Predicting spatial tendency of groundwater head using a machine learning algorithm in the Tangshan plain

Código

I-EBHE0017

Autores

Nebiyou Kassahun

Tema

WG 1.12: Development & application of river basin simulators

Resumo

Abstract: In recent years, datadriven machine learning algorithms for groundwater studies have emerged as alternative and complementary approaches to conventional numerical groundwater modeling. In this research, we employed Random Forest algorithm to predict long-term groundwater table tendency of the Tangshan plain, an economically active city located in north China.Observed data of groundwater depth from 72 wells were temporally analyzed in the time window 2020to 2022. It was identified that groundwater head was the lowest in April in a year whereas the peak appeared in August. The spatial distributions of head at selected months were then used as model targets, while a digital elevation model, locations of rivers, proximity to coast, soil texture, and land cover were feature engineered and used to train the random forest model. Split sample test was used where70% of observed data was made for training and 30% for testing the model performance. Prediction accuracy was evaluated using Root Mean Squared Error, R2, Mean Absolute Error and Mean Error. The model showed acceptable prediction capabilities within the study area. Feature importance analysis was also conducted to increase model interoperability. It was found that proximity to the coast and land cover played dominant roles and therefore were identified as the top two important covariates influencing groundwater conditions in the Tangshan plain.