RADARSAT Constellation Mission Compact Polarisation SAR Data for Burned Area Mapping with Deep Learning

Yu Zhao,Yifang Ban

2024-12-16

Abstract:Monitoring wildfires has become increasingly critical due to the sharp rise in wildfire incidents in recent years. Optical satellites like Sentinel-2 and Landsat are extensively utilized for mapping burned areas. However, the effectiveness of optical sensors is compromised by clouds and smoke, which obstruct the detection of burned areas. Thus, satellites equipped with Synthetic Aperture Radar (SAR), such as dual-polarization Sentinel-1 and quad-polarization RADARSAT-1/-2 C-band SAR, which can penetrate clouds and smoke, are investigated for mapping burned areas. However, there is limited research on using compact polarisation (compact-pol) C-band RADARSAT Constellation Mission (RCM) SAR data for this purpose. This study aims to investigate the capacity of compact polarisation RCM data for burned area mapping through deep learning. Compact-pol m-chi decomposition and Compact-pol Radar Vegetation Index (CpRVI) are derived from the RCM Multi-look Complex product. A deep-learning-based processing pipeline incorporating ConvNet-based and Transformer-based models is applied for burned area mapping, with three different input settings: using only log-ratio dual-polarization intensity images images, using only compact-pol decomposition plus CpRVI, and using all three data sources. The results demonstrate that compact-pol m-chi decomposition and CpRVI images significantly complement log-ratio images for burned area mapping. The best-performing Transformer-based model, UNETR, trained with log-ratio, m-chi decomposition, and CpRVI data, achieved an F1 Score of 0.718 and an IoU Score of 0.565, showing a notable improvement compared to the same model trained using only log-ratio images.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **Using the compact polarisation C - band SAR data of the RADARSAT Constellation Mission (RCM) for burned area mapping and evaluating its effectiveness in the deep - learning framework**.

Specifically, the paper aims to:

1. **Show compact polarisation m - χ decomposition and compact polarisation vegetation index (CpRVI) images**: These images are generated from RCM C - band multi - look complex products and can effectively supplement the log - ratio intensity images, thereby improving the effect of burned area mapping.
2. **Combine compact polarisation RCM data with convolutional neural network (ConvNet) - based and Transformer - based segmentation models**: By using compact polarisation data, the proposed workflow can effectively detect wildfires at 9 time points in 7 study areas in the test dataset. In particular, the Transformer - based model UNETR performs best in terms of F1 score (0.718) and intersection - over - union (IoU) score (0.565), outperforming other benchmark models.

### Core problems of the paper

1. **Limitations of optical sensors**: Although optical satellites such as Sentinel - 2 and Landsat are widely used for burned area mapping, their effectiveness is affected by clouds and smoke, and they cannot penetrate these obstacles to detect burned areas.
2. **Application potential of compact polarisation SAR data**: Although dual - polarisation Sentinel - 1 and quad - polarisation RADARSAT - 1/ - 2 C - band SAR can penetrate clouds and smoke, there are limited studies on using compact polarisation RCM data for burned area mapping. Therefore, this paper aims to fill this research gap and explore the application potential of compact polarisation RCM data in the deep - learning framework.

### Research methods

1. **Data sources**: Use RCM C - band multi - look complex products from the Canadian wildfire seasons in 2023 and 2024, covering Alberta, British Columbia, Quebec and the Northwest Territories.
2. **Data processing**: Process MLC products through the European Space Agency's SNAP platform to generate ground range detected (GRD) images, m - χ decomposition images and compact polarisation vegetation index (CpRVI) images.
3. **Model training and evaluation**: Use six different deep - learning models (including ConvNet - based and Transformer - based models) for training, and evaluate model performance through F1 score and IoU score.

### Main contributions

1. **Prove compact polarisation m - χ decomposition and CpRVI images**: These images significantly supplement the log - ratio intensity images and improve the effect of burned area mapping.
2. **Propose an effective deep - learning workflow**: Combining compact polarisation RCM data with ConvNet - based and Transformer - based segmentation models can effectively detect wildfires in multiple study areas and time points. In particular, the Transformer - based model UNETR performs excellently.

In conclusion, this paper solves the limitations of optical sensors in burned area mapping by introducing compact polarisation RCM data and deep - learning techniques, providing a new solution for burned area monitoring.