Estimation of Top-of-Atmosphere Net Radiation From AVHRR Data

Abstract

The top-of-atmosphere (TOA) net radiation (NR), a key component of the Earth’s energy budget, directly indicates the imbalance between incoming solar radiation from the space and outgoing shortwave/longwave radiation from the Earth’s climate system. However, the spatial resolutions of the existing TOA NR products are too coarse to provide enough details when analyzing the energy budget at regional scales. This letter presents a direct machine learning method to estimate TOA NR by directly linking advanced very high-resolution radiometer (AVHRR) TOA radiances with TOA NR determined by Clouds and the Earth’s Radiant Energy System (CERES) and other information, such as the solar/viewing geometry, land surface temperature (LST), and cloud top temperature determined by Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2). Models are built using a gradient boosting regression tree. Independent test results show that the root mean square error (RMSE) of the model for estimating instantaneous values is 25.16 W/m2. Daily results are converted from the instantaneous results using climatology conversion ratios derived from CERES daily and hourly data. Intercomparisons of the daily results with CERES TOA NR data show that the RMSEs of the estimated AVHRR NR are less than 30 W/m2. The developed algorithm may contribute to generating relatively high-resolution (5-km) AVHRR TOA NR dataset, which will be beneficial in analyzing the regional energy budget.