Search

Accurate estimation of evapotranspiration (ET) can be affected by cloud change, while cloud‐induced bias errors in ET change remain poorly understood. Here we evaluated seven ET data sets from satellite microwave and optical models, atmospheric reanalysis and land surface models over East Asia. Results showed that although the data sets estimated ET changes overall well under mild cloud increase, all of them had a systematic overestimation (by 73.8%–159.8%) and deteriorated capability of capturing ET change under heavy cloud increase, especially over dense vegetation with large cloud cover. Overestimated incident solar radiation under clouds contributed to the deteriorated performances of ET. Cloud‐induced change in radiation and its errors played larger role in summer ET change over densely vegetated lands, while those in air temperature and relative humidity were dominant jointly over grass and barren lands. These results can serve as a reference for improving ET models and their forcing inputs under clouds. Plain Language Summary: The change of cloud cover affects terrestrial evapotranspiration (ET), which is a fundamental process in the interaction between land and the atmosphere. ET can be estimated from satellite observations, reanalysis data sets and land surface models. Knowing the cloud‐induced errors in the estimation of ET and its drivers can be very helpful for us to improve the ET modeling under the changing cloud conditions. In this study, we evaluated seven ET data sets and their forcing data from clear sky to cloudy sky conditions over East Asia. We quantified the importance of incident radiation, air temperature and relative humidity inputs in ET data sets under cloudy sky using a machine learning algorithm. Our findings revealed that all ET data sets were overestimated systematically under heavy cloud increase. Cloud‐induced change in radiation was more important over densely vegetated lands, while air temperature and relative humidity played larger roles over grass and barren lands. The result highlights the importance of cloud effects on ET estimations. Key Points: Seven evapotranspiration (ET) data sets from satellite microwave and optical models, reanalysis and land surface models were evaluated under cloud changeAll data sets overestimated the absolute ET values and cloud‐induced changes in ET under heavy cloud increase conditionsSummer ET change under clouds was dominated by radiation in the southern dense vegetation, but by temperature and humidity in grass lands [ABSTRACT FROM AUTHOR]

• A small bias in EDVI-ET is found over eight river basins using water balance method. • EDVI-ET performs a larger magnitude and higher accuracy than MOD16 over crop lands. • Cloud-induced changes in ET vary from 10% to 40% compared to clear sky conditions. • Cloud impacts are more significant for forests, savannas and crops than barrens. • EDVI-ET is most sensitive to radiation for forests, but to humidity for arid barrens. Evapotranspiration (ET) is an important component of the hydrological cycle and energy balance in a land-atmosphere system. Satellite remote sensing has been widely used to estimate regional and global ET, but most previous methods depend on optical measurements that are limited to cloud-free conditions. This makes ET estimation challenging under cloudy sky. Currently, evaluations of satellite ET estimation under various cloud conditions remain lacking at the regional scale. Owing to the ability to penetrate clouds, satellite passive microwave measurements are powerful tools for retrieving ET under clouds. This study evaluated a satellite microwave-based daily ET method under all sky conditions over the part of China between 18°N and 50°N from 2003 to 2010, using microwave emissivity difference vegetation index (EDVI) as the proxy of vegetation water content (VWC). Validations using the surface water balance method found that the estimated ET (EDVI-ET) had an overall small bias (6.18%) in eight river basins. EDVI-ET displayed consistent spatiotemporal patterns with global MOD16 ET, with high spatial correlation (R >0.71) and monthly temporal correlation (R >0.82) throughout four seasons. Their differences were also small (<0.56mmday−1) in forests, savannas, grass/shrubs, and croplands. Furthermore, cloud impacts on the regional ET were found to be significant and spatiotemporally heterogeneous. Both EDVI-ET and in-situ observations at seven flux towers indicated that cloud-induced reduction in daily ET could exceed 30% when the cloud cover increased by 60% (R2 of 0.42, fitting line slope of 0.80, p <0.001). Under increased cloud conditions in summer, the changes in net radiation dominated the ET over dense vegetation in southern China, while the roles of air temperature and humidity increased over water-stressed barrens and short vegetation in northwest China. VWC affected EDVI-ET under clouds in temperate transitional zones from flatlands to highlands. This study highlighted the importance of cloud impacts in satellite ET estimation. [ABSTRACT FROM AUTHOR]

Although satellite-based light use efficiency (LUE) model is widely used to estimate gross primary production (GPP) of terrestrial ecosystems, microwave observations have not been integrated into LUE models. This study developed a new LUE model coupled with a passive microwave vegetation index (Emissivity Difference Vegetation Index, EDVI) for daily GPP estimation. Normalized EDVI (nEDVI), an indicator of canopy-scale leaf development and biomass change, was used as a proxy of fraction of photosynthetically active radiation (FPAR). EDVI-based evaporative fraction (EDVI-EF) under all sky was used to indicate synoptic-scale water stress on LUE. 8-year in-situ measurements from seven flux tower sites (four forests, two grasslands and one croplands) of ChinaFLUX network were used to evaluate the model. We found that nEDVI-based FPAR better captured the short-term variations in daily in-situ GPP (GPP obs) than other optic-based FPAR schemes. This capability of nEDVI was more noticeable under moderate and heavy cloud cover (Frc) conditions. Validations against daily GPP obs at all sites showed that EDVI-based GPP (GPP EDVI) generated an overall small bias of −0.47 gC m−2 day−1 (−8.1%) and good Taylor score (S) of 0.86 at the daily scale. Better accuracy of GPP EDVI was found at forests sites with R2 of 0.43 to 0.73, bias of-5.29% to 3.03% and S of 0.58 to 0.78, respectively. At a tropical forest site with most frequent cloud cover, the model also well captured the variation in GPP obs from clear sky to cloudy sky (R2 of 0.93) with stable accuracies. Furthermore, the accuracy of daily GPP EDVI was found to be comparable with global satellite optic MOD17 GPP (GPP MOD17) and EC-LUE GPP (GPP ECLUE) from 8-day to yearly scales across the sites. In particular, GPP EDVI performed generally smaller bias at evergreen broadleaf forests, while both of GPP MOD17 and GPP ECLUE were overestimated, suggesting that there could be less saturation for microwave-based LUE model over dense vegetation. Although all three satellite LUE models severely underestimated GPP of crop, GPP EDVI generated lower bias (−29.8%) than GPP MOD17 (−66.8%) and GPP ECLUE (−59.5%). Overall, this study is the first attempt toward the integration of microwave-derived variables into LUE model for daily GPP estimation. The microwave-based LUE model has a potential of mapping spatiotemporally continues daily GPP under various clouds. • A LUE model coupled with microwave vegetation index (EDVI) for daily GPP estimation. • FPAR and evaporative fraction were quantified by EDVI-based schemes. • 8-year in-situ GPP from seven eddy covariance flux sites of ChinaFLUX were used. • Model performed daily GPP estimation stably under various cloud cover conditions. • Model had a better accuracy than MOD17 and EC-LUE GPP at forests and crop sites. [ABSTRACT FROM AUTHOR]