Salinization Monitoring And Reclamation Strategy Analysis In Desert Soils Using Remote Sensing

Salinized soils are the main reserve land resource for potential cultivation in arid and semi-arid area.And it is of great importance to monitor the spatial distribution,salinization degree and spatial-temporal variation of salinized soils for their management and reclamation,as well as the ecological restoration of local environment.In this study,we selected Kongtailike,a classical alluvial fan in Wensu County,Southern Xinjiang,as the study area.To investigate the hyperspectral characteristics of soil salt content,and to monitor and map soil salinization in the top soil and soil profiles at different depths,we have used multi-source datasets?i.e.,satellite remote sensing images,digital elevation model?DEM?,field survey data,and indoor hyperspectral data?,in combination with a series of state-of-the-art modelling approaches,such as Partial Least Square Regression?PLSR?,Random Rorest?RF?,Cubist,and zonal modeling+PLSR/Cubist.The key results have been summarized as follows:?1?The response of the hyperspectral characteristics to the different salt contents in soil is the basis of soil salinization monitoring using satellite-based remote sensing technologies.Due to the differences in salt composition,the corresponding responses of hyperspectral characteristics were also different.In this study,we applied difference analysis method in order to investigate the hyperspectral characteristics of the soil salinities based on 14 artificially configured soil samples with different salt contents made from purified salt of soils in the study area,and 255 top soil samples collected in the field.Meanwhile,the hyperspectral data of the 255 top soil samples were simulated as multispectral data using the spectral response functions of the sensors born on the Landsat 8 satellite.Then,a series of modelling approaches?e.g.,PLSR,RF and Cubist?was used to construct the soil electric conductivity?EC?inversion models based on the above-mentioned multispectral data.The results suggested that:?a?different salt contents in soils could result in different spectral shapes and reflectance;?b?with the increasing of the salt content,the F values and the spectral angle?values between the two corresponding spectral curves also increase;and?c?soil salt content is positively correlated with soil reflectivity.Additionally,there are significant absorption and reflection characteristics.In terms of the salinized soil samples from the study area,for instance,the reflectance spectra of saline soils showed significant absorption peaks with wavelengths of 445 nm,510 nm,900 nm,1410 nm,1940 nm,2210 nm and 2330 nm,respectively,and showed significant reflection peaks with wavelength of 2130 nm.Considerable differences of predicting performance were observed between different inversion models,among the three models derived from PLSR,RF,and Cubist,the Cubist model has the best predicting performance,with R²,RMSE,MAE and RPD of 0.86,4.17 dS m^-1,3.97 dS m^-1 and3.09,respectively.?2?In the study area,the main controlling factors of salinization in the vegetation-covered area and the bare soil area are significantly different.And the relationships between main controlling factors and EC in these two regions are heterogeneous.The potential of acquiring main controlling factors of salinization based on remote-sensed monitoring is severely limited.In response to this problem,we have proposed zoning modeling methods to effectively improve the performance of soil salinization prediction,and the results demonstrated that the Cubist,zoning+PLSR,and zoning+Cubist models significantly outperformed the RF and PLSR global models.The application of zoning+Cubist was used to invert the soil EC distribution of different years in the study region from 1990 to 2017.The statistical results showed that the soil salinization have great spatial-temporal variabilities during different periods of time.Generally speaking,the degree of salinization in the southern part was obviously more serious than those in the northern part,however,the salinization degree in the southern region showed a significant decline trend since2010 due to human reclamation activities.Meanwhile,the degrees of salinization have significant differences at the annual scale,the most severe soil salinization occurred in 2007,while the degree of salinization in 1994 was relatively light.Additionally,the correlation between the average EC in the study area and the meteorological factors at different temporal scales?daily,ten-daily,monthly,seasonal,and yearly?was also calculated.The results showed that:?a?there was a negative correlation between precipitation and soil salinization;?b?temperature and evaporation were positively correlated with the soil salinization;and?c?the correlations between meteorological data and soil salinization at seasonal scale were stronger than those at daily,ten-daily,monthly and annual scales.?3?Significant heterogeneity in both horizontal and vertical distributions of soil salinity was observed,while the satellite-based remote sensing technology mainly provides spectral information of the surface soils,which makes it a great difficult to obtain the soil salinization information with different depth profiles,especially over large regions.Thus,a multi-source datasets?e.g.,DEM,land surface temperature,satellite-based multispectral data with multiple time phases?was used to construct Cubist inversion models of soil profiles at different depths.The results suggested that the maximum or minimum values of the long-term time-series spectral indices could be used to effectively alleviate the interference from causal factors?e.g.,droughts,and plant diseases and insect pests?on monitoring the soil salinization at different depths of profiles.Meanwhile,in combination of DEM and land surface temperature data,monitoring of soil salinization could be realized at different depth profiles,especially at large spatial scale.Cubist models provided the best prediction performance at different depth profiles,with R² of 0.80-0.90,RMSE of 2.97-5.08dS m^-1,MAE of 2.28-3.92 dS m^-1,and RPD of 2.10-3.10.?4?Quantitative assessments on leaching requirement are the significant basis for the formulation of salinized soil reclamation scheme;however,it is a great challenge to obtain the leaching requirement using remote sensing technology.In this study,satellite remote sensing technology was used to obtain the EC in soil profiles at different depths,and Legostatyev's empirical formula was also considered to calculate the leaching requirements for the entire study area.The results suggested that the estimated water consumption for salinized soil reclamation was approximately 34.01×10⁸-62.90×10⁸ m³,and the available surface runoff water resources?1.29×10⁸m³?could not meet the demands for the salinized soil reclamation.In addition,if all the salinized soils were planned to be reclaimed in a short period,a large amount of groundwater needs to be extracted,thereby causing ecological and environmental problems.Therefore,it is necessary to make reasonable plans for the reclamation and management of salinized soils.Considering the available water resources,crop water requirements,and the leaching effect on the quality of the Tarim River from the salinized soils in the process of the reclamation,and followed the principle as reclaiming the soils with light degree of salinization first,the recommanded reclamation strategy has been made.To ensure that the drainage of saline soil does not make the water of Tarim River unavailable for farmland irrigation,the salinity of the river could not exceed the upper limit of 2.0 g L^-1.Thus,within 1-2 years,approximately 1.98×10⁴ hm² of cotton,or 3.05×10⁴ hm² of corn,or 2.44×10⁴ hm²of wheat could be plant in the reclaimed soils.This study explored the hyperspectral characteristics of salinized soils in typical desert area of Southern Xinjiang,a set of indices which could be easily acquired using remote sensing technology was selected,and the comprehensive utilization of multi-source data?e.g.,DEM,land surface temperature,satellite-based multispectral data with multiple time phases?was carried out to realize the soil salinity estimation in soil profiles at different depths based on remote sensing.Aiming at solving the large-scale excessive reclamations of salinized soils in Southern Xinjiang,a reasonable reclamation strategy based on limited water resources was proposed.These results provided scientific solutions and ideas for remote sensing monitoring and rational utilization of salinized soils in desert area of Southern Xinjiang,and also provided reference for related research in other regions under similar situations.