Incorporating hydrology into climate suitability models changes projections of malaria transmission in Africa

Malaria remains a major climate-sensitive vector-borne disease, causing an estimated 435,000 deaths from 219 million cases in 2017, with 92% of deaths in Africa. Temperature strongly influences key components of malaria transmission, and although uncertainties in temperature parameters persist, much of Africa falls within suitable thermal ranges for transmission. Availability of surface water for mosquito larval habitats is another critical control, but global datasets make direct estimation challenging. Consequently, monthly rainfall thresholds (e.g., 60–80 mm month−1) are often used as proxies, leading to wide variability in suitability estimates because hydrological processes (infiltration, evaporation, storage, routing) are not represented. Irrigation and reservoirs can also sustain year-round habitats but are typically omitted in threshold-based models. While process-based hydrology–biology malaria models exist at local scales, continental-scale malaria climatic suitability models lack hydrological realism. This study asks how incorporating hydrological processes alters current and future projections of malaria hydro-climatic suitability across Africa. The authors use downscaled daily climate projections from seven GCMs to drive a continental hydrological model and compare this hydrology-based approach to commonly used rainfall-threshold methods, assessing implications for season length, spatial patterns, and affected populations under climate change.

Prior work has established temperature effects on parasite and vector development, biting, and survival, with laboratory and field studies defining suitable temperature ranges and noting the role of water temperature on larval development. Classic models often use rainfall thresholds (e.g., 60 or 80 mm per month) as proxies for larval habitat availability, but thresholds vary widely across studies due to omission of key hydrological processes and spatial variability. Evidence shows irrigation and large dams can create persistent breeding sites enabling year-round transmission foci, yet these are rarely included in continental models. Historical maps (e.g., Lysenko and Semashko) and national surveys document transmission along major river corridors not captured by simple rainfall thresholds. Detailed local-scale hydrology–malaria models exist, but continental-scale assessments have lacked process-based hydrological representations, motivating the present study.

Climate projections: Seven downscaled climate projections under RCP 8.5 were produced by forcing EC-EARTH3-HR v3.1 with SST and sea-ice conditions from seven CMIP5 GCMs, preserving global extent and downscaled to 0.35° (~40 km). No bias correction was applied. Runoff simulations exhibit variable quality with positive bias in arid regions but skill in sub-seasonal variability. Hydrological modeling: LISFLOOD was configured at 0.5° (~55 km), forced with daily temperature and precipitation from 1971–2100. Potential evapotranspiration was computed via Penman–Monteith from daily meteorological variables. LISFLOOD simulates full water balance processes (snowmelt, freezing, runoff, infiltration, soil moisture redistribution in a three-layer profile, groundwater storage/flow) and routes runoff using a kinematic wave through the river network. Lakes, reservoirs, and retention areas are included. Input datasets included Hydrosheds elevation, river channel geometry, ISRIC SoilGrids, and SPOT-VGT Leaf Area Index (held constant for future projections). Estimating malaria hydrological suitability (LIS-MAL): Daily mean surface air temperature, rainfall, and runoff were computed for 1971–2005, 2011–2040, 2041–2070, and 2071–2100, and transformed to an equal-area projection. A runoff threshold of 30 m³ s⁻1 (≈1 mm depth over a grid cell) defined hydrological suitability; sensitivity analysis showed modest impact (±50% threshold changed suitable area by −6% and +10%). For each month, the mean number of days exceeding the threshold was compared to the temperature-dependent larval development period (Bayoh and Lindsay). Average daily air temperature defined development rate; a +2 °C offset was applied when air temperature was below the 31 °C water-temperature optimum to approximate warmer larval habitat water, without applying the offset above the optimum. Irrigation: Irrigated areas were added using the IWMI Irrigated Area map of Africa (2010) at 250 m resolution (rainfed areas excluded). Irrigated zones were assumed to provide year-round suitable water bodies; a 3 km buffer (reflecting typical Anopheles gambiae flight ranges) was applied, and these areas were considered hydrologically suitable. Irrigation data were combined with modeled runoff for baseline analyses; irrigation was not projected into the future (i.e., not applied to future scenarios). Hydro-climatic suitability and rainfall-threshold comparisons: Temperature suitability used the Mordecai et al. curve (viable 16–34 °C) and was combined with hydrological suitability to create monthly hydro-climatic masks. Season length was defined as the maximum number of consecutive suitable months. Comparisons were made against rainfall-threshold models from the literature, with thresholds converted to monthly values and rules (upper thresholds, catalyst months) preserved. The Tanser et al. approach (60 mm per month with an 80 mm catalyst month) served as a primary comparator. Additional analyses compared temperature suitability curves (Mordecai; Craig et al. with 18–40 °C and frost criterion; Parham & Michael with 20–40 °C) and their impacts on estimated suitable areas. Population exposure: Gridded WorldPop datasets (30 arc-second resolution) were used to estimate populations within climatically suitable areas for total population and under-5s. Baseline used 2015 WorldPop; future scenarios rescaled 2020 WorldPop to match UN country-level projections (medium variant) at the midpoint of each period, with low and high variants evaluated for comparison.

• Sensitivity to rainfall thresholds vs. temperature models: Thermal response curves produce relatively small differences in suitable area (max difference 2.86 million km², ~9% of Africa), whereas rainfall thresholds yield a much larger spread (range 16.14 million km², ~53% of Africa), indicating projections are far more sensitive to precipitation proxy choice than to temperature curve choice.
• Hydrology-based patterns: The LIS-MAL model reveals complex spatial patterns with river corridors (e.g., Nile, Niger, Senegal, Webi Juba, Webi Shabeelie) acting as year-round foci of suitability not captured by rainfall-threshold methods. Historical records support extended suitability along the Nile corridor. Validation suggests hydrology-based patterns reduce some overestimation (e.g., South Africa) seen in rainfall-threshold models, though underestimation occurs in parts of East Africa likely due to 20th-century drying trends and coarse categorical historical mapping.
• Season length and area: LIS-MAL estimates a larger area suitable for year-round transmission but a smaller area suitable for >3 months compared with a 60 mm rainfall threshold. Incorporating mapped irrigation increases both area and season length.
• Future projections (RCP 8.5): Across seven GCMs, models commonly show a decrease in suitable area from 1971–2005 to 2011–2040, then an increase realized by 2041–2070. For stable transmission (>3 continuous months), many models predict increases by 2071–2100, but LIS-MAL and the Tanser threshold indicate very small net increases in total area (+0.08 and +0.03 million km², respectively), implying larger increases predicted by other thresholds may be unfounded.
• Spatial shifts: The magnitude and pattern of geographic shifts are substantial with LIS-MAL, even if net area change is small. Both LIS-MAL and Tanser predict increased suitability in the Ethiopian Highlands driven by warming. Hydrological changes drive increases in parts of East Africa under warmer, wetter futures. LIS-MAL projects increased suitability along South Africa’s Caledon and Orange rivers rather than the eastern highlands focus from rainfall-threshold models. The expanding wedge across Niger seen in rainfall-threshold results is absent in LIS-MAL, which shows increases farther south near the Gulf of Guinea. Widespread aridity-driven decreases in southern Africa predicted by rainfall thresholds are muted in LIS-MAL due to maintained riverine water availability. A pronounced LIS-MAL finding is a substantial reduction in suitability along the Nile in South Sudan due to coincident declines in thermal and hydrological suitability, amplified by temperature-dependent larval development.
• Population exposure: Under RCP 8.5, populations living in areas suitable for stable transmission increase mainly due to demographic growth: by 2071–2100, +2,238 million (Tanser) and +1,765 million (LIS-MAL) additional people are within stable-suitability zones relative to baseline. The choice of UN population variant substantially affects totals, but patterns persist. Rainfall-threshold models estimate larger affected populations in southern and parts of West Africa, whereas LIS-MAL concentrates increases in East Africa (e.g., Somalia, Tanzania).

By explicitly modeling hydrological processes, the study addresses a key limitation of rainfall-threshold approaches that poorly approximate surface water availability for larval habitats. The hydrology-informed LIS-MAL model yields more realistic, spatially complex suitability patterns, especially highlighting river corridors as persistent transmission foci and moderating some aridity-driven declines predicted by threshold models. Although future warming strongly influences the directionality of temperature suitability, the spatial relocation of hydro-climatically suitable areas is equally sensitive to hydrological representation, underscoring the need to include hydrology in continental-scale projections. The small net change in total stable-suitability area under LIS-MAL contrasts with larger changes from simple thresholds, but LIS-MAL indicates large geographic redistributions with important public health implications for surveillance and intervention planning. The analysis is restricted to hydro-climatic suitability and does not attribute past or future transmission changes solely to climate, acknowledging that control measures, socioeconomics, land use, and health systems are critical cofactors.

Incorporating process-based hydrology into malaria climate suitability modeling substantially alters both present-day estimates and future projections across Africa. Compared with rainfall-threshold approaches, LIS-MAL indicates a smaller hydro-climatically suitable area for stable transmission and only minimal increases under late-century high-emissions scenarios, while revealing significant spatial shifts and riverine foci of year-round suitability. These insights refine estimates of populations at risk and highlight regions where suitability is likely to change. Future research should employ higher-resolution hydrodynamic routing to capture floodplain dynamics, explicitly represent vector niches and processes such as larval flushing, integrate evolving irrigation and water infrastructure, and bridge scales from continental projections to operational, landscape-level decision support for malaria control.

• Validation challenges: Malaria transmission depends on many non-climatic factors; quantitative validation of hydro-climatic suitability is inherently limited. Historical comparison (e.g., pre-intervention maps) involves categorical, coarse classifications.
• Climate data and biases: No bias correction was applied to climate projections to preserve variable coherence; runoff simulations show variable quality and positive biases in arid regions.
• Model scope: The analysis quantifies hydro-climatic suitability only and does not include vector/control dynamics, socioeconomic factors, or intervention impacts. Temperature suitability uses broad ranges; uncertainties in thermal response remain.
• Spatial/temporal resolution: LISFLOOD at 0.5° may miss sub-grid hydrological heterogeneity and small-scale habitat dynamics.
• Irrigation and water infrastructure: Irrigation was mapped for circa-2010 conditions and not projected forward, potentially underestimating future human-made water-body effects; reservoirs and lakes are included in LISFLOOD but detailed future changes in water management are not.
• Scenario coverage: Projections focus on RCP 8.5; other emissions pathways are not assessed here.
• Model agreement: Agreement across GCMs declines slightly over time; some regional predictions carry low signal-to-noise ratios.