Snow Parameters Inversion from Passive Microwave Remote Sensing Measurements by Deep Convolutional Neural Networks.

Area of Science:

• Earth Science
• Remote Sensing
• Artificial Intelligence

Background:

• Passive microwave remote sensing (PMRS) is crucial for monitoring snow properties.
• Traditional inverse methods for snow parameter retrieval often face limitations in accuracy and noise sensitivity.

Purpose of the Study:

• To develop a novel deep learning-based inverse method for retrieving snow layer thickness and temperature.
• To enhance the accuracy and robustness of snow parameter inversion using passive microwave data.

Main Methods:

• A deep convolutional neural network (ConvNet) architecture was designed and trained using simulated data.
• The ConvNet comprises convolutional, activation, and fully connected layers for parameter regression.
• Simulated data were generated using conventional computational electromagnetic methods.

Main Results:

• The proposed ConvNet achieved higher accuracy in snow parameter inversion compared to traditional methods.
• The deep learning approach demonstrated strong tolerance to noise in passive microwave data.
• Numerical examples validated the feasibility and effectiveness of the ConvNet for snow parameter retrieval.

Conclusions:

• Deep learning, specifically ConvNets, offers a promising new direction for improving passive microwave remote sensing of snow.
• The developed method provides a significant advancement in accurately and reliably extracting snow properties from remote sensing data.