Effect of observation scale on remote sensing based estimates of evapotranspiration in a semi-arid row cropped orchard environment

Understanding in detail the spatial distribution of evapotranspiration (ET) in row cropped fruit production areas with diverse water requirements is vital for monitoring water use and efficient irrigation scheduling. Spatially distributed ET for these environments can be estimated using remote sensing (RS). However, the computation of RS based ET under such conditions is complicated because of the complex parameterizations that are required to derive ET for the mixed pixels comprising of bare soil and well-watered plants typical of row cropped areas. Also, the parameterization of these processes is not scale invariant, owing to change in the percentage of vegetation cover in the mixed pixels across remote sensing observation scales. In this study, our main objectives were (1) to isolate and evaluate the effect of varying spatial scales (comparable to canopy sizes and larger) of the remote sensing data on ET estimates; and (2) provide an operational method for estimating remote sensing based ET for row cropped conditions. ET was computed using an empirical technique (S-SEBI: Simplified-Surface Energy Balance Index Algorithm) for almond and pistachio orchards from remote sensing imagery collected at a scale comparable to the canopy sizes of the trees (5.8 and 7.2 m) and a scale that was much larger than the canopy size (120 m) using the MASTER and Landsat sensors, respectively. In order to account for the effect of mixed pixels, a Normalized Difference Vegetation Index based correction factor was applied to the derived ET values and the results averaged for different fields were validated with Penman–Monteith based ET estimates. It was found that the corrected mean ET estimates at 120 m were in agreement with the Penman–Monteith based ET estimates (RMSE_average = 0.12 mm/h), whereas they were underestimated at the finer resolutions. Our results indicated that a remote sensing pixel resolution comparable to the row spacing and smaller and comparable to the canopy size overestimated the land surface temperature and consequently, underestimated ET when using operational models that do not account for vegetation and soil temperature separately. The results of the application of the NDVI correction factor indicates that good spatial estimates of crop ET can be made for crops growing in orchards using simple ET models that require minimal data and freely available Landsat imagery. These findings are very encouraging for the regular monitoring of crop health and effective management of irrigation water in highly water stressed agricultural environments.