In the field of landscape epidemiology, the contribution of machine learning (ML) to modeling of epidemiological risk scenarios presents itself as a good alternative. This study aims to break with the ”black box” paradigm that underlies the application of automatic learning techniques by using SHAP to determine the contribution of each variable in ML models applied to geospatial health, using the prevalence of hookworms, intestinal parasites, in Ethiopia, where they are widely distributed; the country bears the third-highest burden of hookworm in Sub-Saharan Africa. XGBoost software was used, a very popular ML model, to fit and analyze the data. The Python SHAP library was used to understand the importance in the trained model, of the variables for predictions. The description of the contribution of these variables on a particular prediction was obtained, using different types of plot methods. The results show that the ML models are superior to the classical statistical models; not only demonstrating similar results but also explaining, by using the SHAP package, the influence and interactions between the variables in the generated models. This analysis provides information to help understand the epidemiological problem presented and provides a tool for similar studies.

Since 2009, Fundación Mundo Sano has implemented an Aedes aegypti Surveillance and Control Program in Tartagal city (Salta Province, Argentina). The purpose of this study was to analyze temporal dynamics of Ae. aegypti breeding sites spatial distribution, during five years of samplings, and the effect of control actions over vector population dynamics. Seasonal entomological (larval) samplings were conducted in 17,815 fixed sites in Tartagal urban area between 2009 and 2014. Based on information of breeding sites abundance, from satellite remote sensing data (RS), and by the use of Geographic Information Systems (GIS), spatial analysis (hotspots and cluster analysis) and predictive model (MaxEnt) were performed. Spatial analysis showed a distribution pattern with the highest breeding densities registered in city outskirts. The model indicated that 75% of Ae. aegypti distribution is explained by 3 variables: bare soil coverage percentage (44.9%), urbanization coverage percentage(13.5%) and water distribution (11.6%). This results have called attention to the way entomological field data and information from geospatial origin (RS/GIS) are used to infer scenarios which could then be applied in epidemiological surveillance programs and in the determination of dengue control strategies. Predictive maps development constructed with Ae. aegypti systematic spatiotemporal data, in Tartagal city, would allow public health workers to identify and target high-risk areas with appropriate and timely control measures. These tools could help decision-makers to improve health system responses and preventive measures related to vector control. As reported in Porcasi et al., in Argentina we are working on an integrated risk stratification system based in geospatial technologies that have moderately consolidated national scale, but need more understanding of its urban scale mechanisms. In this work, relevant results are shown on how Ae. aegypti breeding sites are distributed in dynamic spatial patterns in a small city on northern Argentina. 5 years of entomologic data were obtained by Mundo Sano Foundation, which is implementing an Aedes aegypti Surveillance and Control Program in Tartagal City (Salta Province, Argentina). The focus of this contribution is based on the difference that can found between one year data typical analysis and long term temporal evolution of spatial patterns. Although environmental sanitation activities and chemical control of breeding sites with larvicide were performed after each entomological surveillance all around Tartagal, sectors with higher densities of breeding sites remained present throughout study period. Nonetheless, the distribution of breeding sites showed a spatial dynamic with high density clusters in city outskirts.

Moisture content of live fuels (LFMC) is one of the main factors determining fuel flammability and, therefore, a key indicator of fire danger. In this study, we modeled the relationship between spectral indices derived from satellite imagery and field estimations of LFMC in the Chaco Serrano subregion; then, we analyzed the relationship between fire danger estimations based on LFMC calculations and fire activity. Empirical LFMC models fitted for grasslands, Chaco Serrano forests, and glossy privet forests may be considered very accurate R ² > 0.80, whereas the model corresponding to shrublands still needs to be improved (R ² = 0.57). Monthly maps of fire danger reflected the occurrence of fires consistently during years of both high and low fire activity. Most fires occurred mainly in areas with high or extreme fire danger, demonstrating a clear relationship between LFMC and fire activity in the Chaco Serrano subregion. Our LFMC models may be useful to assess the spatiotemporal distribution of fire danger in the Chaco Serrano subregion using remote sensing data. The associated fire danger maps represent a valuable tool for improving decision making processes to organize early warning and fire suppression activities.

We use a Species Distribution Modeling (SDM) approach along with Geographic Information Systems (GIS) techniques to examine the potential distribution of hantavirus pulmonary syndrome (HPS) caused by Andes virus (ANDV) in southern Argentina and, more precisely, define and estimate the area with the highest infection probability for humans, through the combination with the distribution map for the competent rodent host (Oligoryzomys longicaudatus). Sites with confirmed cases of HPS in the period 1995–2009 were mostly concentrated in a narrow strip (~90 km × 900 km) along the Andes range from northern Neuquén to central Chubut province. This area is characterized by high mean annual precipitation (~1,000 mm on average), but dry summers (less than 100 mm), very low percentages of bare soil (~10% on average) and low temperatures in the coldest month (minimum average temperature −1.5 °C), as compared to the HPS-free areas, features that coincide with sub-Antarctic forests and shrublands (especially those dominated by the invasive plant Rosa rubiginosa), where rodent host abundances and ANDV prevalences are known to be the highest. Through the combination of predictive distribution maps of the reservoir host and disease cases, we found that the area with the highest probability for HPS to occur overlaps only 28% with the most suitable habitat for O. longicaudatus. With this approach, we made a step forward in the understanding of the risk factors that need to be considered in the forecasting and mapping of risk at the regional/national scale. We propose the implementation and use of thematic maps, such as the one built here, as a basic tool allowing public health authorities to focus surveillance efforts and normally scarce resources for prevention and control actions in vast areas like southern Argentina.

Soil-Transmitted Helminths (STH) are highly prevalent Neglected Tropical Disease in Ethiopia, an estimated 26 million are infected. Geographic Information Systems and Remote Sensing (RS) technologies assist data mapping and analysis, and the prediction of the spatial distribution of infection in relation to environmental variables. The influence of socioeconomic, environmental and soil characteristics on hookworm infection at the individual and household level is explored in order to identify spatial patterns of infection in rural villages from Zenzelema (Amhara region). Inhabitants greater than 5 years old were recruited in order to assess the presence of STH. Socioeconomic and hookworm infection variables at the household level and environmental variables and soil characteristics using RS were obtained. The dominant STH found was hookworm. Individuals which practiced open defecation and those without electricity had a significant higher number of hookworm eggs in their stool. Additionally, adults showed statistically higher hookworm egg counts than children. Nonetheless, the probability of hookworm infection was not determined by socioeconomic conditions but by environmental characteristics surrounding the households, including a combination of vigorous vegetation and bare soil, high temperatures, and compacted soils (high bulk density) with more acidic pH, given a pH of 6.0 is optimal for hatching of hookworm eggs. The identification of high-risk environmental areas provides a useful tool for planning, targeting and monitoring of control measures, including not only children but also adults when hookworm is concerned.