Analysis and estimation of tallgrass prairie evapotranspiration in the central United States

TitleAnalysis and estimation of tallgrass prairie evapotranspiration in the central United States
Publication TypeJournal Article
Year of Publication2017
AuthorsWagle, P, Xiao, X, Gowda, P, Basara, J, Brunsell, N, Steiner, J, Anup, KC
JournalAgricultural and Forest Meteorology
Accession NumberKNZ001742
KeywordsArtificial neural network, Eddy covariance, Empirical model, ET modeling, remote sensing, Wavelet cross-correlation analysis

Understanding the factors controlling evapotranspiration (ET) of spatially distributed tallgrass prairie is crucial to accurately upscale ET and to predict the response of tallgrass prairie ecosystems to current and future climate. The Moderate Resolution Imaging Spectroradiometer (MODIS)-derived enhanced vegetation index (EVI) and ground-based climate variables were integrated with eddy covariance tower-based ET (ETEC) at six AmeriFlux tallgrass prairie sites in the central United States to determine major climatic factors that control ET over multiple timescales and to develop a simple and robust statistical model for predicting ET. Variability in ET was nearly identical across sites over a range of timescales, and it was most strongly driven by photosynthetically active radiation (PAR) at hourly-to-weekly timescales, by vapor pressure deficit (VPD) at weekly-to-monthly timescales, and by temperature at seasonal-to-interannual timescales at all sites. Thus, the climatic drivers of ET change over multiple timescales. The EVI tracked the seasonal variation of ETEC well at both individual sites (R2 > 0.70) and across six sites (R2 = 0.76). The inclusion of PAR further improved the ET-EVI relationship (R2 = 0.86). Based on this result, we used ETEC, EVI, and PAR (MJ m−2 d−1) data from four sites (15 site-years) to develop a statistical model (ET = 0.11 PAR + 5.49 EVI − 1.43, adj. R2 = 0.86, P < 0.0001) for predicting daily ET at 8-day intervals. This predictive model was evaluated against additional two years of ETEC data from one of the four model development sites and two independent sites. The predicted ET (ETEVI+PAR) captured the seasonal patterns and magnitudes of ETEC, and correlated well with ETEC, with R2 of 0.87-0.96 and RMSE of 0.35-0.49 mm d−1, and it was significantly improved compared to the standard MODIS ET product. This study demonstrated that tallgrass prairie ET can be accurately predicted using a multiple regression model that uses EVI and PAR which can be readily derived from remote sensing data.