Forecasting changes of the flow regime at deep geothermal wells based on high resolution sensor data and low resolution chemical analyses

Abstract

Abstract. Geothermal waters provide a great resource to generate clean energy, however, there is a notorious lack of high quality data on these waters. The scarcity of deep geothermal aquifer information is largely due to inaccessibility and high analysis costs. However, multiple operators use geothermal wells in Lower Bavaria and Upper Austria for balneological (medical and wellness) applications as well as for heat mining purposes. The state of the art sampling strategy budgets for a sampling frequency of 1 year. Previous studies have shown that robust groundwater data requires sampling intervals of 1–3 months, however, these studies are based on shallow aquifers which are more likely to be influenced by seasonal changes in meteorological conditions. This study set out to assess whether yearly sampling adequately represents sub-yearly hydrochemical fluctuations in the aquifer by comparing yearly with quasi-continuous hydrochemical data at two wells in southeast Germany by assessing mean, trend and seasonality detection among the high and low temporal resolution data sets. Furthermore, the ability to produce reliable forecasts based on yearly data was examined. In order to test the applicability of virtual sensors to elevate the information content of yearly data, correlations between the individual parameters were assessed. The results of this study show that seasonal hydrochemical variations take place in deep aquifers, and are not adequately represented by yearly data points, as they are typically gathered at similar production states of the well and do not show varying states throughout the year. Forecasting on the basis of yearly data does not represent the data range of currently measured continuous data. The limited data availability did not allow for strong correlations to be determined. We found that annual measurements, if taken at regular intervals and roughly the same production rates, represent only a snapshot of the possible hydrochemical compositions. Neither mean values, trends nor seasonality was accurately captured by yearly data. This could lead to a violation of stability criteria for mineral water, or to problems in the geothermal operation (scalings, degassing). We thus recommend a new testing regime of at least 3 samples a year. While not a replacement for the detailed analyses, under the right circumstances, and when trained with more substantial data sets, viertual sensors provide a robust method in this setting to trigger further actions.