Spatio-temporal modelling of climate variability and malaria transmission across time scales to assess causality, improve forecasting and strengthen early warning
Project Abstract
Model-based malaria surveillance that incorporates climate effects is recognised as an adaptation strategy to address the impact of climate variability on malaria outbreaks. Climate is one of the drivers of transmission, but other factors such as control interventions and socio-economic development can influence disease dynamics.
In the first phase of the project, we developed statistical and mathematical models to quantify the contribution of climatic and non-climatic factors to malaria and to evaluate the added value of mathematical transmission models for outbreak forecasting compared to statistical models. Data from the Kisumu HDSS showed that temperature had a protective effect similar to that of bed nets, but this was offset by rainfall. These relations varied over seasonal scales. The high correlations between climatic factors and non-linear relations with malaria made it difficult to obtain consistent results from statistical models, which can estimate associations but cannot identify causal relations or account for non-linear interactions of malaria drivers. The non-stationarity of the malaria data and the varying impact of predictors across seasons suggested that forecasting models should perform better when the non-stationarity is relaxed.
The proposed project for the second phase builds on our findings, insights and identified research gaps. The overall goal of the follow-on project is to deepen our understanding of the impact of climate change on the malaria burden by developing innovative methods that take into account the non-linear interactions of malaria drivers on transmission dynamics and the non-stationarity of the data. The specific objectives are to: (i) assess the time-delayed causal effects of climatic and non-climatic factors on changes in malaria incidence at different time scales and transmission levels; (ii) develop age-structured stochastic metapopulation malaria transmission models that incorporate climate and control intervention effects; (iii) develop non-stationary model formulations for short- and medium-term malaria forecasting that take into account climate variation across time scales; and (iv) evaluate the performance of the new tools on common datasets and their computational feasibility for implementation within a model-based early warning system.
We will accomplish these specific objectives by (a) employing and further developing methods for assessing causality in time series data and coupling wavelet prediction with machine learning and dynamic time-delay embedding models; (b) using powerful computational algorithms such as iterative filtering, particle Markov chain Monte Carlo (pMCMC) and Hamiltonian Monte Carlo (HMC) simulation; and (c) analysing existing data sets; and (c) analysing existing data from the HDSS in Nouna (Burkina Faso) and Kisumu (Kenya), the health information system in Burkina Faso and Kenya, outputs from downscaled climate models, high-resolution hydrometeorological data, satellite-based climate products and other gridded environmental proxies.
