Gap-filling lake surface temperatures from Advanced Himawari Imager (AHI) by a data-driven approach

Abstract

The Advanced Himawari Imager (AHI) accurately measures the surface temperatures of a large lake, except during cloudy conditions, which cause data gaps and often prevent the detection of their daily cycle. Random forest regression (RFR), a machine learning tool, was employed and tested for various possible predictors to gap-fill the AHI-derived lake surface temperatures (LSTs). The results show that the near-surface lake water temperature (LWT) was the best predictor of LSTs when auxiliary meteorological data such as downward long-wave radiation and wind speed (WS) were used together. The gap-fill operation produced reasonable LSTs with the root mean square error of the LST estimates of 0.1–0.2°C against measured LSTs.