Prediction of diffuse solar radiation by integrating radiative transfer model and machine-learning techniques

Area of Science:

• Atmospheric Science
• Agricultural Science
• Remote Sensing

Background:

• Diffuse radiation significantly impacts agricultural ecosystems, influencing carbon exchange and energy/material flows.
• Accurate measurement of diffuse radiation is challenging and costly, limiting data availability in China.
• Existing data gaps hinder comprehensive understanding and modeling of agricultural ecosystem processes.

Purpose of the Study:

• To develop and compare hybrid models for simulating diffuse radiation in China.
• To quantify uncertainties in diffuse radiation simulations arising from measurement errors.
• To analyze the contribution of various factors to diffuse radiation.

Main Methods:

• Developed five hybrid models combining radiative transfer models (RTM) with machine learning (ML) algorithms (RF, XGBoost, MLP, DNN, CNN).
• Utilized data from AERONET, BSRN, Wuhan University, CERN, and GLASS surface albedo.
• Quantified simulation uncertainties and analyzed variable contributions to diffuse radiation.

Main Results:

• The RTM-RF model demonstrated superior performance, achieving high R² values (0.94-0.98) and low RMSE (9.56-13.27 W m⁻²).
• Aerosol optical depth (AOD) and single-scattering albedo were identified as key contributors to simulation uncertainty.
• AOD, solar zenith angle (SZA), and single-scattering albedo were the most significant variables influencing diffuse radiation.

Conclusions:

• The RTM-RF hybrid model is highly effective for estimating diffuse radiation in China.
• Understanding measurement uncertainties in AOD and single-scattering albedo is critical for accurate simulations.
• The RTM-RF approach offers a reliable and recommended method for diffuse radiation estimation in the region.