干旱区生态环境知识资源中心(Arid): Comparison of bagging, boosting and stacking algorithms for surface soil moisture mapping using optical-thermal-microwave remote sensing synergies

Arid

DOI	10.1016/j.catena.2022.106485
	Comparison of bagging, boosting and stacking algorithms for surface soil moisture mapping using optical-thermal-microwave remote sensing synergies
	Das, Bappa; Rathore, Pooja; Roy, Debasish; Chakraborty, Debashis; Jatav, Raghuveer Singh; Sethi, Deepak; Kumar, Praveen
通讯作者	Chakraborty, D
来源期刊	CATENA
ISSN	0341-8162
EISSN	1872-6887
出版年	2022
卷号	217
英文摘要	Soil moisture information is key to irrigation water management, drought monitoring, and yield prediction. It plays a vital role in the water cycle and energy budget between the earth's surface and atmosphere. Hence, its monitoring is crucial for both natural and anthropogenic environments. While the current remote sensing-based global SM products available at coarser resolution (3/15 km) are unsuitable for field-level operations, the most widely used microwave remote sensing suffers from model complexities and in-situ data requirements. Weather conditions limit the alternate approaches such as optical/thermal. This study aims to map surface soil moisture (SSM) at 30 m spatial resolution in a semi-arid region by fusing optical, thermal, and microwave remote sensing data using bagging, boosting, and stacking machine learning approaches. The reference data were collected using a soil moisture meter. The covariates included radar backscatter from Sentinel-1, visible, near-infrared, shortwave infrared, land surface temperature, and spectral indices derived from Landsat 8. Boruta algorithm was used for feature selection which identified radar backscatter, modified normalized difference water index, and land surface temperature as the most critical covariates impacting the SSM. The random forest (RF) showed the highest correlation coefficient (r = 0.71), and least root mean square error (RMSE = 5.17%). The cubist model had the least mean bias error (MBE = 0.21%) during independent validation. Stacking of cubist, gradient boosting machine (GBM), and RF using elastic net (ELNET) as meta-learner further reduced the MBE (0.18%) and RMSE (5.03%) during the validation. Overall, stacking multiple machine learning models improved model prediction and can be recommended to improve the digital soil moisture mapping.
英文关键词	Machine learning Stacking Soil moisture Sentinel-1 Landsat-8
类型	Article
语种	英语
收录类别	SCI-E
WOS记录号	WOS:000822972200004
WOS关键词	LANDSAT ; WATER ; RETRIEVAL ; SENTINEL-1 ; RESOLUTION ; MODELS ; REGION ; FOREST ; INDEX ; AREA
WOS类目	Geosciences, Multidisciplinary ; Soil Science ; Water Resources
WOS研究方向	Geology ; Agriculture ; Water Resources
资源类型	期刊论文
条目标识符	http://119.78.100.177/qdio/handle/2XILL650/392114
推荐引用方式 GB/T 7714	Das, Bappa,Rathore, Pooja,Roy, Debasish,et al. Comparison of bagging, boosting and stacking algorithms for surface soil moisture mapping using optical-thermal-microwave remote sensing synergies[J],2022,217.
APA	Das, Bappa.,Rathore, Pooja.,Roy, Debasish.,Chakraborty, Debashis.,Jatav, Raghuveer Singh.,...&Kumar, Praveen.(2022).Comparison of bagging, boosting and stacking algorithms for surface soil moisture mapping using optical-thermal-microwave remote sensing synergies.CATENA,217.
MLA	Das, Bappa,et al."Comparison of bagging, boosting and stacking algorithms for surface soil moisture mapping using optical-thermal-microwave remote sensing synergies".CATENA 217(2022).