| Cultivated land is the essence of land,which is the most precious agricultural resource and the most important factor of production.cultivated land quality about national food security,agricultural product quality safety and agricultural ecological security,19 report pointed out that agriculture,rural areas,farmers is related to the affairs of the people’s livelihood,especially we as agricultural country,improving the quality of cultivated land is to promote the urgent needs of the food production and agricultural sustainable development.Like other objects,cultivated soil has its own rules of occurrence and evolution.With the continuous change of soil environment and the rapid development of remote sensing technology,higher requirements are put forward for the construction of high-precision cultivated land quality evaluation indexes and cultivated land degradation monitoring.High precision of cultivated land quality monitoring helps to provide the basis for cultivated land quality classification and grading,for the distribution of crop planting structure with fine agriculture,etc many fields or disciplines to provide the reference,conducive to the implementation of strict arable land protection system,strengthen and improve macroeconomic regulation and control,both for the development of land,or national food security strategy is of great significance.Cultivated land quality monitoring is one of the key sources of soil data in land resource management.Realization of high-precision grading of cultivated land quality is conducive to sustainable and efficient use of cultivated land resources.The traditional soil figure rely on national census data,using kriging interpolation mapping of soil properties,however,the implementation of soil census need to invest a lot of cost for collection and analysis of soil profile,including human,money and time cost,both on aging,and pixel/grid scale(time and space continuity)advantage is not strong,therefore,to achieve quantitative monitoring of soil properties needs to be a fast and accurate way.In this study,satellite spectral remote sensing technology and image noise reduction technology were used to compare the spatial identification and mapping of soil nutrients under different noise reduction methods in Mingshui County,a typical black soil region in Heilongjiang Province.Then,the relationship between cultivated land quality grading results and soil degradation was analyzed.In this study,a total of 58 Soil samples were collected in May 2016,May 2018 according to the same proportion of Soil type coverage area to test the content of Soil organic matter(SOM).The Gaofen-5 hyperspectral satellite remote sensing image data in the bare Soil period of 2019,30m spatial resolution SRTM-DEM data,and the second national Soil census data were used.Based on this,14 terrain factors and 39 spectral texture information are derived.Introduced four kinds of noise control way to singular value decomposition(SVD),median filtering(MF),Fourier transform(FT),discrete wavelet transform(DWT)and 2 kinds of spectral information extraction processing(fractional order derivative,spectral index),combined with random regression model to contain different environmental information(topographic factors,texture characteristics)of a study on inversion of soil organic matter content in the study area,analysis the influence of different environmental information for different types of soil mechanism;Then,the inversion results of hyperspectral satellites compared with spatial interpolation results,and on the basis of the evaluation of cultivated land quality classification and gradation,explore different Mingshui county area of arable land quality,at the same time,linked to land degradation,key analysis easily degraded area(such as soil salinization,soil erosion)SOM space distribution characteristics and quality of cultivated land grading,trying to reveal the causes of soil degradation and protection measures.The main research contents and results of this paper are as follows:(1)Noise reduction methods can effectively reduce the noise in hyperspectral remote sensing images.After denoising,the spectral curve is smoother,and the correlation between the reflectivity and SOM is improved obviously.Different noise reduction method and the correlation of SOM content from high to low in turn as the DWT,FT,SVD,MF,among them,the SVD,FT,DWT noise reduction method can improve the predictive accuracy of SOM based on spectral reflectance and SOM content after MF noise correlation is lower than the original reflectivity and the correlation of SOM content,different noise reduction method of SOM prediction accuracy of DWT>FT>SVD>MF.(2)Reflectance fractional derivative processing can express more spectral information,and improve the prediction accuracy of SOM.Compared with the SOM inversion accuracy based on original reflectance(R~2:0.58,RMSE:2.91 g/kg,RPIQ:1.66),the inversion accuracy based on 0.6derivative reflectance is the highest(R~2:0.71,RMSE:2.19 g/kg,RPIQ:2.21).Under different noise reduction methods,the SOM prediction accuracy after DWT noise reduction based on RF regression model is the highest(R~2:0.76,RMSE:2.22 g/kg,RPIQ:2.03).When 0.6FOD was combined with DWT and the decomposition scale was 2,the R~2 of RF was 0.84,RMSE was 1.82 g kg-1,and RPIQ was 2.75.(3)Recursive feature screening technology is an effective way of data compression and dimensionality reduction,that is,it eliminates the redundancy of hyperspectral information,while retaining the integrity of hyperspectral information and the physical significance of the original band.Considering the interaction of multiple categories of features,the RF function was selected as the basic function of feature selection.Multiple decision trees were established to comprehensively consider the influence of different feature combinations on SOM inversion,and the feature combinations with the highest inversion accuracy were selected through the’sbf’function.(4)In this study,the introduction of spectral index,terrain factor and texture feature can all improve the spatial SOM inversion accuracy.Among them,the terrain factor has the most significant effect on the improvement of SOM inversion accuracy.The R~2:0.89,RMSE:1.51g kg-1,RPIQ:3.22based on the"reflectivity+terrain factor".Considering the hyperspectral satellite spectral information and auxiliary variables,the SOM estimation accuracy is the highest,R~2:0.91,RMSE:1.49 g kg-1,RPIQ:3.57.The model can accurately realize the SOM content inversion in the study area.(5)Soil classification helps to improve the accuracy of satellite SOM inversion.Due to the influence of soil formation factors and the scale of the study area,the sensitivity of topographic factors and texture features to different soil types is different.In black soil,DEM,VD and RSP are important topographic factors.This is because black soil is mainly distributed in the north of the study area,where there is a large amount of meadow soil with low soil potential,so the elevation difference is obvious.At the same time,black soil erosion is serious,so VD becomes an important terrain index that black soil is different from other soil types.Meadow soil is developed in low-lying areas,such as depressions and gullies,etc.,which are relatively dispersed in the study area,and the soil boundary position has obvious difference in elevation.(6)The SOM inversion based on Gaofen-5 hyperspectral satellite image is significantly correlated with the SOM spatial distribution obtained by interpolation method under cultivated land quality monitoring.The SOM distribution of cultivated land quality monitoring under different methods is more helpful for the study of soil degradation and cultivated land quality evaluation.In this study,remote sensing technology was used to complete the inversion and mapping of soil attributes,and the spatial prediction ability of different noise reduction methods,environmental information and regression models for soil attributes was compared.The potential of hyperspectral data for soil research was exerted,and the accuracy of soil nutrient inversion and the expression of soil-forming factors were improved.Provide technical support for the soil data update,compared with the commonly used spatial interpolation results,explore the dynamic distribution of SOM content and potential relationship between cultivated land quality degradation,research results can provide reference for the cultivated land classification and grading,to improve the efficiency of agricultural production and land the implementation of the sustainable development strategy is of great significance. |
