Abstract
Introduction:
Banana bunchy top disease (BBTD) is a major emerging viral disease of banana plants in Africa, threatening food security and economies. BBTD has spread to numerous (sub)tropical African countries, with new incursions in multiple countries confirmed even within the last few years.
Methods:
We assessed the continent-wide risk of BBTD, to provide an informative tool of large-scale risk areas of BBTD introduction, establishment, and spread. Using published BBTD survey data across (sub)tropical Africa, a RandomForest model was developed incorporating both climatological suitability and socioeconomic drivers.
Results:
Population density, which serves as a proxy for factors like road infrastructure and trade, dominates our model, along with maximum solar radiation and annual temperature range. Three main regions are identified as current high-risk areas, namely, East-Central Africa (around the African Great Lakes in Uganda, Rwanda, Burundi, eastern DR Congo, and northern Tanzania), western Central Africa (including Gabon and the Republic of the Congo), and southern regions of West Africa. Our projections for ~2055, driven by climate change and human population shifts, predict increased risks in East-Central Africa and along the 5-10°N latitude belt, while the risk in Central Africa is expected to decrease.
Discussion:
We present these African risk maps as a complementary tool to national assessments, offering a broader spatial context to inform effective disease management and policy development. Specifically, we identify the critical need to address the unregulated banana seed system, where stakeholders are free to use and informally trade planting material across banana production regions and borders, to achieve effective prevention and disease management.
Introduction
Banana bunchy top disease (BBTD) is a severe viral disease of banana (Musa spp.) plants caused by the banana bunchy top virus (BBTV; Babuvirus musae, family Nanoviridae; ). Infected plants develop characteristic dark-green streaks on leaf sheaths, petioles, and midribs, with new leaves becoming progressively shorter, narrower, and more fragile, ultimately forming a distinctive bunchy cluster at the top of the plant (). Early infection prevents fruiting, while later infections yield distorted, stunted, and unmarketable fruit (), leading to substantial yield losses. BBTV primarily spreads through infected planting material, with secondary transmission occurring from plant to plant via the banana aphid (Pentalonia nigronervosa) (). Control generally consists of early detection and roguing, i.e., the full removal of infected mats ().
Currently, BBTD is widespread in Southeast Asia, the South Pacific, and Africa (, ). In Africa, BBTD has been identified as one of the main emerging infectious diseases threatening food security and economies (; ). African banana production relies heavily on smallholder farmers, and banana crop diseases such as BBTD threaten household food security and livelihood, further impacting national value chains. BBTV has spread to numerous African countries (, ; ; Supplementary Figure S1), with the most recent confirmed reports in Togo (), Uganda (, ), and Tanzania (; ). Continuous national-level monitoring of disease spread is critical for establishing and implementing effective prevented spread or disease management strategies (e.g., Ximba et al., 2022; ), yet survey efforts are frequently constrained by a combination of competing governmental priorities, insufficient financial allocations, and inadequate human resources. More still, knowledge of hotspots or areas-at risk to support continuous monitoring beyond initial assessments is often lacking.
Modeling efforts to identify areas at risk for banana crop disease offer a useful tool, bridging the gap between geographically constrained/limited survey data and informed predictions of risk. In recent years, several country-/national-level BBTD risk assessments have been developed for Uganda (), Tanzania (), and Rwanda (), in addition to a continent-wide assessment (). These assessments utilized various modeling approaches, including statistical regressions, process-based modeling, and machine learning. The approaches applied relied solely on environmental/climatological suitability for banana production, BBTD establishment, and vector habitat (e.g., ), or a combination of environmental suitability with expert knowledge. This expert knowledge generally includes factors influencing disease spread, such as distance to currently infected areas, production landscape connectivity, and dominant banana cultivars (e.g., ; ). Specifically, expert opinion on the risk of incursions for user-delineated banana production landscapes (i.e., arbitrary zonal borders, often aligning with administrative boundaries) are integrated post-hoc into the overall BBTD risk assessments. While such expert knowledge is highly useful in specific, single-effort studies, the results are not reproducible as is, transferable to other regions, or reliably applicable in changing environments.
Additionally, the spatial scale at which risk assessments are used should be carefully assessed, incorporating realistic pathways and ranges of disease transmission within the areas of interest. These are generally confined to national levels, aligning with the scale at which policy is implemented. In the context of African smallholder banana production systems, BBTD disease transmission is complicated by the highly informal and unregulated banana seed system, i.e., the informal use and trade of vegetatively propagated planting materials (; ). Therefore, national BBTD risk assessments should account for potential cross-border movement and disease incursion from neighboring regions or other regions with regular trade. Combining national risk assessments with evaluations at a larger scale (; ) can provide additional spatial context and broader patterns needed for effective disease management and policy development.
provided a first African continent-wide BBTD risk assessment, with an environmental suitability map developed using a compilation of BBTD field surveys, and a vulnerability map based on expert knowledge. However, while several of the survey datasets used had recorded both presence and absence of BBTD, the data were heavily skewed toward presence, which is a common issue in ecological modeling (). For example, the field survey by focused solely on disease presence targeting villages in territories with the highest observed incidence and severity levels, reflecting the worst-case scenario within a given location. When not accounting for the absence of the disease in healthy fields and regions, and simultaneously addressing the clustering of observations (spatial autocorrelation) in densely surveyed regions (for which solutions are available, e.g., ; ), the risk in such regions can be highly overestimated. Furthermore, expert knowledge on distance to infected areas and geographical connectivity, among other variables, was used to assess vulnerability to BBTD. Such expert evaluation is highly biased toward well-documented existing hotspots, potentially overlooking areas with unknown infections. Compiling comprehensive expert knowledge for an entire continent also proves difficult, resulting in information gaps and irreproducible assessments.
In this study, we reevaluate the continent-wide risk of BBTD in Africa, focusing on a reproducible risk assessment and evolving risks driven by climate change and human population shifts. A RandomForest model was developed incorporating both climatological suitability and socioeconomic drivers related to the spread, introduction, and establishment of BBTD. Here, population density is used as a proxy, applicable at the continental scale, to represent regions with a good road infrastructure and market access, experiencing intensified trade and increased food security pressures. Finally, we evaluate how overall BBTD risk changes by mid-century (~2055) in relation to predicted climate and population shifts under two shared socioeconomic pathway scenarios.
Materials and methods
Data informed spatial analysis
Survey data and covariate selection
Field survey datasets as compiled by were extended with recent surveys from Uganda (), Rwanda, and neighboring regions in eastern Democratic Republic of Congo (DR Congo) (). We note that the compiled survey datasets were produced through various research studies using varying methodologies, often prioritizing the reporting of BBTD presence over its absence.
Selected co-variates for model development included climatological, land-use/land-cover, and socioeconomic variables, which can influence BBTD introduction, establishment, and spread (; ; ). No distinction is therefore made between the risk of these components (introduction, establishment, and spread), but an overall BBTD risk assessment is presented. Standardized bioclimatic variables based on temperature and precipitation (19 in total) from the CHELSA-BIOCLIM+ dataset (), averaged for the years 1981-2010, were included to represent climatological annual trends, seasonality, and extreme or limiting environmental factors. Annual means, range, and limiting months (months with highest and lowest values) of wind speed and solar radiation from 1981 to 2010 were identified as potential factors influencing aphid habitat and spread and were thus added (). Human population density, estimated as a count per grid cell (people/km2) (WorldPop et al., 2018 for year 2020), was included as a proxy for road infrastructure and market access, intensified trade, and food security pressures, all factors affecting the exchange of potentially infected planting material. CROPGRIDS crop area data for banana crop cover (; at a 0.05° grid resolution) was selected to represent banana production areas. The Enhanced Vegetation Index (EVI), a metric for vegetation greenness, was included through EVI MODIS data (; Monthly EVI MOD13A3 v061 at a 1-km grid resolution) as the mean, minima, and maxima for 2014-2023. We did not correct for land-cover changes, due to missing data in future projections, and assumed it balanced through population density.
All data were resampled to a common 30-arcsec grid resolution (~1 km2 near the equator) using bilinear resampling, when necessary. The AppEEARS (version 1.1) R-package was used for EVI data retrieval ().
Feature selection and preprocessing
We applied two feature selection methods in order to reduce cross-correlations among the full list of features. First, we applied the Boruta algorithm which implements a random forest-based feature selection based on variable importance (). Second, we used a simple (Pearson’s) correlation analysis. Features not retained by the Boruta method, or with correlation values exceeding a (rho) 0.9 threshold, were removed from the dataset. Clustered data points can bias model training due to oversampling. Presence and absence data were therefore thinned to the spatial resolution of the driver data. Thinning reduces the data points to 1,384 from the original 2,363 locations. We acknowledge that sampling for BBTD is uneven across presence and absences across the large area considered. Most notably, there is a tendency to monitor for and only report presences of the disease. To address this imbalance, we used pseudo-absences to provide feature space representation outside the densely sampled, most likely, locations. Random samples were generated across the whole spatial domain, but at least 5 km removed from known BBTD presence locations (but constrained by neighboring points to avoid oversampling). These final data were split using an 80%/20% ratio for training and testing data, respectively. During model training, we used a 20-fold spatial block cross-validation on the training data only (withholding the testing data for final model evaluation), using a spatial block design with a 5-km buffer to address spatial-autocorrelation.
Model development
We used a decision tree-based Random Forest approach () to optimally use the data space and validate expert knowledge on a larger spatial scale through model inference, and informal prediction. We tuned the random forest hyperparameters using a regular grid search with the following settings: number of predictors sampled at each split (mtry: 1-4), the minimal node size (min_n: 4-100), and number of trees (trees: 1-500), and this using the 20-fold spatial block cross validation split on the training data. Best hyperparameters were retained for final model training. In all instances, we relied on the receiver operating characteristic area under the curve (AUROC) score in model optimization. To assess the model fit, we report variable importance for the final model using the Gini impurity index. The testing data (remaining 20%) were used in the calculation of a confusion matrix and derived accuracy metrics (e.g., Accuracy, Precision, F1, Recall). To assess the model accuracy across feature space, we calculated the area of applicability (AOA) as a way to assess where the model is applicable in the feature and geospatial domain (; ). Areas outside of the AOA are due to the survey data used to train the model not being representative of the specific conditions, meaning they fall outside the range of the model’s feature space. The AOA allows us to assess the geospatial applicability of the model in contemporary and future climate scenarios (see below).
Our model was developed in the R programming language (R version 4.4.0; ) relying on the R packages tidymodels (version 1.3.0), spatialsample (version 0.06), waywiser (version 0.6.3), Boruta (version 8.0.0), and ranger (version 0.17.0) (; Wright and Ziegler, 2017; ; ; ).
BBTD predictions toward ~2055
BBTD probability for mid to late 21st century (2041-2070) was predicted using the selected best-fit model. Five Coupled Model Intercomparison Project Phase 6 (CMIP6) climate projections were selected for a variety of models (GFDL-ESM4, IPSL-CM6A-LR, MPI-ESM1-2-HR, MRI-ESM2-0, UK-ESM1-0-LL) based on two Shared Socioeconomic Pathway (SSP) scenarios, namely, the middle of the road SSP3-7.0, and more optimistic (SSP1-2.6) climate scenarios (). The choice of these five models was reliant on the completeness of all covariates and the general model performance (). Future, projected, human population densities for 2055, i.e., the middle of the 2041–2070 period, under these same SSP scenarios as developed by were included into our future predictions. Other covariates, including cropland density, EVI, wind, and radiation values are not altered in these future predictions. We calculated ensemble means, relative to the contemporary values, of the five CMIP6 2041–2070 climate projections and population shifts for the SSP1-2.6 and SSP3–7 scenarios.
Results
Model development
The random forest model attained an accuracy of ~89% on the testing data using a model with 63 trees, a node size of 36, and four predictors to each split (mtry). A more comprehensive F1 score, the harmonic mean of precision and recall, has a value of 0.89, where values of 1 indicate perfect precision and recall (Supplementary Tables S1, S2 for the confusion matrix and full list of metrics).
A total of 19 co-variates were retained in the best-fit model (Table 1). Assessing the variables of importance (Supplementary Figure S2), we note that population density dominates our model, together with maximum radiation and annual temperature range (BIO7). The sensitivity to precipitation of the driest month (BIO14), precipitation seasonality (BIO15), mean solar radiation, and the maximum temperature of the warmest month (BIO5) also contribute significantly.
Table 1
| Category | Variable | Rational | Abbreviated variable name | Res | Ref |
|---|---|---|---|---|---|
| Bioclimatic | Mean diurnal range (mean of monthly (max temp − min temp)) | Bioclimatic variables summarize the climate and climate variability. These variables represent climatological annual trends, seasonality, and extreme or limiting environmental factors. Additional variables on radiation and wind conditions are also included as they link to aphid habitat and spread (; ). | BIO2 | 30s | |
| Temperature seasonality (standard deviation × 100) | BIO4 | 30s | |||
| Max temperature of warmest month | BIO5 | 30s | |||
| Min temperature of coldest month | BIO6 | 30s | |||
| Temperature annual range (BIO5-BIO6) | BIO7 | 30s | |||
| Precipitation of driest month | BIO14 | 30s | |||
| Precipitation seasonality (coefficient of variation) | BIO15 | 30s | |||
| Precipitation of wettest quarter | BIO16 | 30s | |||
| Precipitation of warmest quarter | BIO18 | 30s | |||
| Precipitation of coldest quarter | BIO19 | 30s | |||
| Mean annual surface downwelling shortwave radiation (MJ m−2 day−1) | rsds_mean | 30s | |||
| Range of surface downwelling shortwave radiation across mean monthly radiation (MJ m−2 day−1) | rsds_range | 30s | |||
| Surface downwelling shortwave radiation during month with the highest mean maximum radiation (MJ m−2 day−1) | rsds_max | 30s | |||
| Range in near-surface wind speed across mean monthly wind speeds (m s−1) | sfcWind_range | 30s | |||
| Near-surface wind speed during month with lowest wind speeds (m s−1) | sfcWind_min | 30s | |||
| Socioeconomic | Population density as a count | Disease spread is mediated by human factors. | population | 30s | WorldPop et al. (2018) |
| Banana cropland density - CROPGRIDS | Presence of banana crops is a prerequisite for presence and disease transmission. | croparea | 30s | ||
| Vegetation | Mean annual MODIS EVI – 2014–2023 EVI | Vegetation indices provide an indication on land use and land cover, as well as the health of the vegetation. | evi_mean | 30s | |
| Minimum MODIS EVI of month with lowest EVI across 2014 - 2023 | evi_min | 30s |
Final retained covariates for Africa-wide banana bunchy top disease (BBTD) modeling.
Areas at risk for BBTD
Three main regions are identified as current high-risk areas (probabilities > 0.8; Figure 1), namely, (i) East-Central Africa, between and around the African Great Lakes of Lake Albert, Lake Edward, Lake Kivu, the northern side of Lake Tanganyika, and Lake Victoria; (ii) the western regions of Central Africa, including Gabon and the Republic of the Congo; and (iii) southern regions of West Africa, including southern regions of Cote d’Ivoire, Ghana, Togo, Benin, and Nigeria. In DR Congo, more scattered high-risk areas are identified along the southern savanna zones, the eastern highlands, and a few locations in northern DR Congo. These areas correspond to locations with higher population densities in DR Congo.
Figure 1
When scaled spatially within the contemporary time frame, the Area-of-Applicability excludes the high-altitude regions of North-East Africa, including the Ethiopian highlands and mountain ranges along the Great Rift Valley (Kenya). In addition, several coastal areas are outside of the Area-of-Applicability of the model.
Future scenarios
Future predictions based on scenarios of climate and human population changes for 2041-2070 (Figure 2) show increased ensemble mean (arithmetic mean) probabilities of BBTD in Central-East (including Uganda and scattered regions in Tanzania) and North-East Africa (including Ethiopia), and countries approximately along the 5-10°N latitude belt (Cote d’Ivoire, Ghana, Togo, Benin, Nigeria, Central African Republic, South Sudan, and Ethiopia).
Figure 2
The same overall areas of increased BBTD probabilities are identified using the SSP1-2.6 and SSP3–7 scenarios. Several countries in Central Africa are predicted to be at a lower risk in future scenarios (both SSP1-2.6 and SSP3-7), including southern Cameroon, Equatorial Guinea, Gabon, Republic of the Congo, and DR Congo.
Discussion
Three main clustered regions, each crossing multiple country borders, have been identified as high-risk (> 0.8) areas for BBTD introduction, establishment, and wider spread. At the applied continental scale, population density is a critical driving feature for this risk, alongside maximum solar radiation and the annual temperature range.
The resulting risk map is markedly different from the Africa-wide risk assessment carried out by , with a reduced risk pressure in the central Congo Basin and a concentrated pressure in more densely populated regions in West Africa, western Central Africa, and East-Central Africa. While differences in used covariates partially explain the varying outcomes [CHELSA and CROPGRIDS data vs WorldClim () and SPAM ()], the introduction of population density drives the major shift in the risk map outcome.
The trade and use of BBTV-infected planting material is a critical aspect of BBTD transmission over a range of distances. At local scales (i.e., field and village-level), winged banana aphids, moving over relatively short distances, can transmit BBTV from infected to healthy plants, in addition to farmers transplanting suckers from infected mats within their own fields or trading within the community (, ). However, for local spread to occur, BBTV first has to be introduced into the community or landscape. This long-distance spread is mediated predominantly by trade or exchange of planting materials, in addition to rare reports of winged aphids being carried by strong winds over large distances (). The risk associated with such seed trade or exchange depends on the structure of trade networks, social interactions, and landscape connectivity (; Xing et al., 2020). These networks and interactions are generally more developed in larger (agricultural) populations, often going hand in hand with better road infrastructure and market access. The predominant informal nature of the banana seed system in Africa complicates the oversight, regulation, or quality control, which is often non-existent (; ; ). When stakeholders, generally smallholder farmers, trade or exchange planting material outside of the local banana production landscape without quality control, for example to diversify banana production, it opens up entire communities to potential disease incursion.
Toward the mid-century (~2055), predicted change in BBTD risk is related to both population shifts and changes in climate. Currently, the (potential) occurrence of BBTD seems consistent with the African (sub)tropical regions of banana cultivation, which also constitute the primary habitat for banana aphids (). Our future predictions, however, show a reduced BBTD risk in Central Africa, including in Gabon, the Republic of Congo, and DR Congo. Across this region, the thermal climate is predicted to shift from hot to torrid by this timeframe (), based on temperature-related potential evapotranspiration estimates. Such a shift in thermal climate could have three direct effects on BBTD risk within the modeled assessment: (i) reduced banana cropping potential, (ii) unfavorable habitat for banana aphids, and (iii) reduced habitability leading to a decline in human populations and/or large-scale migration. While not a desirable evolution, a reduction in banana production systems would inherently decrease the BBTV host potential and overall disease occurrence. Additionally, the habitat may no longer be within the suitable range for banana aphids and BBTV transmission. For example, in laboratory conditions, banana aphids do not thrive in temperatures above 27 °C-30 °C, leading to diminishing body sizes and populations (; ), although its habitat within complex field microclimates remains unclear. The threatened agricultural development and food security would further burden local populations, compounding already climatologically challenging living conditions. Smallholder farmers, in particular, in the African agricultural sector are highly vulnerable to climate change due to their reliance on rain-fed agriculture, limited access to irrigation, prevalent soil fertility issues, poor access to affordable agricultural inputs, insufficient information and extension services, limited preventative and curative strategies, and restricted access to credit, assistance, and insurance (Wichern et al., 2023; ). Furthermore, environmental stresses such as the increasing occurrence of heat stress will diminish both labor capability and productivity (), thereby intensifying the negative effects on crop yields. Additional exacerbating factors related to climate change such as flooding, land degradation, soil erosion, biodiversity loss, and extreme weather events, not accounted for in the model, further put pressure on the overall agricultural system and food security (; ). Compiled effects of climate change and crop diseases are likely to influence migration and displacement ().
On the other hand, BBTD is predicted to increase in the 5-10°N latitude belt and in Central-East Africa by the mid-century. Averaged predicted climate change (although extreme weather and events are not accounted for in the model) is likely to have a less pronounced impact on banana cultivation potential and aphid habitats in these regions. Conversely, the pressure from increased population density (Supplementary Figure S3) on natural and agronomic systems will likely drive an increased risk of BBTD, particularly from the perspective of movement of diseased planting material (as discussed above).
We acknowledge that the methodology presented to develop the Africa-wide risk map is purely data driven. Detailed inference about the processes involved in the future spread is therefore not possible. This is a commonality across many machine learning methods, where predictions are restricted to already seen data. We addressed this issue, of out-of-distribution data, through our area-of-applicability analysis. Furthermore, the future predictions do not account for changes in land use/land cover, although these are likely partially a reflection of changing pressures with population density combined with crop production possibilities in changing climates.
The developed current and future risk assessment maps at the continental scale should be used at their intended scale, namely, as an informative tool of large-scale risk areas of BBTD introduction, establishment, and spread. We recommend this tool to be used in combination with risk assessments and expert knowledge at smaller geographical scales, particularly at regional and national levels, to develop spread and disease management strategies, and to inform policy. Specifically, human population density, as a proxy for regions with a good road infrastructure and market access, experiencing intensified trade, and increased food security pressures, does not capture all aspects of the banana seed system, particularly at smaller scales. Risk assessments at regional and national levels can provide a more detailed and nuanced view of disease risk and priority areas (; ; ). At the continental scale, risk assessments can be used to identify risk areas in neighboring countries, in regions between which trade is common or present, and to understand national disease management, including prevention, mitigation, and control, in a broader context. The risk landscape in African disease management and in particular cross-border seed trade encompasses issues such as infrastructural challenges, custom inefficiencies, and varying degrees of political stability (; ). Accounting for such large-scale components of BBTD propagation is critical even at regional and national levels. As such, both localized solutions, including community engagement and training, and collaboration with international partners are critical components of effective risk management in the African context. Quantification of country-specific informal trading partners within the banana seed system remains challenging and will depend on localized assessments (e.g., standardized interviews with farmers) and expert knowledge (; ; ).
Nonetheless, in absence of regional or national BBTD risk assessments, our large-scale risk assessment could be used as a preliminary assessment to guide further monitoring and research. To illustrate this, we examined Tanzania, where the International Institute of Tropical Agriculture (IITA) recently conducted field surveys providing insights into BBTD incursions (). The exact survey data are currently unpublished and therefore not incorporated into our model, providing an opportunity to evaluate our model outcome at national level (Supplementary Figure S4). In Tanzania, banana production landscapes bordering Lake Victoria and Burundi are identified as high-risk areas and correspond to field survey findings. Moreover, BBTD field incursions found near larger cities, including Arusha, Tanga, and Dar es Salaam, correspond to clustered modelled risk areas. These areas correspond to locations with higher population densities and good road infrastructure, all pointing to introductions due to the movement of planting materials (), in line with our modelling efforts where population effects dominate predictions. Incursions identified along Lake Tanganyika (beyond the regions bordering Burundi) were however not identified via the large-scale risk mapping. This area generally has fewer banana production, but the robust road infrastructure likely facilitated incursions, even into communities where bananas are not a primary crop. As such, at least for Tanzania, we can conclude that the national subset of the large-scale map would provide a good preliminary guide, but it requires field verifications, monitoring, and overall refinement before national disease management strategies and policy are defined. Particularly, additional aspects might be in play at regional and national scales, not accounted for in the continental model, which should affect decisions in disease management, such as specific socioeconomic factors, banana production systems, cropland connectivity, banana cultivar preference and susceptibility to BBTD, local seed transactions, access to extension and training, and regulations in play (; ; ; Xing et al., 2020; ).
Statements
Data availability statement
The original contributions presented in the study are publicly available. This data can be found here: https://doi.org/10.5281/zenodo.17513942. Field data are available upon request.
Ethics statement
The present study did not carry out any new field surveys and hence did not involve any new on-site interviews with human subjects. The present study uses data from a large number of previous/published field surveys.
Author contributions
GB: Conceptualization, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Writing – original draft, Writing – review & editing. KH: Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Visualization, Writing – original draft, Writing – review & editing, Conceptualization. WO: Methodology, Writing – review & editing. EK: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Validation, Visualization, Writing – original draft, Writing – review & editing.
Funding
The author(s) declared that financial support was received for this work and/or its publication. This study was funded by the Sustainable Farming Science Program of the CGIAR.
Acknowledgments
We would like to acknowledge the Sustainable Farming Science Program of the CGIAR for its financial contribution to this study.
Conflict of interest
The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The author(s) declared that generative AI was not used in the creation of this manuscript.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fagro.2026.1702397/full#supplementary-material
References
1
AbiolaA.Zandjanakou-TachinM.AoudjiK. N. A.Avocevou-AyissoC.KumarP. L. (2020). Adoption of roguing to contain banana bunchy top disease in south-east Bénin: Role of farmers’ knowledge and perception. Int. J. Fruit Sci.20, 720–736. doi: 10.1080/15538362.2019.1673277
2
AlmekindersC. J.WalshS.JacobsenK. S.Andrade-PiedraJ. L.McEwanM. A.de HaanS.et al. (2019). Why interventions in the seed systems of roots, tubers and bananas crops do not reach their full potential. Food Secur.11, 23–42. doi: 10.1007/s12571-018-0874-4
3
AndersenK. F.BuddenhagenC.RachkaraP.GibsonR.KaluleS.PhillipsD.et al. (2019). Modeling epidemics in seed systems and landscapes to guide management strategies: The case of sweet potato in Northern Uganda. Phytopathology109, 1519–1532. doi: 10.1094/PHYTO-03-18-0072-R
4
AnhaltM. D.AlmeidaR. P. P. (2008). Effect of temperature, vector life stage, and plant access period on transmission of banana bunchy top virus to banana. Phytopathology98, 743–748. doi: 10.1094/PHYTO-98-6-0743
5
BaninlaY.SharifiA.AllamZ.TumeS. J. P.GangtarN. N.GeorgeN. (2022). An overview of climate change adaptation and mitigation research in Africa. Front. Climate4, 976427. doi: 10.3389/fclim.2022.976427
6
BentleyJ. W.Andrade-PiedraJ.DemoP.DzomekuB.JacobsenK.KikulweE.et al. (2018). Understanding root, tuber, and banana seed systems and coordination breakdown: a multi-stakeholder framework. J. Crop Improvement32, 599–621. doi: 10.1080/15427528.2018.1476998
7
BoloyF. N.NkosiB. I.LosimbaJ. K.BungamuziC. L.SiwakoH. M.BaloweF. W.et al. (2014). Assessing incidence, development and distribution of banana bunchy top disease across the main plantain and banana growing regions of the Democratic Republic of Congo. Afr. J. Agric. Res.9, 2611–2623. doi: 10.5897/AJAR2014.8751
8
BouwmeesterH.BlommeG.OmondiA. B.OcimatiW. (2023). Banana bunchy top disease in africa—predicting continent-wide disease risks by combining survey data and expert knowledge. Plant Pathol.72, 1476–1490. doi: 10.1111/ppa.13764
9
BrunP.ZimmermannN. E.HariC.PellisierL.KargerD. N. (2022). Global climate-related predictors at kilometre resolution for the past and future. Earth System Sci. Data14, 5573–5603. doi: 10.5194/essd-14-5573-2022
10
BusogoroJ. P.SulemanJ. T.AmenyT.NdamugobaD.NajjembaA.KirichuS.et al. (2023). Reviving the banana industry in Malawi through integrated crop management of local banana germplasm. Tropical Agriculture And Development, 67, 1–14.
11
ChettoR.MdemuM.KihilaJ. (2024). Climate change-related drivers of migration in East Africa: an integrative review of the literature. Environ. Sci. Sustain. Dev.9, 42–56. doi: 10.21625/essd.v9i1.1005
12
Commonwealth Agricultural Bureau (CAB) (1968). Distribution maps of pests. Series A (Agricultural), Map No. 242 (Pentalonia nigronervosa Coq.), Published at 56 Queen’sGate, London, S.W.7: CAB.
13
DajumaA.SyllaM. B.TallM.AlmazrouiM.YassaN.DiedhiouA.et al. (2023). Projected expansion of hottest climate zones over Africa during the mid and late 21st century. Environ. Research: Climate2, 025002. doi: 10.1088/2752-5295/acc08a
14
DaleJ. L. (1987). Banana bunchy top: an economically important tropical plant virus disease. Advanced Virus Res.33, 301–325. doi: 10.1016/S0065-3527(08)60321-8
15
DatoK. M. G.DégbègniM. R.AtChadéM. N.Zandjanakou TachinM.HounkonnouM. N.OmondiA. B. (2021). Spatial parameters associated with the risk of banana bunchy top disease in smallholder systems. PloS One16, e0260976. doi: 10.1371/journal.pone.0260976
16
DescheemaekerK.MasikatiP.MacCarthyD. S. (2025). “ Climate change: impacts and adaptation in smallholder farming systems in sub-Saharan Africa,” in Pathways to African Food Security (London: Routledge), 150–159.
17
DhekneyS. A.ZebeloS.SardaruP.NatarajanP.TubeneS. L.NindoC.et al. (2025). Addressing global food security in Africa through training of next generation of researchers in plant pest diagnostics, in vitro culture, and clean plant production. In Vitro Cell. Dev. Biology Plant61, 535–544. doi: 10.1007/s11627-024-10512-2
18
DidanK. (2021). MODIS/Terra Vegetation Indices Monthly L3 Global 1km SIN Grid V061 ( NASA EOSDIS Land Processes Distributed Active Archive Center). doi: 10.5067/MODIS/MOD13A3.061
19
FickS. E.HijmansR. J. (2017). WorldClim 2: new 1km spatial resolution climate surfaces for global land areas. Int. J. Climatology37, 4302–4315. doi: 10.1002/joc.5086
20
GaidashovaS. V.KajugaJ.BlommeG.OcimatiW.KearsleyE.HufkensK.et al. (2025). Banana bunchy top disease in Rwanda: identifying disease presence and potential areas at risk. Phytopathology. 115, 1728–1736. doi: 10.1094/PHYTO-04-25-0135-R
21
HausfatherZ.PetersG. P. (2020). Emissions - the ‘business as usual’ story is misleading. Nature577, 618–620. doi: 10.1038/d41586-020-00177-3
22
HooksC. R. R.WrightM. G.KabasawaD. S.ManandharR.AlmeidaR. P. P. (2008). Effect of banana bunchy top virus infection on morphology and growth characteristics of banana. Ann. Appl. Biol.153, 1–9. doi: 10.1111/j.1744-7348.2008.00233.x
23
HootenM. B.HobbsN. T. (2015). A guide to Bayesian model selection for ecologists. Ecol. Monogr.85, 3–28. doi: 10.1890/14-0661.1
24
HufkensK.CampitelliE. (2023). appeears: Programmatic interface to the NASA AppEEARS API. Zenodo. doi: 10.5281/zenodo.7938190
25
International Food Policy Research Institute (2019). Global spatially disaggregated crop production statistics data for 2010 version 2.0. doi 10.7910/DVN/PRFF8V
26
JacobsenK.OmondiA. B.AlmekindersC.AlvarezE.BlommeG.DitaM.et al. (2019). Seed degeneration of banana planting materials: strategies for improved farmer access to healthy seed. Plant Pathol.68, 207–228. doi: 10.1111/ppa.12958
27
KjellstromT.BriggsD.FreybergC.LemkeB.OttoM.HyattO. (2016). Heat, human performance, and occupational health: A key issue for the assessment of global climate change impacts. Annu. Rev. Public Health37, 97–112. doi: 10.1146/annurev-publhealth-032315-021740
28
KolombiaY. A.OviasuyiT.DzolaA. K.Ale Gonh-GohA.AtsuT.OresanyaA.et al. (2021). First report of banana bunchy top virus in banana (Musa spp.) and its eradication in Togo. Plant Dis.105, 3312. doi: 10.1094/PDIS-03-21-0473-PDN
29
KuhnM.WickhamH. (2020). Tidymodels: a collection of packages for modeling and machine learning using tidyverse principles. Available online at: https://www.tidymodels.org (Accessed July 15, 2025).
30
KumarP. L.HannaR.AlabiC. O. J.SokoM. M.ObenT. T.VanguG. H. P.et al. (2011). Banana bunchy top virus in sub-Saharan Africa: investigations on virus distribution and diversity. Virus Res.159, 171–182. doi: 10.1016/j.virusres.2011.04.021
31
KumarP. L.SelvarajanR.Iskra-CaruanaM. L.ChabannesM.HannaR. (2015). Biology, etiology, and control of virus diseases of banana and plantain. Adv. Virus Res.91, 229–269. doi: 10.1016/bs.aivir.2014.10.006
32
KursaM. B.RudnickiW. R. (2010). Feature selection with the Boruta package. J. Stat. Software36, 1–13. doi: 10.18637/jss.v036.i11
33
LudwigM.Moreno-MartinezA.HölzelN.PebesmaE.MeyerH. (2023). Assessing and improving the transferability of current global spatial prediction models. Global Ecol. Biogeography32, 356–368. doi: 10.1111/geb.13635
34
MahoneyM. J. (2023). Waywiser: ergonomic methods for assessing spatial models. arXiv preprint arXiv:2303.11312. doi: 10.48550/arXiv.2303.11312
35
MahoneyM. J.JohnsonL. K.SilgeJ.FrickH.KuhnM.BeierC. M. (2023). Assessing the performance of spatial cross-validation approaches for models of spatially structured data. arXiv preprint arXiv:2303.07334. doi: 10.48550/arXiv.2303.07334
36
MahukuG.KumarP. (2023). Banana bunchy top virus (BBTV) delimiting survey in Tanzania: Phase 1 survey summary report (Ibadan: IITA)13.
37
MenardiG.TorelliN. (2014). Training and assessing classification rules with imbalanced data. Data Min. Knowledge Discov.28, 92–122. doi: 10.1007/s10618-012-0295-5
38
MeyerH.PebesmaE. (2021). Predicting into unknown space? Estimating the area of applicability of spatial prediction models. Methods Ecol. Evol.12, 1620–1633. doi: 10.1111/2041-210X.13650
39
MissangaJ. S.MmbandoG. S. (2025). Challenges of food security in Africa in the face of climate change: potential of the clustered regularly interspaced short palindromic repeats (CRISPR)-cas9 technology to address these challenges. Discover Sustainability6, 305. doi: 10.1007/s43621-025-01092-1
40
MondoJ. M.NdekoA. B.AyagirweR. B.MatitiH. M.BahatiC.CivavaR. M. (2024). Seed delivery system in northeastern Democratic Republic of Congo: characteristics, stakeholders mapping and challenges towards building a well-functioning seed system under humanitarian and security crises. Int. J. Ag. Sustain. 22, 2335128.
41
Mouafo-TchindaR. A.EthertonB. A.Plex SuláA. I.Andrade-PiedraJ.OgeroK.OmondiB. A.et al. (2024). Pathogen and pest risks to vegetatively propagated crops in humanitarian contexts: Towards a national plant health risk analysis for Cameroon and Ethiopia. bioRxiv, 2024–2002. doi: 10.1101/2024.02.12.580019
42
MulugoL.KyazzeF. B.OmondiA. B. (2024). Strengthening banana seed systems among smallholder farmers in bugiri district, Uganda. Available online at: https://ssrn.com/abstract=4987642 or http://dx.doi.org/10.2139/ssrn.4987642 (Accessed July 15, 2025).
43
NgatatS.HannaR.Doumtsop FotioA. R.LienouJ. A.Nanga NangaS.Fotso KuateA.et al. (2024). Distribution and diversity of emergent Banana bunchy top virus infecting banana and plantain in Cameroon, Central Africa. J. Phytopathol.172, e13279. doi: 10.1111/jph.13279
44
NiyongereC.LosengeT.AtekaE. M.NkezabahiziD.BlommeG.LepointP. (2012). Occurrence and distribution of banana bunchy top disease in the Great Lakes region of Africa. Tree Forestry Sci. Biotechnol.6, 102–107.
45
NiyongereC.OmondiB. A. O.BlommeG. (2015). “ The Banana bunchy top disease,” in Virus diseases of tropical and subtropical crops. Eds. TennantP.FerminG. ( CAB International Plant Protection Series 12/2015, Wallingford, United Kingdom), 17–26. Available online at: https://hdl.handle.net/10568/78254 (Accessed July 15, 2025).
46
OcimatiW.BouwmeesterH.GrootJ.TittonellP.BrownD.BlommeG. (2019). The risk posed by Xanthomonas wilt disease of banana: mapping of disease hotspots, fronts and vulnerable landscapes. PloS One14, e0213691. doi: 10.1371/journal.pone.0213691
47
OcimatiW.OgwalG.TazubaA. F.KubiribaJ.TugumeJ.ErimaR.et al. (2024). Mapping the vulnerability of banana production landscapes in Uganda to banana bunchy top disease. Front. Agron.6, 1401478. doi: 10.3389/fagro.2024.1401478
48
OcimatiW.TazubaA. F.TushemereirweW. K.TugumeJ.OmondiB. A.AcemaD.et al. (2021). First report of banana bunchy top disease caused by Banana bunchy top virus in Uganda. New Dis. Rep.44, e12052. doi: 10.1002/ndr2.12052
49
OkoyeC. C.OfodileO. C.TulaS. T.NifiseA. O. A.FalaiyeT.EjairuE.et al. (2024). Risk management in international supply chains: A review with USA and African Cases. Magna Scientia Advanced Res. Rev.10, 256–264. doi: 10.30574/msarr.2024.10.1.0024
50
OmondiB. A.SokoM. M.NduwimanaI.DelanoR. T.NiyongereC.SimbareA.et al. (2020). The effectiveness of consistent roguing in managing banana bunchy top disease in smallholder production in Africa. Plant Pathol.69, 1754–1766. doi: 10.1111/ppa.13253
51
R Core Team (2024). _R: A Language and Environment for Statistical Computing_. Vienna, Austria: R Foundation for Statistical Computing. Available online at: https://www.R-project.org/.
52
RetkuteR.GilliganC. A. (2025). Developing a spatio-temporal model for banana bunchy top disease: leveraging remote sensing and survey data. Front. Plant Sci.16, 1521620. doi: 10.3389/fpls.2025.1521620
53
RobsonJ. D.WrightM. G.AlmeidaR. P. P. (2007). Biology of Pentalonia nigronervosa (Hemiptera, Aphididae) on banana using different rearing methods. Environ. Entomology36, 46–52. doi: 10.1603/0046-225X(2007)36[46:BOPNHA]2.0.CO;2
54
ShahD. A.De WolfE. D.PaulP. A.MaddenL. V. (2023). Into the trees: random forests for predicting Fusarium head blight epidemics of wheat in the United States. Phytopathology113, 1483–1493. doi: 10.1094/PHYTO-10-22-0380-R
55
ShimwelaM. M.MahukuG.MbanzibwaD. R.MkamiloG.MarkD.MoshaH. I.et al. (2022). First report of banana bunchy top virus in banana and plantain (Musa spp.) in Tanzania. Plant Dis.106, 1312. doi: 10.1094/pdis-07-21-1387-pdn
56
SileshiG. W.GebeyehuS. (2021). Emerging infectious diseases threatening food security and economies in Africa. Global Food Secur.28, 100479. doi: 10.1016/j.gfs.2020.100479
57
Soley-GuardiaM.Alvarado-SerranoD. F.AndersonR. P. (2024). Top ten hazards to avoid when modeling species distributions: a didactic guide of assumptions, problems, and recommendations. Ecography2024, e06852. doi: 10.1111/ecog.06852
58
TangF. H.NguyenT. H.ConcheddaG.CasseL.TubielloF. N.MaggiF. (2024). CROPGRIDS: a global geo-referenced dataset of 173 crops. Sci. Data11, 413. doi: 10.1038/s41597-024-03247-7
59
TatineniS.HeinG. L. (2023). Plant viruses of agricultural importance: Current and future perspectives of virus disease management strategies. Phytopathology113, 117–141. doi: 10.1094/phyto-05-22-0167-rvw
60
WangX.MengX.LongY. (2022). Projecting 1 km-grid population distributions from 2020 to 2100 globally under shared socioeconomic pathways. Sci. Data9, 563. doi: 10.1038/s41597-022-01675-x
61
WichernJ.HammondJ.van WijkM. T.GillerK. E.DescheemaekerK. (2023). Production variability and adaptation strategies of Ugandan smallholders in the face of climate variability and market shocks. Climate Risk Manage.40, 100490. doi: 10.1016/j.crm.2023.100490
62
WorldPop(www.worldpop.org - School of Geography and Environmental Science, University of SouthamptonDepartment of Geography and Geosciences, University of LouisvilleDepartement de Geographie, Universite de Namur) and Center for International Earth Science Information Network (CIESIN), Columbia University (2018). Global High Resolution Population Denominators Project - Funded by The Bill and Melinda Gates Foundation (OPP1134076). doi: 10.5258/SOTON/WP00674
63
WrightM. N.ZieglerA. (2017). ranger: A fast implementation of random forests for high dimensional data in C++ and R. J. Stat. Software77, 1–17. doi: 10.18637/jss.v077.i01
64
XimbaS. P. F.TshabalalaJ.GubbaA.JoosteA. E. C. (2022). Monitoring the distribution of banana bunchy top virus in South Africa: a country-wide survey. Arch. Virol.167, 1433–1441. doi: 10.1007/s00705-022-05451-5
65
XingY.Hernandez NopsaJ. F.AndersenK. F.Andrade-PiedraJ. L.BeedF. D.BlommeG.et al. (2020). Global cropland connectivity: A risk factor for invasion and saturation by emerging pathogens and pests. BioScience70, 744–758. doi: 10.1093/biosci/biaa067
Summary
Keywords
banana bunchy top virus, BBTD, climate change, Musa spp., risk mapping
Citation
Blomme G, Hufkens K, Ocimati W and Kearsley E (2026) Informing African agricultural health: integrating human population dynamics and climate change into banana bunchy top disease risk assessment. Front. Agron. 8:1702397. doi: 10.3389/fagro.2026.1702397
Received
09 September 2025
Revised
15 January 2026
Accepted
27 January 2026
Published
17 February 2026
Volume
8 - 2026
Edited by
Pedro Laborda, Nantong University, China
Reviewed by
Nagamani Balagurusamy, Autonomous University of Coahuila, Mexico
Renata Retkute, University of Cambridge, United Kingdom
Updates
Copyright
© 2026 Blomme, Hufkens, Ocimati and Kearsley.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Guy Blomme, g.blomme@cgiar.org
Disclaimer
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.