Improving the detection accuracy of micropulse lidar based on RBF neural network

Abstract

Atmospheric aerosol monitoring is essential for understanding climate change and assessing environmental quality. Micro Pulse Lidar (MPL) enables detection of aerosol extinction coefficients (AEC), but its performance is often limited under complex conditions, particularly at low altitudes and in the presence of clouds, due to uncertainties in key parameters such as the AEC at calibration altitude and aerosol extinction-to-backscatter ratio (AEBR). This study introduces a Radial Basis Function (RBF) neural network framework to address these limitations. Through simulations, the RBF network effectively captures the complex nonlinear relationship between backscattered signals and AEC, achieving high accuracy across over 4000 training samples and overcoming MPL’s detection range limitations. Cloud subtraction preprocessing further improves accuracy from ground level to cloud heights. Synchronous experiments with MPL and Raman Lidar (RL) validate the RBF network’s superior performance under both clean and polluted conditions, especially in reducing cloud interference. This research provides a novel pathway for applying MPL in complex atmospheric conditions, enhancing the precision and utility of aerosol monitoring.