Estimating evapotranspiration based on the satellite-retrieved near-infrared reflectance of vegetation (NIRv) over croplands

Abstract

ABSTRACT Accurate information on cropland evapotranspiration (ET) can facilitate effective agricultural management. However, the application of existing physical models over broad regions may be impeded due to the need for difficult to acquire information about environmental factors that constrain ET. The recently developed near-infrared reflectance of vegetation (NIRv), which can reasonably characterize ecosystem photosynthesis without the need for additional environmental information, is potentially useful for estimating cropland ET and reducing the difficulty in ET modeling. As such, we proposed two simply formulated semi-empirical models that utilize NIRv as a major factor constraining cropland ET. The first model, termed Penman–Monteith+ (PM+), computed canopy transpiration using the PM equation along with canopy conductance values estimated from NIRv-derived gross primary productivity (NIRv-GPP) and calculated soil evaporation using an empirical approach. Another model, termed underlying water-use efficiency+ (uWUE+), used the uWUE approach along with the NIRv-GPP to predict ET. We calibrated and validated PM+ and uWUE+ over 32 cropland flux sites and then compared them with six complex models. The better model between PM+ and uWUE+ was applied to estimate regional ET over North China Plain (NCP), where typical C3 and C4 crops were planted during 2010–2018, along with remote sensing and meteorological data. Results indicated that the two new models can reasonably estimate cropland ET. The PM+ model reproduced an eight-day value of ET (denoted as eight-day ET) with R2 = 0.741 and RMSE = 5.638 mm/8d, slightly better than the uWUE+ model (R2 = 0. 674 and RMSE = 6.275 mm/8d) for a cross-site validation over 17 flux sites of the validation dataset. Site-level validation revealed consistently better performance of PM+ compared to uWUE+ over most flux sites. For the comparisons with six existing models, PM+ can perform better than all these models and uWUE+ is better than five of them. Subsequently, the PM+ (uWUE+) model reasonably reproduced eight-day ET over four sites under a dry climate (Arid Index ≤ 0.5) with site-level R2 = 0.691 (0.617) and RMSE = 6.663 (7.906) mm/8d on average, which is better than the six existing models. In addition, the regional mean annual ET varied from 590 mm per year (mm/yr) to 680 mm/yr over NCP with a significant increasing trend of 9.05 mm/yr (p < 0.01), and the ET values in maize growing season were higher than that in wheat growing season. Our results demonstrated that the simply formulated PM+ and uWUE+ models can provide simple and robust approaches to estimate regional and global cropland ET.