Use of GIS, Statistics and Machine Learning for Groundwater Quality Management: Application to Nitrate Contamination

Groundwater NO3 contamination (GNC) threatens the drinkability of water in many countries worldwide. It could cause serious health problems and sometimes lead to death. This paper aims to introduce a comprehensive approach that combines GIS, statistics and Machine Learning (ML) for the groundwater quality management including both water quality assessment and prediction. The performances of this approach are discussed through its application on assessing and predicting nitrate (NO3) concentrations in the Eocene Aquifer, Palestine. Spatiotemporal records of NO3 over the period 1982-2019 are integrated in a database and used in this research. The database includes the following factors: well depth, well use, anthropogenic on-ground activities, watersheds, soil type and land use. Geo-statistical assessment using GIS and statistical boxplots is employed to assess the variability of NO3 concentrations and how they are affected by the independent indicators. Assessment outcomes (NO3 distribution and the influencing factors) were used to build the Random Forest (RF) prediction model. Such model is used to predict GNC level in groundwater based on multi-influencing factors. Assessment results indicate increasing and decreasing trends of GNC in the southern and middle parts of the study area, respectively. It also provides the RF model by the main influencing factors affecting GNC in the study area which are: well depth, well use, anthropogenic on-ground activities, watersheds and land use. Results indicate that RF has an average and maximum prediction accuracy of 88.5 and 91.7%, respectively. The well depth has the highest influence on GNC. This research could support water authority decision-makers toward the adoption of sustainable groundwater protection plans in Palestine.

Automated summary

This paper introduces a comprehensive approach combining GIS, statistics and Machine Learning (ML) for groundwater quality management, including assessment and prediction of NO3 concentrations. The study in Palestine shows increasing and decreasing trends of GNC in different regions. The RF model, using factors such as well depth, well use, anthropogenic on-ground activities, watersheds, and land use, predicts GNC with an average accuracy of 88.5% and a maximum accuracy of 91.7%. This research can guide decision-makers in the adoption of sustainable groundwater protection plans in Palestine.