Snow Coverage Mapping by Learning from Sentinel-2 Satellite Multispectral Images via Machine Learning Algorithms

Yucheng Wang,Xiaojun Zhai,Fanlin Meng,Cunjia Liu,Jinya Su

doi:10.3390/rs14030782

Yucheng Wang, Xiaojun Zhai + Show 3 more

Open Access

https://doi.org/10.3390/rs14030782

Copy DOI

Journal: Remote sensing	Publication Date: Feb 8, 2022
Citations: 11	License type: CC BY 4.0

Affiliation: University of Essex, Loughborough University

Abstract

Snow coverage mapping plays a vital role not only in studying hydrology and climatology, but also in investigating crop disease overwintering for smart agriculture management. This work investigates snow coverage mapping by learning from Sentinel-2 satellite multispectral images via machine-learning methods. To this end, the largest dataset for snow coverage mapping (to our best knowledge) with three typical classes (snow, cloud and background) is first collected and labeled via the semi-automatic classification plugin in QGIS. Then, both random forest-based conventional machine learning and U-Net-based deep learning are applied to the semantic segmentation challenge in this work. The effects of various input band combinations are also investigated so that the most suitable one can be identified. Experimental results show that (1) both conventional machine-learning and advanced deep-learning methods significantly outperform the existing rule-based Sen2Cor product for snow mapping; (2) U-Net generally outperforms the random forest since both spectral and spatial information is incorporated in U-Net via convolution operations; (3) the best spectral band combination for U-Net is B2, B11, B4 and B9. It is concluded that a U-Net-based deep-learning classifier with four informative spectral bands is suitable for snow coverage mapping.

Highlights

Remote sensing technology (e.g., satellite-based, unmanned aerial vehicle (UAV)based, ground-based) can provide a nondestructive way of spatially and temporally monitoring the targets of interest remotely, which has drawn increasing attention recently with the rapid development of autonomous system innovation, sensing technology and imageprocessing algorithms, especially with convolutional neural network-based deep-learning approaches
The proposed database would on the one hand be used to evaluate the performance of different snow prediction tools, and on the other hand enable the future development of more advanced algorithms for snow coverage mapping
This work investigates the problem of snow coverage mapping by learning from Sentinel-2 multispectral satellite images via conventional machine- and recent deep-learning methods

Summary

Introduction

Remote sensing technology (e.g., satellite-based, unmanned aerial vehicle (UAV)based, ground-based) can provide a nondestructive way of spatially and temporally monitoring the targets of interest remotely, which has drawn increasing attention recently with the rapid development of autonomous system innovation, sensing technology and imageprocessing algorithms, especially with convolutional neural network-based deep-learning approaches. More importantly, the accurate early forecasting of crop disease outbreak by making use of the remote sensing data has attracted much attention in recent years [6,7,8]. We first compared the reflectance distributions of snow, cloud and background in the 12 spectral bands from the Sentinel-2 L2A product. Snow and cloud showed similar and relatively high reflectance values (median reflectance values are greater than 5000) in the first ten spectral bands; they both have high reflectance variations in these ten bands. B9, B1 and B2 are the three bands that have a relatively larger distribution difference between snow and cloud. Even though snow and cloud have very similar reflectance distributions in the first 10 bands, the cloud has a slightly higher median value than snow in nine out of the ten bands

Results

Discussion

Conclusion