http://refhub.elsevier.com/S0308-521X(25)00175-1/optXoQlOfImE5 http://refhub.elsevier.com/S0308-521X(25)00175-1/optXoQlOfImE5 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0260 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0260 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0260 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0265 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0265 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0265 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0265 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0270 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0270 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0270 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0270 http://refhub.elsevier.com/S0308-521X(25)00175-1/optcbic5xGz3Z http://refhub.elsevier.com/S0308-521X(25)00175-1/optcbic5xGz3Z http://refhub.elsevier.com/S0308-521X(25)00175-1/optcbic5xGz3Z http://refhub.elsevier.com/S0308-521X(25)00175-1/optcbic5xGz3Z http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0275 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0275 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0275 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0280 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0280 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0280 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0280 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0285 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0285 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0285 http://refhub.elsevier.com/S0308-521X(25)00175-1/optc5ckYFwVwP http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0290 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0290 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0295 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0295 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0300 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0300 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0305 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0305 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0305 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0310 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0310 http://refhub.elsevier.com/S0308-521X(25)00175-1/optFcIv1JHtiC http://refhub.elsevier.com/S0308-521X(25)00175-1/optFcIv1JHtiC http://refhub.elsevier.com/S0308-521X(25)00175-1/optFcIv1JHtiC http://refhub.elsevier.com/S0308-521X(25)00175-1/optFcIv1JHtiC http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0320 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0320 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0325 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0325 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0330 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0330 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0330 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0330 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0330 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0335 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0335 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0335 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0335 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0335 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0340 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0340 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0340 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0345 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0345 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0345 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0345 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0350 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0350 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0355 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0355 http://refhub.elsevier.com/S0308-521X(25)00175-1/optywClE34zXw http://refhub.elsevier.com/S0308-521X(25)00175-1/optywClE34zXw http://refhub.elsevier.com/S0308-521X(25)00175-1/optywClE34zXw http://refhub.elsevier.com/S0308-521X(25)00175-1/optywClE34zXw http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0360 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0360 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0360 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0365 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0365 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0365 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0370 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0370 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0370 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0370 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0375 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0375 http://refhub.elsevier.com/S0308-521X(25)00175-1/opt1xUtkURJhh http://refhub.elsevier.com/S0308-521X(25)00175-1/opt1xUtkURJhh http://refhub.elsevier.com/S0308-521X(25)00175-1/opt1xUtkURJhh http://refhub.elsevier.com/S0308-521X(25)00175-1/opthGDMwxdV86 http://refhub.elsevier.com/S0308-521X(25)00175-1/opthGDMwxdV86 http://refhub.elsevier.com/S0308-521X(25)00175-1/opthGDMwxdV86 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0380 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0380 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0385 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0385 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0385 http://refhub.elsevier.com/S0308-521X(25)00175-1/optl5x2fV8cXH http://refhub.elsevier.com/S0308-521X(25)00175-1/optl5x2fV8cXH http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0395 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0395 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0395 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0395 http://refhub.elsevier.com/S0308-521X(25)00175-1/rf0395 http://refhub.elsevier.com/S0308-521X(25)00175-1/optbOfYSAdglw http://refhub.elsevier.com/S0308-521X(25)00175-1/optbOfYSAdglw http://refhub.elsevier.com/S0308-521X(25)00175-1/optbOfYSAdglw http://refhub.elsevier.com/S0308-521X(25)00175-1/optbOfYSAdglw http://refhub.elsevier.com/S0308-521X(25)00175-1/optbOfYSAdglw Decision support for managing an invasive pathogen through efficient clean seed systems: Cassava mosaic disease in Southeas ... 1 Introduction 2 Methods 2.1 Invasive species distribution model of CMD establishment 2.1.1 Geographic distribution of disease (response variables) 2.1.2 Predictor selection and dimension reduction 2.1.3 Classification models 2.1.4 Machine learning 2.1.5 CMD risk 2.2 Network model of pathogen spread 2.3 Model components 2.3.1 Seed system network 2.3.2 Short-distance whitefly dispersal 2.4 Simulation model 2.4.1 Uncertainty 2.5 Simulation experiments 2.5.1 Experiment one: seed trade restriction 2.5.2 Experiment two: clean seed intervention strategies 2.6 Model feedback and stakeholder participation 3 Results 3.1 Environmental predictor selection and model performance 3.2 Epidemic spread and model parameter sensitivity 3.3 Effect of halting seed trade 3.4 Clean seed provisioning strategies 3.4.1 Stakeholder workshop feedback 4 Discussion CRediT authorship contribution statement Declaration of competing interest Acknowledgements Appendix A Supplementary data Data availability References