Evapotranspiration from Landsat and Modis Images

The author is Vivek Sharma, ASABE Member, graduate research Assistant, Department of Biological Systems 1 Engineering, University of Nebraska-Lincoln, Nebraska, USA. Corresponding author: Vivek Sharma, 158. 2 L.W. Chase Hall, University of Nebraska-Lincoln, NE 68583-0726; phone: 402-419-1687; e-mail: 3 [email protected] 4 5 Abstract: Remote sensing-derived surface energy fluxes enable quantification and understanding of spatial 6 distribution of these fluxes with varying scales and resolutions. However, since the major uncertainty lies in 7 regional estimates of surface energy balance fluxes due to underlying heterogeneous surface and difference in 8 spatial resolution of different satellite sensors [e.g., Landsat (30 m); MODIS (250 m, 500 m, and 1000 m)], it is 9 essential to understand the uncertainty in evapotranspiration (ETc) estimated from different satellite sensors. In 10 this study, the effects of pixel scales on ETc estimation and other parameter that are used to calculate ETc was 11 investigated over different vegetation surfaces in south central Nebraska, U.S.A. Surface Energy Balance 12 Systems (SEBS) was used to estimate spatially-distributed ETc by combining ground-based meteorological data 13 for Landsat and MODIS imagery. The SEBS-estimated surface energy fluxes were compared to the Bowen Ratio 14 Energy Balance System (BREBS)-measured fluxes before up-scaling to the coarser resolution product (60 m to 15 990 m). In order to quantify the differences in ETc estimates both output flux aggregation and input up-scaling 16 procedure were conducted using simple average and nearest neighbor aggregation techniques. Validation 17 results showed a strong correlation with Landsat-based ETc estimates and land use type. For Landsat ETc 18 retrievals, the regression models explained 91% of the variability in the observed data [root mean square error 19 (RMSD) = 0.064 mm/hr]. However, for MODIS-based ETc, the regression model explained only 59% of the 20 variability in observed ETc with a larger error (RMSD = 0.17 mm/hr). On average, MODIS-based ETc was 21 about 31% higher than the BREBS-measured ETc. Aggregation results indicated that for flux aggregation 22 procedures using both simple average and nearest neighbor resampling methods preserve the image scale mean 23 values of the resampled images across all resolutions. A substantial decline in standard deviation (SD) between 24 estimated and measured ETc was observed when fluxes were resampled using simple average. There was about 25 24% and 18% reduction in SD from 30 m to 990 m spatial resolution on May 30 and August 02, 2009, 26 respectively. No change in the SD was observed for nearest neighbor method. Output flux aggregation pixel by 27 pixel comparison resulted larger relative difference, ranging from 5% to 35% between original 30 m Landsat 28 flux and aggregated flux. In input up-scaling relative pixel scale error ranged from 25% to 60% in ETc and 29 other energy balance fluxes. Larger error in input up-scaling is due to the changes in the surface roughness 30 parameters due to aggregation in SEBS model. Simple average output flux aggregation provided less errors in 31 aggregating fluxes to higher resolution than the nearest neighbor approach. Comparison between up-scaled 32 ETc at 480 m spatial resolution with original MODIS image at 500 m illustrated a relative difference of only 33 13.4% using simple average resampling method, compared to 20% when using nearest neighbor resampling 34 method. The simple average aggregation method provided closer representation of ETc values, when compared 35 with the original MODIS ETc at 500 m resolution as compared with the nearest neighbor resampling method. 36