Monitoring groundwater nitrate concentrations using geodetector-based geofactors and neural network approach in the Wanka Watershed, Ethiopia

Achieving Sustainable Development Goal 6 (Clean Water and Sanitation) and Goal 3 (Good Health and Well-Being) is reliant on having access to safe drinking water. Groundwater is the main sources of drinking water in rural and urban areas. However, water quality is still poorly monitored in Ethiopia''s rural areas, including the Wanka Watershed. Improper land use and agricultural growth, nitrate contamination is a growing concern for groundwater, endangering human health and water sustainability. High nitrate concentration (NO 3 -N) pose serious health risks such as blue baby syndrome, cancers, and neural tube defects. However, routine monitoring is limited because laboratory analysis is expensive, logistically difficult, and often inaccessible in remote communities. Hence, the primary objective of this research is to develop suitable statistical methods and apply appropriate machine learning algorithms to estimate the current status of nitrate concentration in data-limited area (Wanka watershed), upper Blue Nile basin, Ethiopia. Therefore, geodetector-based geofactors and neural network methods have been applied to predict the nitrate level in the groundwater. Forty-one groundwater nitrate concentration samples were used as the training dataset in 2013, whereas 12 nitrate concentration samples were collected from the same well locations in 2022 for validation. Nitrate influencing factors were quantified as factor detector output such as recharge (0.210), static water level (0.015), transmissivity (0.119), lithology (0.079), mean air temperature (0.021), Topographic Wetness Index (0.078), soil (0.115), and land use (0.232) as input variables. The results revealed that land use, recharge, transmissivity, and soil have a strong power of influence on changing the nitrate concentration of shallow groundwater in agriculturally intensified areas, compared with lithology, static water level, and temperature have a weak power of influence. Geofactors and neural network (NN) were acceptable with respect to metrics such as, RMSE (0.39, 0.53), MAE (0.26, 0.33), RSSE (1.83, 3.37), and R 2 (0.995, 0.99), respectively. Comparatively, geofactors has a better performance for nitrate monitoring in this study area. The predicted results of NO3-N (in mg/L) have been mapped for monitoring and management in the agriculturally intensified areas. It has a better implication for policymakers regarding to water resources protection.