An improved groundwater vulnerability evaluation model based on random forest algorithm and spatio-temporal change prediction method

Groundwater resource management and protection depend heavily on the assessment of groundwater vulnerability. This study with emphasis on the shallow groundwater in the piedmont alluvial fan region of northern Henan Province, China, aiming to improve the accuracy and applicability of existing groundwater vulnerability evaluation methods. By incorporating additional indicators such as land use types and groundwater quality, the traditional DRASTIC model's indicator system was refined. The random forest algorithm was introduced to determine the weights of these indicators, leading to the establishment of a novel groundwater vulnerability evaluation model. The spatiotemporal patterns of groundwater vulnerability were then predicted by combining this model with a three-dimensional groundwater flow model and examining fluctuations in groundwater level. Evaluation statistics of the research area revealed that the high-risk area of groundwater contamination covered 141.41 km2, 39.78 % of the entire area. Over next 1 and 5 years, the low-risk area of groundwater contamination expanded by 0.34 % and 0.68 %, respectively, due to increased groundwater levels in the southern region. The primary causes of groundwater contamination in the study area were identified as high proportion of plowland and severe pesticide pollution. The shallow groundwater depth, gentle topographic slope, and high permeability coefficient of the vadose zone facilitated the entry of pollutants into the groundwater system. These research findings provide valuable insights for improving groundwater vulnerability evaluation methods.