Upscaling of longwave downward radiation from instantaneous to any temporal scale: Algorithms, validation, and comparison

Abstract

Surface longwave downward radiation (LWDR) is a key factor affecting the surface energy balance. The daily LWDR and the diurnal variations of LWDR are of great significance for studies of climate change and surface processes. How to obtain LWDR at an averaged temporal scale from instantaneous LWDR is one of the long-standing problems in the field of radiation budget from remote sensing. In this paper, two temporal upscaling methods are introduced, namely, a method based on the diurnal variations of LWDR (diurnal variation based, DVB) and a method based on random forest regression (RFR). The results reveal that: (1) The DVB method has a global hourly and daily LWDR root-mean-square error (RMSE) of less than 21 W/m2 and 15 W/m2, respectively, and the RMSE of the daily LWDR based on RFR is less than 7 W/m2; (2) When compared with four existing statistical interpolation methods, the DVB method can not only ensure the accuracy, but also can overcome the problem of missing samples and/or an abnormal samples during upscaling; (3) Except for directly predict daily LWDR, the DVB methods can also obtain more accurate LWDR diurnal variations such as hourly, half-hourly etc. The RFR method enables high-efficiency and accurate estimation of daily averaged LWDR from instantaneous measurements. Compared with existing methods and products, the proposed methods are not only efficient, but also have a superior applicability and reliable accuracy. The proposed strategies provide new ideas for the community in estimating LWDR at continuous temporal scales from remotely sensed measurements.

Keywords

• Longwave downward radiation
• Temporal upscaling
• Statistical interpolation
• Diurnal variation
• Random forest regression