Improving climate suitability for *Bemisia tabaci* in East Africa is correlated with increased prevalence of whiteflies and cassava diseases

Climate change is widely expected to affect ranges and abundance of arthropod pests, weeds, and diseases, but detecting such changes against noisy climatic and biological variability is challenging. Insect-vectored emerging infectious diseases are theoretically sensitive to climate change. Cassava is a critical subsistence crop in Africa and pandemics of cassava mosaic disease (CMD) and cassava brown streak disease (CBSD) have been reported in East and Central Africa since the late 1990s, with production losses up to 47% raising food security concerns. Hypotheses for recent cassava disease epidemics include a novel recombinant begomovirus, range expansion or genetic changes within native Bemisia tabaci populations, and synergistic interactions between high B. tabaci populations and viruses. B. tabaci is a cryptic species complex; in East and Central Africa, SSA1 (subgroups SG1 and SG2), SSA2, and SSA3 predominate on cassava. Previous correlative species distribution models for B. tabaci have had limitations and implausible projections. CLIMEX, a process-based climatic niche model, allows mechanistic exploration of climatic suitability and seasonal dynamics. A CLIMEX model developed for B. tabaci MEAM1 using long-term climate averages encompassed the known distributions of B. tabaci complex members, including SSA taxa. Research question/hypotheses: Are historical changes in climate suitability for B. tabaci correlated with observed increases in B. tabaci abundance and cassava disease prevalence in East and Central Africa? Is the MEAM1-based CLIMEX model applicable to SSA B. tabaci? The study applies an end-to-end approach: compare SSA development responses with MEAM1, test correlation between modelled suitability and a 13-year Ugandan abundance time series, evaluate probability of occurrence across Uganda, Tanzania, and Malawi, and assess temporal trends in climatic suitability over 39 years in regions affected by pandemics.

The paper reviews prior attempts to link climate and pest/disease dynamics. Simple correlative species distribution models (e.g., consensus models, MaxEnt) applied to B. tabaci and cassava risk factors have provided limited mechanistic insight and showed poor fit, low sensitivity, and implausible projections into unsuitable climates, with overfitting and artefacts. High AUC values can be misleading for bioclimatic models. Extrapolating correlative models to climate change scenarios is unreliable due to novel climate spaces. In contrast, CLIMEX has been used to assess invasive organisms and simulate seasonal and long-term changes by cross-validating parameters against distribution, phenology, and experimental data. A previously developed CLIMEX model for B. tabaci MEAM1, fitted to 1981–2010 climate, encompassed known records of B. tabaci complex members including SSA taxa and the cassava epidemic zone, suggesting overlapping fundamental niches among cryptic species with realized differences driven by biotic factors.

Process-based climatic suitability modelling (CLIMEX): The Compare Locations model simulates weekly population responses to climate, calculating annual Growth Index (GI_A) as the average of weekly growth (temperature and moisture indices) during favourable conditions. Stress indices (cold, wet, hot, dry) and interactions (cold-wet, cold-dry, hot-wet, hot-dry) account for survival during unfavourable seasons. The Ecoclimatic Index (EI) summarizes growth and stress to scale suitability from 0 to 100. For baseline Compare Locations, a 30-year average of monthly climate (1981–2010) was derived. Minimum/maximum temperatures and precipitation were used with calculated 9:00 and 15:00 relative humidity (via Magnus equation and vapour pressure), aggregated into an SQLite database. Time-series modelling (Compare Locations/Years): A 39-year monthly time-series (1978–2017/1979–2017) from CRU (0.5° x 0.5°) provided maximum temperature, minimum/maximum temperature, vapour pressure, and precipitation totals. The B. tabaci MEAM1 CLIMEX model was run to produce yearly EI, GI_A, Temperature Index (TI), and Moisture Index (MI). Linear time trend models were fitted at each grid cell in R to estimate average annual changes; standardized effect sizes (t values) and 95% confidence intervals were mapped. Laboratory development rate comparison (SSA vs MEAM1): To assess applicability of MEAM1 parameters to SSA taxa, development times from egg to adult were measured for SSA1-SG1, SSA1-SG2 (colonies established from Kayingo, Uganda, Feb 2016) and SSA2 (Kiboga, Uganda, Aug 2013) under six constant temperatures (15, 20, 25, 30, 35, 40 °C). Each temperature had up to five replicates on cassava cultivar Ebwanateraka in Lock and Lock containers under controlled photoperiod and RH. Adults were allowed to oviposit for 24 h at 25 °C; nymph emergence counted at 10 days; plants exposed to treatment temperature for 5 days, then held at 25 °C and monitored every 2 days for adult emergence. Development rate was calculated as inverse of development time; mean times across replicates were plotted against temperature. Uganda field abundance analysis (2004–2017): Field survey data (7,627 summaries across 96 districts; gap in 2016) of adult B. tabaci counts on cassava and CBSD prevalence were aggregated to 121 CLIMEX climate grid cells. A linear mixed-effects model (REML) related log(count+1) to modelled EI per cell, assessing consistency across years (year as random effect/interaction examined). East Africa probability of presence (2015–2016): Cassava fields were surveyed in seven regions across Uganda, Tanzania, and Malawi in Sept 2015 and Apr 2016. Up to 10 fields per region of known varieties were sampled; adult B. tabaci on the top five leaves of 30 plants per field were categorized (0, 1–9, 10–99, 100–200, >200). For analysis, presence/absence was used, presuming SSA identity (with later molecular confirmation). Geocoded locations were intersected with CLIMEX outputs for East Africa generated using a 30-year climate average dataset centered on 1995 (CRU). A logistic regression (glm, binomial link) modeled probability of adult presence as a function of EI. The aim was pattern agreement rather than precise abundance prediction.

- SSA development rates align with MEAM1 parameters: Observed SSA1-SG1, SSA1-SG2, and SSA2 development times showed shortest durations between 25–30 °C and strong increases beyond 35 °C (especially SSA2). The CLIMEX cardinal temperatures used for MEAM1 (DV0 12 °C, DV1 28 °C, DV2 32 °C, DV3 42 °C) are consistent with SSA taxa, supporting use of the MEAM1 model for SSA species. - Baseline climatic suitability pattern: High EI in northwestern East Africa (e.g., Uganda) and southeastern coastal Tanzania with a depression in central Tanzania; global suitability covers tropics/subtropics and warm temperate regions, consistent with distribution records. - Uganda 2004–2017 field abundance: A statistically consistent, positive relationship between modelled EI and observed adult B. tabaci counts across 13 years; interannual variation in slope appeared random with 2009 an outlier. Data encompassed 7,627 field summaries across 96 districts aggregated to 121 climate cells. - East Africa 2015–2016 presence: Probability of adult presence on cassava displayed a sigmoid relation with EI. Differences among sites at moderate EI suggest non-climatic influences (crop type, age, landscape crops, natural enemies, competition, insecticide use). - Time-series trend (1979–2017): Despite strong interannual variability, significant increases in EI and GI occurred across large areas of East and Central Africa, especially the Democratic Republic of the Congo, Uganda, Rwanda, Burundi, and western Tanzania and Kenya. Slight decreases were detected in some lower-lying coastal regions (e.g., parts of Mozambique, sub-coastal Tanzania, northern Kenya), often associated with decreasing rainfall. Increases were primarily driven by rising temperature suitability (TI); decreases in some locations reflected reduced soil moisture suitability and drought stress (MI). Locations in the epidemic center (e.g., Luweero, Uganda; Mwanza, Tanzania) showed improving temperature and moisture conditions over time. - Epidemiological concordance: Spatial-temporal increases in climatic suitability coincided with reported increases in B. tabaci abundance and spread of CMD/CBSD in East and Central Africa. The modelled decrease in coastal Tanzanian suitability aligns with observed three-fold abundance declines at coastal sites (1994–2009).

Findings support that historical climatic changes improved suitability for B. tabaci in regions where cassava disease pandemics expanded. The mechanistic CLIMEX model fitted for MEAM1 is relevant for SSA taxa, as evidenced by congruent development responses and consistent field relationships. The end-to-end attribution approach indicates that observed changes are unlikely due to internal variability alone; applying a process-based model to historical climate produced statistically significant spatial patterns matching increases in whitefly abundance and cassava disease prevalence. While the movement of infected cassava cuttings can seed new foci, sustaining epidemics in landscapes is strongly influenced by vector dynamics; B. tabaci abundance increases driven by hotter and, in some epidemic centers, drier conditions likely facilitated virus transmission. Regions with decreasing suitability were already marginal for cropping and became drier. Overall, rising temperatures were the dominant driver of increased suitability, with reductions in rainfall sometimes enhancing favorability for B. tabaci (consistent with susceptibility of nymphs to raindrop dislodgement). These results provide among the first evidence linking historical climate changes to increased abundance of an insect vector contributing to a plant disease pandemic and underscore the need to integrate climate considerations into cassava disease management.

This study demonstrates that modelled climatic suitability for Bemisia tabaci increased over the past four decades across large parts of East and Central Africa where cassava disease pandemics expanded, and that these trends align with observed increases in whitefly abundance and disease prevalence. Laboratory data confirm that SSA taxa share thermal development responses with MEAM1, validating the use of the existing CLIMEX model. The CLIMEX Compare Locations/Years tool, combined with gridded climate time series, provides a practical mechanism for joint attribution of climate change impacts on biological systems. Given sustained increases in suitability, long-term, continual investment is needed in cassava breeding for resistance to evolving virus strains, complemented by biological and cultural control of B. tabaci. There is an urgent need to assess risks to West Africa should CBSD/UCBSV or other viruses be introduced and to clarify the vector competence of West African whiteflies. Future work should apply climate change scenarios to project potential extents of epidemics and evaluate integrated management strategies under evolving climates.

- Species/identity in field data: Many field abundance datasets did not determine B. tabaci species identity; analyses presumed SSA in some surveys and used adults presence/absence in 2015–2016. - Proxy modelling: The CLIMEX model was parameterized for MEAM1 and applied to SSA taxa based on overlapping development responses; subtle interspecific differences or non-climatic factors (hosts, natural enemies, competition) may affect realized niches and abundance. - Survey heterogeneity: Ugandan surveys varied in intensity and spatial coverage across years (gap in 2016), potentially influencing abundance estimates despite mixed-effects modeling. - Non-climatic confounders: Crop variety, age, landscape composition, agronomic practices, and insecticide use influence abundance and were not explicitly modeled. - Spatial/temporal resolution: Climate inputs at 0.5° resolution may miss local microclimatic and irrigation effects; monthly time steps may smooth extreme events. - Attribution scope: While trends were linked to historical climate changes, formal partitioning of anthropogenic vs natural forcings was not performed; conclusions are based on concordance using an end-to-end approach rather than detection/attribution with climate models. - Quantitative predictive precision: The relationship between EI and abundance is correlative and intended for pattern agreement rather than precise abundance prediction.