Soil salinity monitoring model based on the synergistic construction of ground‐UAV‐satellite data

• Published in: Soil use and management Vol. 40; no. 1
• Main Authors: Jia, Jiangdong, Chen, Ce, Liu, Qi, Ding, Binbin, Ren, Zheng, Jia, Yanxin, Bai, Xuqian, Du, Ruiqi, Chen, Qinda, Wang, Shuang, Luo, Linyu, Zhang, Zhitao, Geng, Hongsuo
• Format: Journal Article
• Language: English
• Published: Bedfordshire Wiley Subscription Services, Inc 01.01.2024
• Subjects: Accuracy; administrative management; Agricultural development; Algorithms; China; Environmental monitoring; Field tests; irrigation; Machine learning; machine learning algorithm; Modelling; Monitoring; Remote sensing; Salinity; Salinity data; Salinity effects; Salinization; satellite; Satellite imagery; Satellites; Soil salinity; soil salinity monitoring; Soil salinization; Soils; Spectral bands; Sustainable development; UAV; Unmanned aerial vehicles; upscaling; vegetation; China
• ISSN: 0266-0032; 1475-2743
• DOI: 10.1111/sum.12980

Soil salinization poses a significant constraint on the sustainable development of agriculture. While satellite remote‐sensing data enables salinity monitoring over large spatial scales, its coarse resolution limits monitoring accuracy. On the other hand, unmanned aerial vehicle (UAV) remote‐sensing data offers greater accuracy in salinity monitoring but covers a smaller area compared to satellite remote sensing. To address the need for both high precision and wide‐range monitoring, we developed a soil salinity monitoring model integrating ground, UAV, and satellite data. Field experiments were conducted in the Shahaoqu irrigation area, Inner Mongolia, China, from August 11 to 15, 2019. During this period, we collected satellite remote‐sensing images, UAV remote‐sensing images, and soil salinity data. Spectral bands from the remote‐sensing data were utilized to construct separate vegetation and salinity indices, which were further filtered using the variable importance in projection (VIP) algorithm. The soil salinity monitoring model was then constructed using the extreme learning machine (ELM) algorithm. Several soil salinity monitoring models were developed, including SSMM‐UAV (based on ground‐UAV data at a resolution of 6.5 cm), SSMM‐UAV‐upscaling (obtained by upscaling the results of SSMM‐UAV to a 16 m scale), SSMM‐satellite (based on ground‐satellite data at a 16 m scale), and SSMM‐UAV‐satellite (constructed using ground‐UAV‐satellite data at a 16 m scale). The results revealed that SSMM‐UAV accurately monitored soil salinity at the UAV scale, with R2 values exceeding .81 and RMSEs below 0.11% for both the model modelling set and validation set. SSMM‐UAV‐upscaling demonstrated consistency with SSMM‐UAV and effectively represented salinity conditions at the 16 m scale. In contrast, SSMM‐satellite exhibited inferior performance, with R2 values of .42 and .32 for the modelling set and validation set, respectively, and RMSEs of 0.15% and 0.17%, respectively. By incorporating ground‐UAV‐satellite data, SSMM‐UAV‐satellite improved the R2 of SSMM‐satellite by more than .09 and reduced the RMSEs by at least 0.08%. Furthermore, the area covered by satellite data was ca. 85 times larger than that covered by UAV data. The synergistic use of UAV and satellite data in salinity monitoring enhances the accuracy of satellite remote sensing and expands the monitoring range of UAV remote sensing. The findings of this study provide a reference for high precision and large‐scale salt monitoring through the synergistic integration of ground, UAV, and satellite data.