<![CDATA[To understand the environmental distribution of Burkholderia pseudomallei, it is essential to study the causative agent of melioidosis for effective public health risk assessment. This study integrates geostatistical analysis and machine learning to predict the spatial distribution of Burkholderia pseudomallei in paddy soils of northeastern Thailand. A total of 92 soil samples were collected and analysed using culture-based methods. Environmental covariates were derived from remote sensing and topographic data, including land surface temperature, normalised difference salinity index, bare soil index, digital elevation model, distance to water, slope, aspect, and soil drainage. Indicator kriging was used to generate a spatial probability map of Burkholderia pseudomallei presence. An extreme gradient boosting machine learning model was applied to predict bacterial presence. Of the 92 soil samples analysed, 40.22% tested positive for Burkholderia pseudomallei. Indicator kriging demonstrated clustered distributions primarily in low-lying, poorly drained areas. The extreme gradient boosting model achieved an F1-score of 0.70 on the testing dataset. Shapley additive explanations analysis highlighted the digital elevation model, bare soil index, and slope as the most influential predictors. The resulting risk maps provide valuable tools for identifying high-risk areas, supporting targeted surveillance and public health interventions in melioidosis-endemic regions.]]>
Copyrights © 2026
