Abstract: The dynamic TSI dataset presented here provides a new product with far richer spatial and temporal information than any other presently available for Africa. As spatiotemporal malaria modeling endeavors evolve, dynamic predictor variables such as the malaria temperature suitability data developed here will be essential for the rational assessment of changing patterns of malaria risk.
Introduction
Malaria transmission is a biological process fundamentally dictated by temperature. Ambient heat affects both the longevity of mosquito vectors and the rate at which the Plasmodium falciparum parasite develops within them. Traditionally, malaria risk has been mapped using "synoptic" models based on long-term monthly averages. While useful for identifying general endemic zones, these static maps fail to capture inter-annual variability.
In reality, anomalous weather patterns—such as an unusually warm spring or a cool autumn—can drastically shift transmission windows. These fluctuations are critical for public health, as they often trigger outbreaks in populations with low acquired immunity. This research addresses the need for a dynamic, high-resolution spatiotemporal dataset reflecting actual temperature conditions across Africa over a 13-year period.
Methodology
The researchers developed a multi-stage pipeline to transform raw satellite data into a biological Temperature Suitability Index (TSI):
- LST Processing: Land Surface Temperature (LST) data from MODIS satellites (MOD11A2) served as the primary input. To address cloud cover gaps, a "nested space-time interpolation" algorithm was employed to maintain spatial detail.
- LST to Air Temperature Conversion: Because mosquitoes inhabit the air rather than the soil surface, LST was converted to ambient air temperature ($T_{min}$ and $T_{max}$).
The $T_{max}$ model was further enhanced by incorporating diurnal temperature ranges and day length to improve predictive power beyond raw LST correlations.
- Temporal Upsampling: Satellite composites (8-day) were interpolated into 2-hour slices using sine-wave and exponential-decay models to simulate the diurnal cycle.
- Biological Simulation: A mechanistic model simulated "mosquito cohorts" emerging every two hours. The model tracked each cohort’s survival and the parasite's development (sporogony) based on the specific temperature trajectory experienced by that cohort.
Key Results and Validation
The study produced 154 monthly raster surfaces at 1km resolution covering April 2000 to December 2012. The gap-filling procedure proved highly robust ($R^2 \approx 0.97$).
Validation against 407,857 daily observations from 154 meteorological stations confirmed the accuracy of the final air temperature models, with an $R^2$ of 0.82 for $T_{min}$ and 0.80 for $T_{max}$. The resulting TSI revealed massive inter-annual swings in transitional zones like the Sahel and southern Africa, where the standard deviation of suitability often exceeded 0.4.
Discussion: The Importance of Dynamic Modeling
A primary finding is the "muting" effect of long-term averages. Synoptic models tend to lower the peaks of high-suitability years and raise the troughs of low-suitability years, potentially underestimating risk during epidemic periods. By tracking cohorts over time, the model accounts for the parasite's incubation period; a temperature spike is only relevant if the vector survives long enough for sporogony to complete.
Furthermore, the 1km resolution is vital for "montane" regions. In highland areas, malaria risk can vary significantly between a valley floor and a nearby ridge—a nuance lost in coarser global models.
Limitations and Future Directions
While a significant advancement, the model currently focuses on adult mosquitoes and the parasite, omitting the temperature effects on larval development. Additionally, temperature is only one component of transmission; future iterations should integrate dynamic rainfall and humidity data. While currently tuned for P. falciparum, the framework is adaptable for other pathogens like P. vivax or other vector-borne diseases.
Summary updated