Hyperspectral Remote Sensing Estimation Of Soil Nutrients In Alpine Grassland

As one of China's three grassland livestock husbandry areas,the Qinghai-Tibet Plateau's special climatic conditions and regional conditions cause grassland ecosystem imbalances and ecological environment deterioration,especially the deterioration of the alpine grassland ecosystem.Rapid,accurate and efficient monitoring of vegetation growth and soil degradation status in alpine grassland is of great significance to effectively curbing the deterioration of the grassland ecological environment and sustaining the sustainable development of grassland resources.At present,with the rapid development of hyperspectral techniques,many laboratories have achieved accurate inversion based on vegetation growth information and soil nutrient content inversion models established by ground hyperspectral.However,with the combination of satellite remote sensing and breakthroughs in the limitation of research scales,accurate retrieval of soil information at large spatial scales is still the focus and difficulty of research.Based on the actual needs of obtaining information on the growth and soil information of alpine grassland in the Qiangtang Plateau,the study focused on the GPS survey,spectral measurement of the alpine grassland,and field conditions in Stipa purpureus in Shenza County,northern Tibet.Soil sampling,around the spectral index extraction of alpine grassland based on hyperspectral data,the remote sensing inversion and monitoring model of soil elements based on alpine grass canopy hyperspectral and HJ-1A satellite HSI hyperspectral imagery were studied.The main research conclusions are:(1)Analysis of canopy hyperspectral characteristics and soil nutrient response bands based on alpine grasslandThe special environment of the Qinghai-Tibet Plateau made the plants of alpine grassland short and small,and there were few leaves.The“peak and valley”features of the green vegetation spectrum were not obvious.After the Savitzky-golay method was used for convolution smoothing,first-order differential processing and de-envelope processing,the reflection and absorption characteristics of the prominent spectral curve were enhanced.The first-order differential mean curve of the soil reflectivity and the peaks and valleys of the continuum curve are more obvious,which shows that the characteristics of special bands are more prominent.With the increase of grassland coverage in alpine grassland,the albedo spectral reflectance curve,the first derivative curve and the peaks and valleys of the continuum curve are more obvious,and the chlorophyll absorption bands at 450 nm and 670 nm and the chlorophyll reflection at 550 nm are more obvious.The band and the moisture absorption band at 1450 nm show obvious features.There is also a certain degree of moisture absorption valley characteristics at 960 nm and 1160 nm in the near-infrared region.The curve of correlation coefficient between soil organic carbon,total nitrogen,total potassium,total phosphorus and canopy spectral reflectance and first-order differential value of vegetation in the alpine grassland were basically the same,and organic carbon>total nitrogen>total phosphorus>total potassium.In the range of400-720 nm,the spectral reflectance of alpine grassland was negatively correlated with soil organic carbon,total nitrogen,total potassium,and total phosphorus;and the correlation was highest between 670-680 nm;in the 720-1400 nm range,the alpine grassland vegetation Canopy spectral reflectance was positively correlated with soil organic carbon,total nitrogen,total potassium,and total phosphorus.After the first-order differential treatment,the correlation has been significantly improved.The maximum correlation coefficient of organic carbon,total nitrogen,total potassium,and total phosphorus has been greatly improved.The correlation coefficient curves fluctuate drastically,oscillating violently between extremely significant positive correlations and extremely significant negative correlations,and the bands of significant correlations are more refined and differentiate into narrower bands.(2)Construction of response model of spectral index of alpine grassland to soil nutrientsIn order to better show the relationship between the selected 21 categories of spectral indices,the correlation coefficients between the spectral indices were calculated,and it was found that the correlation between the spectral indices was relatively high,that is,there was a relatively serious autocorrelation in the model independent variables.Therefore,multivariate stepwise regression and partial least squares regression were used to establish the model.The Pearson correlation analysis was performed on the spectral index of 21 types of alpine grassland and the measured soil chemical elements.Overall,organic carbon>total nitrogen>total phosphorus>total potassium.The maximum positive correlation between soil and spectral EVI was the most negative correlation with SD_y.In the multivariate stepwise regression model established,the organic carbon model has the highest coefficient of determination R~2,which is 0.56,followed by the determination coefficient of total phosphorus,R~2 is0.529,and the R~2 of total nitrogen is 0.483,both of which are related by the property level?=0.001 test.Due to multiple autocorrelation in the independent variable,the model retains fewer independent variables.In the partial least-squares regression model established,the R~2 of the total phosphorus model was the highest,which was0.602,the R~2 of the organic carbon model was the second,it was 0.576,the R~2 of the total nitrogen model was 0.46,and the correlation of the alleviation level?=0.001.test.Extract the number of total phosphorus>organic carbon>total nitrogen.Comparing and verifying the two models,it is found that the partial least-squares regression model is better than the multivariate stepwise regression in terms of the utilization of the independent variable information,model determination coefficient,model prediction accuracy,and error analysis.The relevance,stability,predictability,and reliability.Among them,the total phosphorus model has the best lifting effect,followed by total nitrogen and organic carbon,and the total potassium model has poor results.(3)Construction of soil nutrient space estimation model based on HJ-1A hyperspectral dataIn this study,the global stripping method was used to remove the strips of remote sensing images,and after FLAASH atmospheric correction,the analysis was performed on the vegetation of any two spectral bands in the range of 0.45-0.95 um and calculated in the matrix.The index was sensitive to soil organic carbon,total nitrogen,total potassium,and total phosphorus.The optimal near-infrared and infrared bands for various vegetation indices corresponding to soil organic carbon,total nitrogen,total potassium,and total phosphorus were selected.Among the spectral characteristics of alpine grassland,only SD_y has a significant correlation with soil organic matter,total nitrogen,and total phosphorus,while only TVI index has no significant correlation with soil organic carbon,total nitrogen,and total phosphorus in the vegetation index.In general,the spectral index based on HJ-1A hyperspectral data has the greatest correlation with total potassium,almost the correlation with organic carbon and total nitrogen,and the correlation with total phosphorus is slightly poor.From the goodness of fit of the model,organic carbon>total phosphorus>total nitrogen>total potassium;from the perspective of the predictability and stability of the model,total phosphorus>total potassium>organic carbon>total nitrogen;reliability from the model For example,total phosphorus>organic carbon>total potassium>total nitrogen.Based on the statistical mean of the reference data sets in different types of soil unit polygons,it was found that the predicted values of soil organic carbon,total nitrogen,total potassium,and total phosphorus were only similar to the reference values of some soil types,and existed in the soil element range as a whole.A certain amount of deviation.Through qualitative analysis,it was found that the results of the model can still largely conform to the relationship between different types of soil organic carbon,total nitrogen,total potassium,and total phosphorus.