Estimation of high spatial resolution all-weather near-surface air temperature using FY-4A AGRI observations

Abstract

Near-surface air temperature (Tair) at high spatiotemporal resolution is critical to meteorological, hydrological, and ecological processes. At present, the satellite based Tair estimation is generally restricted to clear skies, while the Tair estimation method under cloudy skies needs further development. In this study, a neural network-based approach was proposed to estimate the instantaneous all-weather Tair with high spatial resolution (i.e., 250 m). The approach includes two independent Tair estimation models for clear and cloudy sky conditions. The inputs of the model mainly include FY-4A Advanced Geostationary Radiation Imager (AGRI) products, Global Forecast System (GFS) Tair forecasts, elevation, MODIS normalized difference vegetative index and other auxiliary parameters. The AGRI land surface temperature was used to estimate Tair under clear skies (Ta,clear), while cloud top height and temperature were used to estimate Tair under cloudy skies (Ta,cloudy). Validation using the in-situ Tair revealed that the correlation coefficient (R) and root mean square error (RMSE) were 0.986 and 1.42 °C for Ta,clear, and 0.983 and 1.46 °C for Ta,cloudy, respectively. In addition, the AGRI estimated Tair was compared with ERA5-Land product, and the accuracy of AGRI Tair was better than the ERA5-Land data (RMSE = 2.2 °C, R = 0.984). The proposed model has good potential for all-weather Tair estimation with high spatiotemporal resolution and can be easily applied to other geostationary satellites.