Satellite-UAV-ground Integrated Inversion Method Of Soil Salinity Of Main Crops In Coastal Area Of The Yellow River Delta

Soil salinization has become a global issue,and it is a major factor affecting crop development and yield.The Yellow River Delta is one of China’s three biggest deltas;its abundant natural resources and large land area make it an important reserve land resource in China.Soil salinization is one of the key constraints limiting agricultural development in this region,and quantitative,accurate,and rapid acquisition of salt content and geographic distribution information of regional saline soil is the basis of its control and exploitation.Satellite remote sensing provides good technical support for gathering large-scale soil salinization information.However,the low spatial resolution and imaging quality are inadequate,making high-precision real-time monitoring impossible.UAV technology can obtain images with high temporal and spatial resolution,but the observation range is limited which makes it difficult to monitor large areas.Ground hyperspectrum can build high-precision inversion models,while the point-like information is difficult to monitor in a continuous spatial range.It can be seen that the data of satellite,UAV and ground platform all have their own advantages and disadvantages,and the surface information acquired by a single remote sensing system is frequently insufficient.Therefore,it is necessary to fully exploit the potential of multi-source remote sensing data,complement each other’s advantages,and realize the integrated remote sensing data system of satellite,UAV and ground platform.On the basis of clarifying the approaches of satellite-UAV-ground integration,this study aims to explore the key links and methods of satellite-UAV-ground integration from the aspects of platform data fusion,quantitative remote sensing modeling inversion,upscaling,and satellite-UAV collaboration.Then aiming at the coastal area of the Yellow River Delta in Kenli District,using Sentinel-2A satellite images,UAV images in the test areas,ground imaging hyperspectrum and ground measured salinity as data sources,using UAV-ground inversion and satellite-UAV collaboration,UAV-ground fusion and model upscaling,satellite-ground fusion and satellite-UAV collaboration integration have achieved soil salinity inversion in winter wheat,corn and cotton planting regions in the study areas,respectively.The following are the key research contents and findings:（1）The major ways,crucial links,and methodologies of satellite-UAV-ground integration were proposed.There are seven main approaches for the integration of satellite,UAV and ground,which are:UAV-ground inversion and satellite-UAV collaboration,UAV-ground modeling and satellite inversion,UAV-ground fusion and satellite-UAV collaboration,UAV-ground fusion and satellite inversion,satellite-UAV fusion and satellite-ground inversion,UAV-ground inversion and satellite-ground fusion,UAV-ground modeling and satellite-ground fusion.The following are the key link of the integration of satellite-UAV-ground integration:data fusion of the ground,UAV and satellite,modeling and inversion of the UAV-ground and satellite-ground,upscaling and coordinated inversion of the satellite-ground and satellite-UAV.The key methods include remote sensing data fusion,quantitative remote sensing modeling,and upscaling methodologies.Three representative,efficient and feasible methods of satellite-UAV-ground integration were proposed,which were UAV-ground inversion and satellite-UAV collaborative integration,satellite-UAV fusion and model upscaling integration,satellite-ground fusion and satellite-UAV collaborative integration.（2）Soil salinity monitoring in wheat fields based on UAV-ground inversion and satellite-UAV collaboration integration.To perform UAV-ground modeling,the machine learning approaches were applied,and the best model was chosen for inversion of the test areas.The upscaling method was investigated using the pixel correspondence of different sizes of the satellite and UAV.The UAV-scale soil salinity was pushed up to the satellite scale,and the satellite-UAV collaborative inversion model was established to realize the satellite-UAV-ground integrated inversion of winter wheat soil salinity.The findings revealed that the random forests model exhibited the best fitting accuracy（R~2=0.878）,and the inversion findings were quite consistent with the real soil salinity distribution.Upscaling enhances the accuracy of the measure salinity value associated with satellite pixels.The satellite-UAV collaboration’s random forests inversion model was the best,with R~2=0.885.The R~2is 0.15 higher when compared to the inversion findings of the model built directly from ground data and Sentinel-2A image.The satellite-UAV-ground integration method can significantly improve the accuracy of soil salinity inversion results and yield a more precise geographical distribution of soil salinity in winter wheat fields.The soil in the study area’s winter wheat planting area is dominated by mild and moderate salinization,accounting for80.85%of the area,which is primarily dispersed in the southwest and northeastern regions.The area of severe salinization and saline soil is smaller,accounting for 19.1%of the total,and is dispersed throughout the wheat area.（3）Soil salinity monitoring in corn fields based on UAV-ground fusion and model upscaling integration.The nonlinear UAV-ground fusion method was studied based on the UAV multi-spectrum and ground imaging hyperspectrum.A high-precision inversion model was constructed based on the fused UAV images,and the UAV scale model was pushed up to the satellite scale according to the upscaling algorithm,realizing the satellite-UAV-ground integrated inversion of soil salinity in corn field.The results revealed that fusing UAV-ground bands efficiently enriched the spectral information of UAV images and improved the correlation with soil salinity.The accuracy of the partial least squares inversion model based on fused UAV images was R~2=0.743,which was 0.11 greater than the accuracy of the model built directly on UAV images.The inversion results were highly consistent with the measured soil salinity distribution,and 84.6%of the samples were consistent with the measured soil salinity grade.The partial least squares inversion model of soil salinity at the Sentinel-2A satellite scale is S₁₀=6.214-6.336NDVI-7.588DVI+0.131GRVI after upscaling,and thus we obtained the soil salinity grade distribution of corn fields in the study area.The results of soil salinity inversion based on satellite-UAV-ground integration were in good agreement with measured soil salinity R~2=0.716 and the inversion model had good universality.The distribution of non-salinization in the study area is less,with mild and moderate salinization accounting for 88.36%of the total area,concentrated in the southwest and north-central areas,while the area of severe salinization and saline soil is small and scattered in corn growing area.（4）Soil salinity monitoring in cotton fields based on satellite-ground fusion and satellite-UAV collaborative integration.UAV-ground modeling was realized by using machine learning methods;we selected the optimal model for inversion.The differential fusion method of satellite multispectral and ground imaging hyperspectral based on vegetation index is studied.The satellite-UAV-ground integration inversion was performed using the optimal inversion results of the UAV as training samples for the fusion satellite to construct the inversion model.The results reveal that the satellite spectrum after satellite-ground fusion was closer to the original hyperspectrum of the ground,which effectively enriched the spectral information of Sentinel-2A image and improved the correlation with soil salinity.The XGBoost inversion model based on UAV image was the best,and the inversion results were in good agreement with the actual distribution of soil salinity in the test area,therefore it was utilized as training samples for constructing the convolutional neural network inversion model based on fused satellite images.The soil salinity obtained by the satellite-UAV-ground integration method was highly consistent with the soil salinity grade distribution obtained by UAV inversion in the verification area,and had a high consistency with the measured salinity in the study area,with R~2=0.805,which was better than the R~2=0.613 obtained directly by satellite inversion.In addition,the integrated inversion model had good universality.The cotton soil in the study area is mainly moderate and severe salinization,accounting for69.43%of the total area,scattered distribution,relatively concentrated in the central and southwest regions,the area of non-salinization,mild salinization and saline soil is small.This research systematically analyzes and proposes the approaches,key links and methods of satellite-UAV-ground integration,and respectively constructs the monitoring method for the integration of satellite-UAV-ground in the main crop planting areas of the study area.It improves the theoretical system of satellite-UAV-ground integration,realizes three-dimensional fusion of satellite,UAV and ground remote sensing data,fully exploits the advantages of remote sensing data from different platforms,improves the fine expression ability of remote sensing technology for large-scale soil salinity,and provides more effective technical support for agricultural production in coastal areas.