New Methods for Deriving Clear-Sky Surface Longwave Downward Radiation Based on Remotely Sensed Data and Ground Measurements

Accurate knowledge of spatiotemporal characteristics of land surface energy budget is essential to understand the Earth's surface system. To some extent, the continuous development of remote sensing technology makes the estimation of global high-resolution surface energy budget possible. However, it is still challenging to accurately derive the global longwave downward radiation (LWDR) from space. In this study, two advanced methods were proposed to estimate clear-sky LWDR based on Moderate-resolution Imaging Spectroradiometer (MODIS) products and corresponding ground measured LWDR from several global distributed radiation networks. One is a sensor-based Random Forest (RF) method that uses MODIS's surface elevation, sensor zenith angle, column precipitable water vapor (PWC), top of atmosphere radiance of band 28, 31, 34, and 36 as inputs; the other is a sensor-independent nonlinear regression method, which only uses the land surface temperature (LST) and PWC as inputs. The validation results show that the root mean square errors of RF and nonlinear regression method are less than 25.5 W/m² and bias are less than 0.5 W/m² under different tenfold cross-validation schemes. In addition, the sensitivity analysis indicates that the two models are very stable, and the retrieval errors are independent of changes of LST and PWC, and the accuracy does not rely on variations of land cover types, different observing angles, and different seasons. The comparison analysis indicates that the proposed methods are comparable or even better than existing algorithms. More importantly, the proposed sensor-independent regression method was designed for generality purpose so can easily be utilized with reasonable accuracy.