No abstract

No abstract

No abstract

<p>Salinization is one of the main threats to groundwater quality around the world, particularly in arid and semi-arid regions (IPCC, 2007). Some of the major causes of high salinity include natural geological conditions, seawater intrusion, climate change affecting patterns of precipitation and evaporation, overexploitation of groundwater and poor irrigation practices (Amer and Vengosh, 2001; Russ et al., 2020). Salinity can reduce the availability of water for humans and wildlife and can negatively impact crop productivity and promote desertification. Desert regions in Somalia, Ethiopia and Kenya have natural characteristics that favour high salinity in groundwater. 80% of the population in the region depends on groundwater (UNICEF, 2020), and 69% of groundwater sources have salinity levels above the WHO health-based drinking water guideline of 1500 µS/cm.</p><p>Here, we use machine learning to spatially predict patterns of high salinity with a dataset of 6300 groundwater quality measurements and various environmental predictors. More than 60 predictor variables were tested and 100 iterations of the random forest were performed. Most of the salinity data were clustered, which can lead to sampling issues due to spatial autocorrelation (SAC). As traditional non-spatial validation methods ignore SAC in the data and therefore do not guarantee independence between training and testing data, we instead use spatial cross-validation to address this spatial phenomenon as well as variograms to identify the extent of autocorrelation among variables. Preliminary results indicate that fractured ancient marine deposits, recharge, precipitation, evaporation and proximity to the ocean are the main factors related to high salinity levels. The model performs well with a combined overall accuracy of ~80% and an Area Under the Curve (AUC) of 0.80. Predictive spatial maps of groundwater salinity will be presented along with an analysis of the drivers of salinity.</p><p><img src="https://contentmanager.copernicus.org/fileStorageProxy.php?f=gepj.f5764ba1e0c165360300461/sdaolpUECMynit/22UGE&app=m&a=0&c=a657542f78ae3ce12aea2c3ee686e9a3&ct=x&pn=gepj.elif&d=1" alt="" width="568" height="587"></p><p>Figure 1. Topographic map of the study area and salinity concentration represented by electrical conductivity (EC).</p>

No abstract