Predicting the Effects of Climate Change on Dengue Vector Densities in Southeast Asia through Process-Based Modeling

Background: Aedes aegypti and Ae. albopictus mosquitoes are major vectors for several human diseases of global importance, such as dengue and yellow fever. Their life cycles and hosted arboviruses are climate sensitive and thus expected to be impacted by climate change. Most studies investigating climate change impacts on Aedes at global or continental scales focused on their future global distribution changes, whereas a single study focused on its effects on Ae. aegypti densities regionally. Objectives: A process-based approach was used to model densities of Ae. aegypti and Ae. albopictus and their potential evolution with climate change using a panel of nine CMIP6 climate models and climate scenarios ranging from strong to low mitigation measures at the Southeast Asian scale and for the next 80 y. Methods: The process-based model described, through a system of ordinary differential equations, the variations of mosquito densities in 10 compartments, corresponding to 10 different stages of mosquito life cycle, in response to temperature and precipitation variations. Local field data were used to validate model outputs. Results: We show that both species densities will globally increase due to future temperature increases. In Southeast Asia by the end of the century, Ae. aegypti densities are expected to increase from 25% with climate mitigation measures to 46% without; Ae. albopictus densities are expected to increase from 13%–21%, respectively. However, we find spatially contrasted responses at the seasonal scales with a significant decrease in Ae. albopictus densities in lowlands during summer in the future. Discussion: These results contrast with previous results, which brings new insight on the future impacts of climate change on Aedes densities. Major sources of uncertainties, such as mosquito model parametrization and climate model uncertainties, were addressed to explore the limits of such modeling. https://doi.org/10.1289/EHP11068


Table of Contents
. Study area. The study area covered Southeast Asia and was restricted to the countries of Myanmar, Lao PDR, Thailand, Cambodia and Vietnam. Borders are displayed here in red dotted lines and the color scale represents elevation. Black lines represent the 600 m contour line.  1 In blue, the aquatic stages (E: eggs, L: larvae, P: pupae); in orange, the adult female stages (Aem: emerging, A1: nulliparous, A2: parous, with h: host-seeking, g: resting, o: ovipositing). Figure S3a. Species Distribution Models predictions. An Ensemble Modeling approach was used to produce spatial predictions of the probability of observation of both species. For each species, the mean of predictions from the 30 models weighted by their AUC score (provided the model's AUC was greater than a threshold = 0,85) were computed to produce the final output. Maps of predicted probability of observation were thus computed for Ae. aegypti and Ae. albopictus. The dataset used to build the models are displayed as green and grey circles, respectively corresponding to true occurrences and to randomly-generated pseudo-absences.       Percentage changes correspond to ratios of projected modeled density over contemporary modeled densities, averaged over the nine simulations associated with the nine climate models. Positive (red) and negative (blue) percentage changes correspond respectively to higher and lower predicted mosquito density compared to the contemporary situation. Purple grid cells correspond to percentage changes greater than a threshold of 130%, mostly corresponding to areas of null to very low contemporary adult female density. Diagonal hashing represent grid cells where percentage changes associated with the nine climate models are significantly different from 0 (Student's t-test, p-value < 0.05). Figure S10. Contemporary climate and future projections during summer and winter months. Panels A describe the spatial variations of contemporary temperature and precipitation in summer (April, May, June) and winter months (December, January, February). Panels B describe projections of temperature and precipitation averaged over the nine climate models in summer and winter months at the middle (2041-2060) and end of the century (2081-2100) according to SSP1 2.6 and SSP3 7.0. Diagonal hashing represent grid cells where anomalies associated with the nine climate models are significantly different from 0 (Student's t-test, p-value < 0.05). Figure S11. Evolution of modeled Aedes density anomalies. Densities of Ae. albopictus and Ae. aegypti were modeled from climate data averages from all climate models, along the SSP3 7.0 pathway and for four periods to the end of the century. Percentage changes were calculated as a ratio of projected densities over contemporary densities, averaged over the nine simulations associated with the nine climate models. Diagonal hashing represent grid cells where percentage changes associated with the nine climate models are significantly different from 0 (Student's ttest, p-value < 0.05).

Figure S12. Comparison of modeled Aedes density anomalies across Shared Socio Economic
Pathways. Densities of Ae. albopictus and Ae. aegypti were modeled over climate data averages from all climate models, for SSP1 2.6, SSP2 4.5, SSP3 7.0 and SSP5 8.5 and for the 2081-2100 period. Percentage changes were calculated as a ratio of projected densities over contemporary densities, averaged over the nine simulations associated with the nine climate models. Diagonal hashing represent grid cells where percentage changes associated with the nine climate models are significantly different from 0 (Student's t-test, p-value < 0.05). Figure S13. Spatial trends of the future evolution of Aedes albopictus density versus topography. Percentage changes correspond to ratios of projected modeled Ae. albopictus density in summer months (2081-2100 SSP3 7.0) over contemporary modeled densities, averaged over the nine simulations associated with the nine climate models. Positive (red) and negative (blue) percentage changes correspond respectively to higher and lower predicted mosquito density compared to the contemporary situation. Contour lines represent the 300 m and 600 m isopleths.      2019), were compared to those used in the present study. Although the two models differed in several aspects, notably on the number of life stages considered, comparing the functions describing transitions rates is still pertinent. Table S1. Species Distribution Models quality metrics. Figure S1. Study area. The study area covered Southeast Asia and was restricted to the countries of Myanmar, Lao PDR, Thailand, Cambodia and Vietnam. Borders are displayed here in red dotted lines and the color scale represents elevation. Black lines represent the 600 m contour line.

Figure S3a. Species Distribution Models predictions.
An Ensemble Modeling approach was used to produce spatial predictions of the probability of observation of both species. For each species, the mean of predictions from the 30 models weighted by their AUC score (provided the model's AUC was greater than a threshold = 0,85) were computed to produce the final output. Maps of predicted probability of observation were thus computed for Ae. aegypti and Ae. albopictus. The dataset used to build the models are displayed as green and grey circles, respectively corresponding to true occurrences and to randomly-generated pseudo-absences.       Panels describe projections of modeled Aedes adult female density at the end of the century (2081-2100) according to SSP1 2.6 and SSP3 7.0. Percentage changes correspond to ratios of projected modeled density over contemporary modeled densities, averaged over the nine simulations associated with the nine climate models. Positive (red) and negative (blue) percentage changes correspond respectively to higher and lower predicted mosquito density compared to the contemporary situation. Purple grid cells correspond to percentage changes greater than a threshold of 130%, mostly corresponding to areas of null to very low contemporary adult female density. Diagonal hashing represent grid cells where percentage changes associated with the nine climate models are significantly different from 0 (Student's t-test, p-value < 0.05). Figure S10. Contemporary climate and future projections during summer and winter months. Panels A describe the spatial variations of contemporary temperature and precipitation in summer (April, May, June) and winter months (December, January, February). Panels B describe projections of temperature and precipitation averaged over the nine climate models in summer and winter months at the middle (2041-2060) and end of the century (2081-2100) according to SSP1 2.6 and SSP3 7.0. Diagonal hashing represent grid cells where anomalies associated with the nine climate models are significantly different from 0 (Student's t-test, p-value < 0.05).
A B Figure S11. Evolution of modeled Aedes density anomalies. Densities of Ae. albopictus and Ae. aegypti were modeled from climate data averages from all climate models, along the SSP3 7.0 pathway and for four periods to the end of the century. Percentage changes were calculated as a ratio of projected densities over contemporary densities, averaged over the nine simulations associated with the nine climate models. Diagonal hashing represent grid cells where percentage changes associated with the nine climate models are significantly different from 0 (Student's t-test, p-value < 0.05).
Figure S12. Comparison of modeled Aedes density anomalies across Shared Socio Economic Pathways. Densities of Ae. albopictus and Ae. aegypti were modeled over climate data averages from all climate models, for SSP1 2.6, SSP2 4.5, SSP3 7.0 and SSP5 8.5 and for the 2081-2100 period. Percentage changes were calculated as a ratio of projected densities over contemporary densities, averaged over the nine simulations associated with the nine climate models. Diagonal hashing represent grid cells where percentage changes associated with the nine climate models are significantly different from 0 (Student's t-test, p-value < 0.05).
Figure S13. Spatial trends of the future evolution of Aedes albopictus density versus topography. Percentage changes correspond to ratios of projected modeled Ae. albopictus density in summer months (2081-2100 SSP3 7.0) over contemporary modeled densities, averaged over the nine simulations associated with the nine climate models. Positive (red) and negative (blue) percentage changes correspond respectively to higher and lower predicted mosquito density compared to the contemporary situation. Contour lines represent the 300 m and 600 m isopleths. Figure S14. Evolution of projected climate across the 21 st century following four Shared Socio-Economic Pathways scenarios. Grid cells were split in two groups considering their elevation (Highlands: elevation < 600 m, Lowlands: elevation > 600 m). For each variable (temperature and precipitation), scenario, period and month, points and bars represent respectively mean and standard deviation of percentage change values associated to all climate models and all grid cells. Figure S15. Effects of climate mitigation measures on Aedes densities. Process-based model simulations were performed over various constant values of temperature and precipitations. Model outputs were extracted after 30 days of simulation and plotted against corresponding temperature and precipitations (Panels A and C). Changes in temperature and precipitation associated with a transition from SSP3 7.0 to SSP1 2.6 at specific locations and season at the end of the century (2081-2100) are represented with white dots and arrows. Changes in Ae. aegypti and Ae. albopictus adult densities corresponding to a transition from SSP3 7.0 to SSP1 2.6 at the end of the century are mapped respectively in panels B and D. Figure S16. Lengthening/shortening of the Aedes high density season. For both species, the median value of daily contemporary density was extracted for each grid cell, thus corresponding to the value at which contemporary densities are greater for 183 days of the year (6 months). For each grid cell, considering a 6-month long season of high contemporary adult density, the length of the season at the 2081-2100 horizon was considered as the number of days projected densities (SSP3 7.0) were greater than the contemporary median value. The lengthening/shortening of the season was assessed as the number of days longer/shorter than the 183 days of the contemporary season. Here are displayed in red and blue the areas where the model respectively predicts a lengthening and a shortening of Ae. aegypti's and Ae. albopictus's season. The values of season lengthening and shortening for each grid cell are displayed for both species as a boxplot (right panel). Boxes extend from quartiles Q1 to Q3, whiskers extend to 1.5 x (Q3 -Q1) from edges of the box. Overall median values are displayed as green bars and correspond to an overall lengthening of 10 days for Ae. aegypti and an overall shortening of 4 days for Ae. albopictus. Figure S17. Process-based model's underlying transition and mortality functions as a function of temperature. Panels describe mortality and transition rates between life stages as a function of temperature, as used by the process-based model to simulate population dynamics of Ae. albopictus and Ae. aegypti. Figure S18. Comparison of temperature-dependent transition rates used in process-based models. Temperaturedependent transition rates used in the model developed by Liu-Helmersson et al. (2019), 3 were compared to those used in the present study. Although the two models differed in several aspects, notably on the number of life stages considered, comparing the functions describing transitions rates is still pertinent. For each species, the Ensemble modeling process was performed over 6 model families, the 5-fold cross-validation process thus resulted in building 30 models. For Ae. aegypti, the 30 models displayed an average AUC of 0.87 (+-0.03 SD). Ensemble forecasting was performed using the 26 models presenting an AUC>0.85 (colored in black in the following table). For Ae.