Understanding and Correcting Biases in INSAT 3D Cloud Fraction Using High-Resolution Satellite Data

Clouds play a crucial role in the Earth’s energy balance, thereby influencing its climate system.  Cloud fraction (CF) is one of the important Essential Climate Variables. The discrepancies among satellite CF products are due to four effects: the Resolution effect, the View angle effect, the ability of the sensor to detect clouds, and the difference in satellite overpass time. Additionally, the reanalysis data is not a direct observation but rather depends on the model parameters. To understand the diurnal variation of CF, we need to use data from a geostationary satellite after bias correction, if any. To understand the cloud processes that are inevitable in the climate system, we need to compare and study existing cloud products and understand the CF data and biases.  

This study leverages INSAT 3D geostationary satellite data to monitor cloud fraction changes over the Indian region from 2014 to 2024, providing high temporal and spatial resolution insights. We examine diurnal and seasonal patterns in CF and compare them against bias-corrected MODIS, MISR data, and study diurnal variation using ERA5 reanalysis datasets. Preliminary analysis reveals systematic biases in INSAT-3D CF, with differences in amplitude and phase relative to ERA5. Unless the biases in INSAT 3D are quantified and corrected, the diurnal pattern in CF cannot be understood robustly over the Indian region.  

To overcome the Resolution effect, we employ a pattern recognition technique having feature vector to correct the CF bias in the INSAT 3D data using the CF from high spatial resolution satellites such as Sentinel 2 (QA60 cloud mask band). The optimized feature vector includes - Ae (standard method estimate of CF), Aedge (fraction of cloudy pixels that border a clear pixel on at least 1 of their eight sides or vertices), the first moment invariant (Hu moment), Mean, Variance and entropy of the grey levels in the scene, which makes it a six dimensional vector. The radiance from the thermal band of Landsat 8/9 can be used as an extra dimension during nighttime. The cloud masks that have similar spatial features will have similar true CFs and the degree of correction depends upon ratio of cloud size to pixel size and distribution of true cloud area.