Bias correction framework for satellite precipitation products using a rain/no rain discriminative model

Area of Science:

• Hydrology and Remote Sensing
• Atmospheric Science and Meteorology

Background:

• Satellite precipitation products (SPPs) offer global coverage but have accuracy limitations in critical applications like hazard forecasting and water resource management.
• Rain/no-rain detection errors significantly impact the accuracy of daily SPPs, necessitating improved detection capabilities.
• Existing SPPs, such as the Integrated Multi-satellitE Retrievals for Global Precipitation Measurement (GPM) mission (IMERG), require bias correction for enhanced reliability.

Purpose of the Study:

• To develop and validate a precipitation bias correction framework (PBCF) for improving the accuracy of daily SPPs.
• To enhance the rain/no-rain detection ability of SPPs using an artificial neural network (ANN) based discriminative model.
• To assess the performance of the proposed PBCF in correcting IMERG precipitation data over the Hanjiang River Basin (HRB).

Main Methods:

• A precipitation bias correction framework (PBCF) was developed, incorporating a rain/no-rain discriminative model.
• An artificial neural network (ANN) was employed to construct the discriminative model using land and climate variables from the ERA5-Land reanalysis dataset.
• The PBCF was applied to correct daily IMERG precipitation data over the HRB (2004-2018), with validation using meteorological station data.

Main Results:

• The PBCF significantly reduced IMERG's bias, increasing the correlation coefficient (R) by 19.4% and decreasing RMSE and MAE by 19.0% and 29.8%, respectively.
• The R/NR discriminative model achieved a classification accuracy of 86.5% and improved the equitable threat score (ETS) from 0.15 to 0.58.
• The proposed PBCF demonstrated superior performance compared to the cumulative distribution function mapping method for correcting IMERG data.

Conclusions:

• The developed PBCF effectively improves the accuracy of daily SPPs by enhancing rain/no-rain detection.
• The ANN-based R/NR discriminative model is a viable approach for correcting SPP biases.
• This study offers a novel and efficient method for bias correction of satellite precipitation data, benefiting hydrological and water resource applications.