NASA JPL: Subsidence Modeling

About the Project

Majority of the water that is consumed by billions of people comes from groundwater. Due to excessive pumping of groundwater, land is sinking at a rapid rate (20 cm per year). Displacement is principally downward. California central valley is responsible for 17 billion of annual agricultural output producing 25 percent of nations food.Central valley’s land is at a risk of sinking due to pumping of groundwater It is very important to understand groundwater depletion is the key factor for the permanent subsidence in Central Valley. Some areas are increasingly subsiding even with new precipitation. These areas cannot store new groundwater in that region after subsidence. By understanding subsidence patterns, we can predict these areas and prevent permanent subsidence. We have real-world datasets collected from NASA from 2015 to 2019.This dataset is integrated multimodal data that include INSAR, groundwater, rainfall, precipitation, and soil composition by interpolating data with same spatial and temporal resolution. Our goal is predicting subsidence using machine learning models and multimodal data. We choose to explore and compare various forecasting models including RNNs, GRUs,LSTMs, Random Forest, and Arima. Along with this, the key factors such as groundwater and precipitation is implemented into our models to improve predictions. By exploring these models and using various predictors, we identify the best models and performance for this problem.