Snow Mapping from Passive Microwave Brightness Temperature and MODIS Snow Product with Machine learning Approaches

Snow cover is one of the cryosphere's most critical components, representing a vital geophysical variable for climate and hydrology. Monitoring snow cover in Arctic regions has gained increasing significance, particularly considering recent climate warming. Given the complex spatiotemporal variability, inconvenience of transportation, and the remote locations of many snow-covered areas, remote sensing emerges as an ideal technique for data collection to monitor snow cover across various spatiotemporal scales. \nIn contrast to optical remote sensing, passive microwave (PMW) and active microwave (AMW) satellite sensors remain unaffected by clouds and solar illumination, making them widely employed in snow detection. PMW observations have lower spatial resolution and high temporal resolution than AMW, which are suitable for large-scale snow mapping. Integrating optical data and PMW data can significantly enhance the quality of snow cover information. Various machine learning (ML) methods have been pivotal in environmental remote-sensing research in recent years. With the surge in Earth observation big data and the rapid advancements in ML techniques, an array of innovative methods has emerged to facilitate environmental monitoring on a global scale. Thus, a snow-monitoring method has been proposed based on multi-source remote sensing data and ML. The brightness temperature (Tb) data derived from the Advanced Microwave Scanning Radiometer E/2 (AMSR-E/2) Level 3 product and Moderate Resolution Imaging Spectroradiometer (MODIS) snow product serves as the reference for snow cover area (SCA). This study predominantly selects Oct, Dec, Feb, and Apr from 2012 to 2022 as the study periods. The research uses three ML methods, Logistic Regression (LR), Random Forest (RF) and Support Vector Machine (SVM), for snow cover detection based on PMW and MODIS data in the Arctic. The overall accuracy (The ratio of correctly classified as snow plus correctly classified as non-snow points to the total number of points) of ML models in snow detection surpasses 80%, yet it exhibits regional and seasonal variations. Notably, distinctions in the distribution of MODIS snow and PMW snow become evident in two types of areas: regions where MODIS estimates exceed PMW and those where PMW estimates surpass MODIS. ML-based estimation significantly enhances the accuracy of snow monitoring in the latter category, reducing misclassifications and augmenting the precision of snow cover assessment. When comparing the ML-derived SCA, PMW-derived SCA, and MODIS-derived SCA with the snow depth dataset-derived SCA, the ML method exhibited the highest consistency

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.