Cristobal, J; Poyatos, R; Ninyerola, M; Llorens, P; Pons, X (2011). Combining remote sensing and GIS climate modelling to estimate daily forest evapotranspiration in a Mediterranean mountain area. HYDROLOGY AND EARTH SYSTEM SCIENCES, 15(5), 1563-1575.
Evapotranspiration monitoring allows us to assess the environmental stress on forest and agricultural ecosystems. Nowadays, Remote Sensing and Geographical Information Systems (GIS) are the main techniques used for calculating evapotranspiration at catchment and regional scales. In this study we present a methodology, based on the energy balance equation (B-method), that combines remote sensing imagery with GIS-based climate modelling to estimate daily evapotranspiration (ET(d)) for several dates between 2003 and 2005. The three main variables needed to compute ET(d) were obtained as follows: (i) Land surface temperature by means of the Landsat-5 TM and Landsat-7 ETM+ thermal band, (ii) air temperature by means of multiple regression analysis and spatial interpolation from meteorological ground stations data at satellite pass, and (iii) net radiation by means of the radiative balance. We calculated ET(d) using remote sensing data at different spatial and temporal scales (Landsat-7 ETM+, Landsat-5 TM and TERRA/AQUA MODIS, with a spatial resolution of 60, 120 and 1000 m, respectively) and combining three different approaches to calculate the B parameter, which represents an average bulk conductance for the daily-integrated sensible heat flux. We then compared these estimates with sap flow measurements from a Scots pine (Pinus sylvestris L.) stand in a Mediterranean mountain area. This procedure allowed us to better understand the limitations of ET(d) modelling and how it needs to be improved, especially in heterogeneous forest areas. The method using Landsat data resulted in a good agreement, R(2) test of 0.89, with a mean RMSE value of about 0.6 mm day(-1) and an estimation error of +/- 30 %. The poor agreement obtained using TERRA/AQUA MODIS, with a mean RMSE value of 1.8 and 2.4 mm day(-1) and an estimation error of about +/- 57 and 50 %, respectively. This reveals that ET(d) retrieval from coarse resolution remote sensing data is troublesome in these heterogeneous areas, and therefore further research is necessary on this issue. Finally, implementing regional GIS-based climate models as inputs in ET(d) retrieval have has provided good results, making possible to compute ET(d) at regional scales.