Optimization of Data from Local Groundwater Head Monitoring Network Using Principal Component Analysis

Ascending near-surface groundwater is an important source of water supply to crops and grasslands. Tapping this sub-surface water source, they exhibit a more stable productivity over time as compared to groundwater-distant sites. Assessing crop and grass productivity at high resolution with the help of simulation models therefore requires groundwater table distance information in an apposite spatial and temporal resolution. Groundwater level monitoring networks offer point-based data with variable observation frequency and data quality, impairing assessments of local variations in the hydrographs at each observation well. We propose an efficient and structured process for applying principal component analysis (PCA) in optimizing the groundwater level monitoring network. The PCA functions were used to determine the relative contributions of individual observation wells in determining the spatio-temporal variations in hydrographs. For each well, the Principal Components (PCs) derived from the PCA were used as predicted variables to draw reference hydrographs that describe the expected normal behavior of individual observation wells. This reference hydrograph was then compared with the observed hydrograph so that the residuals describe the local deviations from the normal behavior of the observation well. Deriving a time series of the residuals facilitates a rapid screening for idiosyncrasies unique to each well. Based on a ranking of all wells in the network according to their degree of deviation from the reference, we discarded irrelevant monitoring wells and time series. In a case study using 1300 observation wells in Brandenburg State, Germany, with mean monthly data from 2000 to 2022, we showed in preliminary results that the overall difference in groundwater level between the original observation well network and the optimized network developed with PCs is less than 5%, while the total number of observation wells in the network is reduced by 10%, which will save the time and cost to monitor groundwater levels in the area.