Estimation of cloud top parameters from Himawari-8/AHI measurements with infrared spectral bands using the Random Forest method

摘要

We have developed a rapid simplified algorithm to estimate cloud top properties from infrared bands of Himawari based on machine learning. The new-generation geostationary satellite of Himawari-8 with Advanced Himawari Imager (AHI) provides high temporal (every 10 min) and high spatial resolution. CALIPSO (Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations) provides cloud top parameters with high accuracy, but with limited temporal-spatial resolution. This paper reports on a study to derive the cloud top properties from combined AHI and CALIPSO using Random forests (RFs) algorithm, an advanced machine learning (ML) method with better accuracy than that from the traditional physical algorithms. Further sensitivity and validation analyses help determine the optimal RF classification and regression models for predicting process. The selected RF regression model is found to predict cloud top properties with highly consistent with CALIPSO observations (correlation coefficients are 0.89、 0.89、 0.90 for CTH, CTP, CTT respectively). New algorithm provides a robust and rapid algorithm of cloud top properties and we find significant accuracy improvements compared to AHI. Based on the accuracy evaluation of the model estimation results, the characteristics of cloud top properties in time and space are analyzed, and a typical case is selected to study. The application of the algorithm and error analysis are carried out to evaluate the estimation ability of cloud top parameters. The new approach could be used to process data from advanced geostationary imagers for climate and weather applications.