干旱区生态环境知识资源中心

Accurate estimation of evapotranspiration (ET), as the second largest term of the water budget after precipitation, is necessary for arid zone catchments, where conservation of limited water resources is vital for the natural environment and human lives but where reliable ground measurements are sparse due to low population densities. Emergence of remotely sensed products has given promising results in better quantification of spatial and temporal behaviour of evapotranspiration, but these products are mainly calibrated in more humid zones with almost stable land cover, and the emphasis is on transpiration as the main term of water loss. Therefore, utilization of remotely sensed ET products requires further evaluation in arid regions. MOD16 and CMRSET are two of the most well-known actual ET (AET) datasets available for Australia, provided by NASA and CSIRO respectively. In order to validate the accuracy of these products, a 480 km reach of Cooper Creek in the Lake Eyre Basin (LEB), with an area of 22,000 km(2), was chosen. Transmission losses (TL) in this large-scale, low gradient floodplain were estimated by considering rainfall, gauged (inflow and outflow) and ungauged streamflow volumes for 14 flood events between 2000 and 2012. It was assumed that satellite AET retrieved during and post-flooding should approximate to the transmission loss estimated for individual flood events. Results indicate that CMRSET AET performs well in large events (TL>9500 GL) with less than 8% absolute error, but its performance decreases for medium sized floods (TL=4000-8000 GL), with an absolute error of 17-48%, and the highest error occurs at 406% for the smallest flood (TL=496 GL). AET time-series of 8-day ET volume reveals that CMRSET estimates a higher than expected minimum of 85 GL/8-day for the floodplain during prolonged droughts but reproduces dynamics that well correlate with variation in the inundation area and responds to both water and vegetation land cover variations. MOD16 underestimates AET for all the floods, only accounting for 17-37% of total loss and it is does not correlated well with the inundation extent. Pixel based comparison of MOD16 time series also shows very low ET values for peak vegetation in this arid region. For CMRSET, it is expected that a lower limit for minimum AET from non-inundated pixels could improve water balance estimates for medium sized floods but will have a lesser effect for small floods. The error in the latter case will also be due to the limitation of MODIS spatial resolution in properly capturing land cover changes (i.e. inundation extents and vegetation responses). Misclassification of land cover and the dependency of the MOD16 model on estimating water cover fraction based on relative humidity are suggested to be the main sources of error for this product.