Recharge estimation

The estimation of groundwater recharge can be improved by using different methods and comparing the results. This also increases the reliability of recommendations made for the management of groundwater resources.

Hydro-pedotransfer functions

Figure 1: Deriving HPTFs by relating water supply (CWS) to evapotranspiration (ETIa/E0) (Wessolek et al., 2008)

One approach is the derivation of Hydro-pedotransfer functions (HPTFs). These functions are used to calculate groundwater recharge with the help of annual percolation rates on a regional scale, depending on the land cover of the study area (for example, farmland, grassland, coniferous or deciduous forest). An advantage of this method is that only easily accessible data is required (Figure 1) – for instance, climate data such as precipitation (P) or evaporation rates (E0), soil parameters including the water available to plants in the effective root zone (Wa) or the resulting capillary rise of groundwater (Qa). Thus, HPTFs integrate approaches from soil, plant and water sciences.

So far, this approach is only applicable to land conditions in Central Europe. As part of the MedWater project, the transfer of this method to the characteristics of karst aquifers and a Mediterranean climate will be examined. The estimated groundwater recharge rate will then be compared to results from a SWAT-model and a neural network and be integrated into a numerical groundwater model of the Western Mountain Aquifer.

SWAT-Modeling

The Soil & Water Assessment Tool (SWAT) was developed by the U.S. Department of Agriculture to simulate the hydrologic cycle on a watershed-level.

For generating a model, SWAT uses the following input data:

  • Digital Elevation Map: generates topographical sub-basins
  • Soil map: provides information about the area’s characteristic soil parameters
  • Land use map: determines evapotranspiration
  • Weather data: precipitation, temperature, solar radiation, relative humidity, wind velocity
Figure 1: Schematic illustration of the hydrologic cycle (Neitsch et al., 2009)

SWAT uses this information to numerically calculate the course and amount of induced water. SWAT uses the water balance equation, which states that a system’s stored water equals the difference between inflow and outflow (Figure 2).

In the MedWater project, a SWAT-Model is being developed to simulate the recharge zone of the Western Mountain Aquifer. The created model will be calibrated and validated on the basis of runoff and soil moisture data. Subsequently, the model can be used to determine groundwater recharge, as well as the impacts of land use and climate change.

Neural networks

Figure 3: A neural network relates precipitation and other meteorological data to spring discharge using machine learning

Neural networks are nonlinear data modeling tools to simulate complex relationships between model inputs and outputs. These machine learning methods are novel tools of data science, which have recently gained popularity for geoscience applications. Neural networks “learn” the non-linear relationships between physical parameters and processes and can be used for predicting the output of these relationships without actual physics-based modeling.

Within MedWater, the goal of neural network methods is to create a predictor for groundwater recharge based on precipitation time series. A trained neural network could provide model output estimates without the computational costs of numerical modeling and could potentially mimic the complex recharge behavior of karst aquifers. The network will be trained using precipitation data and the observed changes in the flow of selected streams.

The major challenge is to successfully train a neural network in a relatively data-scarce scenario. To mitigate this, the training dataset will be enhanced and artificially extended with randomly generated data. The possibility of model generalization will be also explored, i.e. whether a neural network trained on a specific catchment can be used at other locations.