Abstract
Distributed hydrological models are traditionally evaluated against
discharge stations, emphasizing the temporal and neglecting the spatial
component of a model. The present study widens the traditional paradigm
by highlighting spatial patterns of evapotranspiration (ET), a key
variable at the land–atmosphere interface, obtained from two different
approaches at the national scale of Denmark. The first approach is based
on a national water resources model (DK-model), using the MIKE-SHE
model code, and the second approach utilizes a two-source energy balance
model (TSEB) driven mainly by satellite remote sensing data. Ideally,
the hydrological model simulation and remote-sensing-based approach
should present similar spatial patterns and driving mechanisms of ET.
However, the spatial comparison showed that the differences are
significant and indicate insufficient spatial pattern performance of the
hydrological model.
The differences in spatial patterns can partly be explained by the fact that the hydrological model is configured to run in six domains that are calibrated independently from each other, as it is often the case for large-scale multi-basin calibrations. Furthermore, the model incorporates predefined temporal dynamics of leaf area index (LAI), root depth (RD) and crop coefficient (Kc) for each land cover type. This zonal approach of model parameterization ignores the spatiotemporal complexity of the natural system. To overcome this limitation, this study features a modified version of the DK-model in which LAI, RD and Kc are empirically derived using remote sensing data and detailed soil property maps in order to generate a higher degree of spatiotemporal variability and spatial consistency between the six domains. The effects of these changes are analyzed by using empirical orthogonal function (EOF) analysis to evaluate spatial patterns. The EOF analysis shows that including remote-sensing-derived LAI, RD and Kc in the distributed hydrological model adds spatial features found in the spatial pattern of remote-sensing-based ET.
The differences in spatial patterns can partly be explained by the fact that the hydrological model is configured to run in six domains that are calibrated independently from each other, as it is often the case for large-scale multi-basin calibrations. Furthermore, the model incorporates predefined temporal dynamics of leaf area index (LAI), root depth (RD) and crop coefficient (Kc) for each land cover type. This zonal approach of model parameterization ignores the spatiotemporal complexity of the natural system. To overcome this limitation, this study features a modified version of the DK-model in which LAI, RD and Kc are empirically derived using remote sensing data and detailed soil property maps in order to generate a higher degree of spatiotemporal variability and spatial consistency between the six domains. The effects of these changes are analyzed by using empirical orthogonal function (EOF) analysis to evaluate spatial patterns. The EOF analysis shows that including remote-sensing-derived LAI, RD and Kc in the distributed hydrological model adds spatial features found in the spatial pattern of remote-sensing-based ET.
Original language | English |
---|---|
Pages (from-to) | 5987-6005 |
Number of pages | 19 |
Journal | Hydrology and Earth System Sciences |
Volume | 21 |
Issue number | 12 |
DOIs | |
Publication status | Published - 30 Nov 2017 |
Keywords
- DK-model
Programme Area
- Programme Area 2: Water Resources