Research And Software Implementation Of Cloud Detection Algorithm Over Ice-Snow Based On DPC And POSP Data Collaboration

Ice-snow surface and cloud are important objects for remote sensing observation.On the one hand,clouds play an important role in the Earth’s radiation balance and climate environment changes.On the other hand,cloud cover can obstruct the optical satellite imaging channel during ground observation,reducing the number and quality of available pixels in the image.Ice-snow is a special type of land surface form.Due to the similarity in radiation characteristics between ice-snow surfaces and clouds,it is difficult to detect clouds over ice-snow regions.The Atmospheric Aerosol Multi-angle Polarization Camera(Directional Polarization Camera,DPC)and the high-precision Particulate Observing Scanning Polarimeter(Particulate Observing Scanning Polarimeter,POSP)are installed on the high-spectral observation satellite(GF-5B).The "polarization interplay" scheme of the two payloads can provide more effective observation information and improve cloud detection capability.This paper mainly focuses on the data obtained from DPC and POSP in orbit.and carries out research from three aspects:establishing BRDF(Bidirectional Reflectance Distribution Function)and BPDF(Bidirectional Polarization Distribution Function)models for ice-snow surfaces,studying cloud detection algorithms over ice-snow,and designing data processing software for cloud detection over ice-snow.In the analysis process of selecting the BRDF and BPDF models for ice-snow surfaces,the atmospheric correction is first applied to the multi-angle polarized radiation data obtained from the DPC,which results in the ice-snow surface reflectance and polarized reflectance in different directions.These parameters are then used as inputs for four BRDF and four BPDF models.The optimal model parameters are obtained by using the mean squared error between the model simulation values and the DPC measurements as the objective function and applying genetic algorithm iterations.Taking the Greenland region as an example,the optimal BRDF and BPDF models for ice-snow surfaces are selected based on the comparison between the DPC multi-angle polarized observation data and the model simulation results.The best threshold for identifying ice-snow surfaces is obtained by comparing the measured values and model simulation values of ice-snow surface samples in three different regions.The selected BRDF and BPDF models for ice-snow surfaces provide prior knowledge for extracting ice-snow surface pixels in the observation scenes of cloud detection over ice-snow regions.During the process of ice-snow cloud detection,an algorithm was proposed for the collaborative discrimination of DPC and POSP data.The algorithm mainly includes cloud detection using oxygen A-band apparent pressure,water cloud detection using multi-angle polarization signals,convective cloud detection using the 1380nm shortwave infrared feature band,and ice cloud detection using the improved NDSI(Normalized Difference Snow Index)normalized snow index.By analyzing a large number of samples from multiple observation areas of DPC and POSP,the optimal threshold for each detection determination of ice-snow cloud detection was determined.The detection results in three regions,including Greenland,Antarctica,and Northeast,were compared with the inversion results of MODIS(Moderate-resolution Imaging Spectroradiometer),and the consistency was 83.3%,94.4%,and 88.89%,respectively.The results show that the designed algorithm can effectively detect cloud pixels over ice-snow surfaces,verifying the effectiveness of the algorithm.In the design of the data processing software for ice-snow cloud detection,based on functional and software requirements analysis,C++was selected as the programming language and Qt was used for page development framework.In terms of functionality,four main modules were designed:data extraction,data preprocessing,cloud detection,and product production and archiving.The modules independ on each other,and the interface is designed to be simple and user-friendly,allowing the entire process from data extraction to product production to be completed programmatically.This greatly improves operational efficiency and facilitates maintenance and updates in the future.