Mapping Global Land Subsidence Using Remote Sensing and Machine Learning
Abstract
Generally, it is very difficult to quantify groundwater storage and its temporal loss in the absence of a dense global groundwater monitoring network. Due to different monitoring capabilities, strategies, and trans-boundary water conflicts, data scarcity makes it challenging to form such a global network. In spite of these challenges, it is critical to monitor groundwater storage loss in order to find sustainable groundwater management solutions. Interferometric Synthetic Aperture Radar (InSAR) is a promising technique for groundwater monitoring, and has been used to monitor land subsidence, which is linked to groundwater storage loss, at mm-to-cm scale accuracy. One challenge of using InSAR data to monitor groundwater systems is that processing the data and removing or accounting for tropospheric noise can be computationally intensive. In this study, land subsidence data from 20 existing studies over different regions of the world were used to train a random forests model to predict land subsidence globally. Predictors in the model included land use, evapotranspiration, relative surface water/groundwater use, water storage anomalies from GRACE, and soil type. The global-scale machine learning model predicts land subsidence at 2 km resolution, and can help identify regions with high groundwater storage loss. The model also finds relationships between global hydrologic, satellite and land subsidence datasets. The model has an accuracy of 0.93 on the validation dataset. The prediction is particularly helpful for data scarce regions with little or no ground-based hydrologic data to monitor groundwater. -Approved for public release, 21-775
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFMNS25B0425H