Remote sensing based evapotranspiration and runoff modeling of agricultural, forest and urban flux sites in Denmark: From field to macro-scale

E. Bøgh, R.N. Poulsen, M. Butts, P. Abrahamsen, Ebba Dellwik, S. Hansen, Charlotte Bay Hasager, Andreas Ibrom, J.-K. Lørup, Kim Pilegaard, H. Søgaard

    Research output: Contribution to journalJournal articleResearchpeer-review


    Evapotranspiration (E) and runoff (RT) was modeled for the island of Sjælland (≈7330 km2) in Denmark at multiple spatial scales encompassing agricultural, forest and urban land surfaces. National data were used to represent spatial variations in climate, soil properties and lower boundary conditions, and the EOS/MODIS Normalized Difference Vegetation Index (NDVI) was used to map (a) the temporal development in leaf area index for agricultural fields, (b) a dynamic “canopy” coefficient (Kc) of forests being scaled between its minimum and maximum values for use in the FAO Penman–Monteith equation, and (c) the impervious land cover fraction of urban regions. At field level, the use of local-scale model parameters, NDVI time series and site-specific methodologies to simulate E of the 3 major land surface types (agricultural land, forests and urban regions) explained 67–79% of the observed variability in eddy covariance latent heat fluxes. The “effective” spatial resolution needed to adopt local-scale model parameters for spatial-deterministic hydrological modeling was assessed using a high-spatial resolution (30 m) variogram analysis of the NDVI. The use of the NDVI variogram to evaluate land surface heterogeneity is based on the assumption that sub-class soil heterogeneity can be indirectly represented by the observed spatial variations in NDVI due to its close affiliation with vegetation growth, soil water uptake and evapotranspiration. Multiple spatial resolution water balance simulations were compared to validate the identified effective spatial resolution (500 m) model representation of land cover, NDVI and drainage pattern. Simulated RT of 30 catchments were compared with the fast-flow component of stream discharge data (Q − Qb) which is insensitive to groundwater abstraction and most sensitive to the spatial land surface representation. A good agreement was observed in the timing and size of peak flows in catchment dominated by agricultural, forest and urban land uses in periods when E has important control on the water balance and soil water percolation to groundwater is negligible (Days 125–300). The presence/absence of pipe drains, urban surface runoff and forest parameterization cause very large differences in the water balance of agricultural, forest and urban regions. The results show that the use of local-scale standard model parameters and NDVI time series representing agricultural, forest and urban land surfaces in physically based hydrological modeling makes it possible to reproduce much of the observed variability (48–73%) in stream flow (Q − Qb) when data and modeling is applied at an effective spatial resolution capable of representing land surface heterogeneity. In order to further improve the results, (1) advanced spatial parameterization methods are needed to improve the modeling of bare soil E of agricultural fields, (2) the impact of local conditions, such as tree age and nutrient levels, should be used to parameterize the maximum Kc used for forest E modeling, (3) high accuracy remote sensing based estimation of vegetation parameters is particularly important during sparsely vegetated conditions, and (4) the use of component stream flow data to evaluate the physical consistency of spatial-deterministic models appears to be feasible and should be further explored.
    Original languageEnglish
    JournalJournal of Hydrology
    Issue number3-4
    Pages (from-to)300-316
    Publication statusPublished - 2009


    • Wind energy
    • Meteorology

    Fingerprint Dive into the research topics of 'Remote sensing based evapotranspiration and runoff modeling of agricultural, forest and urban flux sites in Denmark: From field to macro-scale'. Together they form a unique fingerprint.

    Cite this