Enhancing Cloud Phase Identification With the Vulture Algorithm-Optimized Random Forest

Cloud phase identification is a critical aspect of cloud microphysical characteristics research, pivotal for climate studies, weather forecasting, and validation of climate models. Traditional methods, employing polarization lidar and millimeter cloud radar (MMCR), often rely on simplistic threshold algorithms, leading to low classification accuracy and reliability. This letter introduces a refined cloud phase recognition technique based on an optimized random forest (RF) model, addressing the nuanced identification challenges of cloud phases. The approach involves optimizing the decision trees and the number of randomly selected features within the RF model using the African vulture optimization algorithm (AVOA) to achieve near-optimal classification performance and improve accuracy to 88.10%. The proposed method has been validated and assessed using actual measurement data and compared with results from multiple data sources, including polarization lidar, MMCR, and atmospheric temperature, demonstrating high consistency.