Groundwater Depth prediction
Water is the most critical resource for life. In Italy, 72% of available water is from surface water and 28% from groundwater. About 70% of the underground resources are in northern Italy, while groundwater in southern Italy is confined in the short stretches of coastal plains and in a few inner areas. Petrignano Aquifer is located in Petrignano, southern Italy. The aquifer can be considered a water table of groundwater and is fed by the Chiascio River. Prediction of changes in aquifer water levels can be crucial for water management for the area that is on water shortage. Rainfall, seasonal variations of the river, underground water level, and drainage can influence the amount of water in an aquifer. I am curious about which feature contributes the most to water amount changes in this aquifer and what is the future depth to groundwater of a well belonging to this aquifer over the next quarter.
2009-2020 rainfall, temperature, river hydrometry, and drainage volume are used. Drainage volume has the highest correlation coefficient with the depth to groundwater. River hydrometry is the second most correlated to changes in depth to groundwater. When more water is taken from the drinking water treatment plant, the groundwater level is low. Rainfall and temperature are not highly correlated with the depth of groundwater. The correlation of features increases after shifting 3 months. This suggests there is a lag between changes in those features and changes in groundwater depth. The temperature is negatively correlated with groundwater depth. When a day is hot, the groundwater level is low since water demand may increase.
I used prophet, XGBOOST Regressor (Extreme Gradient Boosting) and LSTM (long short-term memory) models to do multivariate time series prediction. Prophet is an open-source library for time series forecasting developed by Facebook. It works best with time series that have strong seasonal effects and several seasons of historical data. XGBoost is an open-source library that provides an efficient and effective implementation of gradient boosted decision trees designed for speed and performance. LSTM is a special kind of recurrent neural network and is capable of learning long-term dependencies.
Out of the three models, LSMT performs the best. Drainage volume and river hydrometry contribute the most to the changes in groundwater depth. River hydrometry and drainage volume are two crucial factors for changes in groundwater depth.
