Evaluations of MODIS and microwave based satellite evapotranspiration products under varied cloud conditions over East Asia forests

Abstract

Satellite remote sensing is an important tool to retrieve terrestrial evapotranspiration (ET). Widely-used MOD16 ET product (MOD-ET) is a representative of Penman-Monteith method coupled with MODerate Resolution Imaging Spectroradiometer (MODIS) optic observations. Although MOD-ET has been extensively evaluated over the world, its accuracy under various cloud conditions remains unevaluated. Combining MODIS-observed cloud cover (Frc) and in-situ measurements at sixteen forests sites in East Asia, we evaluated 8-day MOD-ET and its primary MODIS inputs (i.e. LAI, FPAR and albedo) from clear to cloudy sky. A new satellite microwave ET method based on microwave Emissivity Difference Vegetation Index (EDVI-ET) was also compared with MOD-ET. Results showed that the accuracy of MOD-ET was highly variable under the changing Frc over the forests. The largest bias (>30%) in MOD-ET was found under Frc < 10%, while slight increase of Frc (10%–30%) reduced the bias (10%–20%). Further increase of Frc (>30%) deteriorated the bias in MOD-ET (20%–30%). In contrast, EDVI-ET performed stably with lower bias (<13%) under similar sky conditions. A higher Taylor score (0.84) for EDVI-ET was found under Frc of 4% than MOD-ET (0.66), while two ET had close scores (>0.81) under other sky conditions. Further investigation found that MOD-ET over four tropical coastal forests contributed most to the bias, especially under least cloudy sky. A case study at a tropical forest showed that MODIS LAI/FPAR and surface albedo products were overestimated, which could directly cause the overestimation of canopy-scale conductance and the underestimation of net solar radiation in MOD-ET method, respectively. Analysis showed that the bias in MOD-ET was significantly related to the bias in MODIS LAI under various Frc, but it was weakly related to that in MODIS albedo, suggesting that LAI-based conductance might dominate the overestimation of MOD-ET. During a consecutive cloud cover period when fewer reliable MODIS pixels are available, slight increase of clouds partly reduced MODIS-observed signals of LAI/FPAR and increased those of albedo over the tropical forest, resulting in the lower bias in MOD-ET. More clouds reduced surface MODIS albedo and might increase the uncertainty in all-sky shortwave radiation from reanalysis data, which deteriorated MOD-ET accuracy under overcast sky. Our study highlighted the importance of cloud impacts on the satellite ET estimation.