On groundwater monitoring using machine learning and satellite remote sensing
Schelter, Lennart Noel; Schüttrumpf, Holger (Thesis advisor); Amann, Florian (Thesis advisor)
Aachen : RWTH Aachen University (2021, 2022)
Dissertation / PhD Thesis
Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2021
Groundwater is by far the largest component of the active hydrological cycle. It is also the main source of fresh water in many regions of the world and is only becoming even more important for agriculture, industry and domestic use in the light of climate change and more extreme weather events. An effective and sustainable management of this important resource is therefore necessary. Unfortunately, conventional monitoring of groundwater resources requires expensive networks of monitoring wells, which are not available in many regions of the world, making effective management problematic. The aim of this thesis is to quantify how precisely groundwater can be monitored using only remote sensing data and machine learning models. For this reason, three different problems are investigated: increasing the spatial resolution of available groundwater level data, calculating groundwater level changes in a time period before groundwater level data is available and calculating groundwater levels in a region with no available groundwater level data. Each task is approached using four different model types, a "Multivariate Linear Regression"(MVLR), a "Random Forest" (RF), a "Multilayer Perceptron" (MLP) and a "Long Short-Term Memory" (LSTM) model, applied in two different study areas in Germany and the US. The input data consists of GRACE total water storage data and various meteorological and hydrological parameters provided by different satellite missions. The results show that good correlation can be achieved by multiple models for the increase of spatial resolution with a coefficient of determination (R2) of 0.76 for the best MLP model. Temporal and spatial extrapolation both require further optimization to achieve similar precision. The presented approach is open to anyone using only publicly available data and some local groundwater measurements in open-source algorithms. It can effectively increase spatial resolution of available groundwater observations to a 0.05 degree, monthly resolution.The application in regions with sparse groundwater monitoring data could significantly increase the quality of information available to water authorities and consequently improve an effective and sustainable water resources management everywhere.