Remote estimation of a managed pine forest evapotranspiration with geospatial technology
Remote sensing has increasingly been used to estimate evapotranspiration (ET) and its supporting parameters in a rapid, accurate, and cost-effective manner. The goal of this study was to develop remote sensing-based models for estimating ET and the biophysical parameters canopy conductance (gc), upper-canopy temperature, and soil moisture for a mature loblolly pine forest ( L.) in the Parker Tract in eastern North Carolina. To validate the remote sensing approach, we acquired long-term on-site eddy flux measurements, including micrometeorological variables and water and energy fluxes. Other measured and derived ET-associated parameters include forest gc, leaf area index, canopy absorbed radiation, canopy temperature, and soil moisture. Multi-temporal cloud-free (â‰¤10% cloud cover) Landsat 7 ETM+ satellite images from 2006-2012 were acquired. Field data for the 2 h (12:00 noon to 2:00 p.m.) means were used in the model, coinciding with the image acquisition time. Individual Landsat bands (1 through 7) and developed image vegetation indices (NDVI, SAVI, and VVI) for the study site were obtained through automated geospatial models and were correlated to measured ET flux and related parameters. An excellent coefficient of determination (R2 = 0.93, n = 42) was obtained for the upper-canopy temperature versus band 6 (thermal infrared) model. However, a low correlation (R2 = 0.36, n = 6) was obtained for gc versus band 5 (mid-infrared) model. The correlation for soil moisture versus band 7 was poor (R2 = 0.05, nÂ = 42), perhaps due to heavy canopy and pine litter ground cover. However, the ET estimation model with multiple image information variables, such as bands 5 and 7, provided a good correlation (R2 = 0.55, n = 35) with less spatial and temporal variation in the datasets, along with no data mining application in model building. Therefore, this study suggests that the remote sensing approach is promising for estimating ET with good accuracy (average model prediction residual error = 25.46 W m-2, 6.18% of the average ET values used in the analysis) for a mature homogenous pine forest. Further work is needed to develop robust remote sensing-based ET models by including spatial variability, sound data mining, high-resolution imagery, and advanced image processing to account for potential modeling uncertainties.