Spatial Optimization Methods for Malaria Risk Mapping in Sub-Saharan African Cities Using Demographic and Health Surveys.

Vector-borne diseases, such as malaria, are affected by the rapid urban growth and climate change in sub-Saharan Africa (SSA). In this context, intra-urban malaria risk maps act as a key decision-making tool for targeting malaria control interventions, especially in resource-limited settings. The Demographic and Health Surveys (DHS) provide a consistent malaria data source for mapping malaria risk at the national scale, but their use is limited at the intra-urban scale because survey cluster coordinates are randomly displaced for ethical reasons. In this research, we focus on predicting intra-urban malaria risk in SSA cities-Dakar, Dar es Salaam, Kampala and Ouagadougou-and investigate the use of spatial optimization methods to overcome the effect of DHS spatial displacement. We modeled malaria risk using a random forest regressor and remotely sensed covariates depicting the urban climate, the land cover and the land use, and we tested several spatial optimization approaches. The use of spatial optimization mitigated the effects of DHS spatial displacement on predictive performance. However, this comes at a higher computational cost, and the percentage of variance explained in our models remained low (around 30%-40%), which suggests that these methods cannot entirely overcome the limited quality of epidemiological data. Building on our results, we highlight potential adaptations to the DHS sampling strategy that would make them more reliable for predicting malaria risk at the intra-urban scale.

Poultry farm distribution models developed along a gradient of intensification.

Efficient planning of measures limiting epidemic spread requires information on farm locations and sizes (number of animals per farm). However, such data are rarely available. The intensification process which is operating in most low- and middle-income countries (LMICs), comes together with a spatial clustering of farms, a characteristic epidemiological models are sensitive to. We developed farm distribution models predicting both the location and the number of animals per farm, while accounting for the spatial clustering of farms in data-poor countries, using poultry production as an example. We selected four countries, Nigeria, Thailand, Argentina and Belgium, along a gradient of intensification expressed by the per capita Gross Domestic Product (GDP). First, we investigated the distribution of chicken farms along the spectrum of intensification. Second, we built farm distribution models (FDM) based on censuses of commercial farms of each of the four countries, using point pattern and random forest models. As an external validation, we predicted farm locations and sizes in Bangladesh. The number of chicken per farm increased gradually in line with the gradient of GDP per capita in the following order: Nigeria, Thailand, Argentina and Belgium. Interestingly, we did not find such a gradient for farm clustering. Our modelling procedure could only partly reproduce the observed datasets in each of the four sample countries in internal validation. However, in the external validation, the clustering of farms could not be reproduced and the spatial predictors poorly explained the number and location of farms and farm sizes in Bangladesh. Further improvements of the methodology should explore other covariates of the intensity of farms and farm sizes, as well as improvements of the methodology. Structural transformation, economic development and environmental conditions are essential characteristics to consider for an extrapolation of our FDM procedure, as generalisation appeared challenging. We believe the FDM procedure could ultimately be used as a predictive tool in data-poor countries.

Downscaling livestock census data using multivariate predictive models: Sensitivity to modifiable areal unit problem.

The analysis of census data aggregated by administrative units introduces a statistical bias known as the modifiable areal unit problem (MAUP). Previous researches have mostly assessed the effect of MAUP on upscaling models. The present study contributes to clarify the effects of MAUP on the downscaling methodologies, highlighting how a priori choices of scales and shapes could influence the results. We aggregated chicken and duck fine-resolution census in Thailand, using three administrative census levels in regular and irregular shapes. We then disaggregated the data within the Gridded Livestock of the World analytical framework, sampling predictors in two different ways. A sensitivity analysis on Pearson's r correlation statistics and RMSE was carried out to understand how size and shapes of the response variables affect the goodness-of-fit and downscaling performances. We showed that scale, rather than shapes and sampling methods, affected downscaling precision, suggesting that training the model using the finest administrative level available is preferable. Moreover, datasets showing non-homogeneous distribution but instead spatial clustering seemed less affected by MAUP, yielding higher Pearson's r values and lower RMSE compared to a more spatially homogenous dataset. Implementing aggregation sensitivity analysis in spatial studies could help to interpret complex results and disseminate robust products.

Global mapping of highly pathogenic avian influenza H5N1 and H5Nx clade 2.3.4.4 viruses with spatial cross-validation.

Global disease suitability models are essential tools to inform surveillance systems and enable early detection. We present the first global suitability model of highly pathogenic avian influenza (HPAI) H5N1 and demonstrate that reliable predictions can be obtained at global scale. Best predictions are obtained using spatial predictor variables describing host distributions, rather than land use or eco-climatic spatial predictor variables, with a strong association with domestic duck and extensively raised chicken densities. Our results also support a more systematic use of spatial cross-validation in large-scale disease suitability modelling compared to standard random cross-validation that can lead to unreliable measure of extrapolation accuracy. A global suitability model of the H5 clade 2.3.4.4 viruses, a group of viruses that recently spread extensively in Asia and the US, shows in comparison a lower spatial extrapolation capacity than the HPAI H5N1 models, with a stronger association with intensively raised chicken densities and anthropogenic factors.

Clade-level Spatial Modelling of HPAI H5N1 Dynamics in the Mekong Region Reveals New Patterns and Associations with Agro-Ecological Factors.

The highly pathogenic avian influenza (HPAI) H5N1 virus has been circulating in Asia since 2003 and diversified into several genetic lineages, or clades. Although the spatial distribution of its outbreaks was extensively studied, differences in clades were never previously taken into account. We developed models to quantify associations over time and space between different HPAI H5N1 viruses from clade 1, 2.3.4 and 2.3.2 and agro-ecological factors. We found that the distribution of clades in the Mekong region from 2004 to 2013 was strongly regionalised, defining specific epidemiological zones, or epizones. Clade 1 became entrenched in the Mekong Delta and was not supplanted by newer clades, in association with a relatively higher presence of domestic ducks. In contrast, two new clades were introduced (2.3.4 and 2.3.2) in northern Viet Nam and were associated with higher chicken density and more intensive chicken production systems. We suggest that differences in poultry production systems in these different epizones may explain these associations, along with differences in introduction pressure from neighbouring countries. The different distribution patterns found at the clade level would not be otherwise apparent through analysis treating all outbreaks equally, which requires improved linking of disease outbreak records and genetic sequence data.