Hyperspectral Response Characteristics Of The Main Attributes Of Topsoil In Gannan Navel Orange Orchards

Monitoring soil conditions is important to guide field production and improve crop yields.Achieving rapid determination of soil physical and chemical properties allows for more efficient monitoring of soil conditions.Traditional sampling and survey methods have problems such as slow detection speed,low accuracy,and small range,and require a large amount of human and material resources.The use of hyperspectral technology can achieve accurate and rapid monitoring of soil physical and chemical properties,which plays an important role in promoting the development of precision agriculture.The research area of this study is Xunwu County,Ganzhou City,Jiangxi Province.SOM content,p H value,Clay content and soil spectral reflectance were measured after field sampling.This study investigates the form and degree of influence of SOM content,p H value,Clay content and soil moisture content on the spectral reflectance of surface soil in Gannan navel orange orchards.The correlations between the post-treatment spectra,the spectra under different particles sizes and different moisture content with SOM content,p H value,Clay content were analyzed.After dividing the training and validation sets using the K-S method,the differences in the effectiveness of hyperspectral modeling under the partial least squares regression method and the BP neural network method were analyzed from different perspectives（spectral preprocessing methods,modeling full/sensitive bands,different soil grain sizes,different soil types,etc.）.This is done to provide support for further selection of the optimal modeling path.The main conclusions of the study are as follows:（1）The soil organic matter content in the study area has a large variation,and its variation range is 10.02-45.49 g·kg^-1.According to the nutrient grading standard of the second national soil census,the soil organic matter content belongs to the third level.Most of the soils in the study area are acidic,and a small part of them are neutral.Soil clay content has a large variation,and its variation range is 8.52-55.65%.The texture type is mainly clay loam,and a small part of the texture type is silty clay loam,silty clay,silty loam and loam.Some of the samples are at the junction of different texture types,and it is difficult to classify their texture types.（2）The absolute values of correlation coefficients of soil spectral reflectance and organic matter,p H,and clay content are maximized as the differential order becomes larger and show a general trend of increasing and then decreasing.Fractional order differencing can improve the correlation between soil spectra and soil physical and chemical properties,but the differential order is not necessarily large or small,and there is a certain optimal range for the selection of differential order.After fractional order differencing,the correlation between soil spectra and soil physicochemical properties can be enhanced in some bands.（3）The sensitivity bands of soil spectra for organic matter were more similar but slightly different for the same soil type under different pretreatments.The organic matter sensitive bands of all soil samples were basically the same as those of rice soil and red soil spectra superimposed on each other.The sensitive bands of rice soil spectrum for organic matter were mainly concentrated around 400-600 nm,1400 nm,1900 nm,and 2200-2450 nm;the sensitive bands of red soil spectrum for organic matter were mainly concentrated around 400-800 nm,1400 nm,1900 nm,and 2200-2450 nm.The sensitive bands of soil spectra for p H are mainly concentrated around 400-1200 nm and 1900-2450 nm.The sensitive bands of soil spectra for clay content are similar to those for organic matter,mainly around 400-600 nm,900-1000 nm,1900 nm,and 2200-2450 nm.The sensitive bands of soil spectra for organic matter at different particle sizes were mainly concentrated around 400-900 nm,1400 nm,and 1900-2450 nm.The sensitivity bands of soil spectra for p H varied according to soil types.The sensitivity bands of rice soil for p H were mainly concentrated around 400-800 nm,900-1000 nm,1600-1800 nm,2200 nm and 2400 nm,and the sensitivity bands of red soil for p H were mainly concentrated around 400-1200 nm and 1900 nm.The sensitive bands of soil spectrum for clay content are mainly concentrated around 400-800 nm and 1900-2400 nm.（4）In the PLSR and BPNN prediction models constructed using full-band spectra and sensitive-band spectra,the prediction models constructed using sensitive-band spectra were sometimes better and sometimes worse than those constructed using full-band spectra under different pretreatments.In addition,the prediction accuracy of the model can be improved by using fractional order differentiation compared with the original band,but the prediction accuracy of the model varies under different pretreatments,and there is no certain order of fractional order differentiation that shows excellent prediction results in different models.In the organic matter content prediction model,the best prediction results among all samples,rice soil samples and red soil samples under different pretreatments were 0.75 order,1.0 order and2.0 order fractional order differential treatment respectively,and the R² of the validation set was 0.77,0.85 and 0.67 respectively;the best prediction results among all samples,rice soil samples and red soil samples under different particle sizes were The models were constructed under 0.5 mm,2 mm,and 2 mm,and the R² of the validation set were 0.73,0.78,and 0.67,respectively.In the p H prediction models,in general,the common performance of PLSR and BPNN models modeling under different pretreatments,different soil types,full band or sensitive band was that the modeling effect of p H was not satisfactory,and there was no significant difference in their prediction accuracy.Among the Clay content prediction models,the best results among all samples were obtained using the PLSR model in full band modeling under 2.0 fractional order differentiation treatment with R² of 0.72,RMSE of 4.42 and RPD of2.02 in the validation set,with very good model prediction ability.The best result in the rice soil sample using BPNN model is full-band modeling under fractional order differentiation of1.25,with R² of 0.39,RMSE of 2.79 and RPD of 1.88 in the validation set,which has good model prediction ability.The best results in the red soil samples were modeled using PLSR model in sensitive bands under 0.75 order fractional order differentiation treatment,with R² of0.80,RMSE of 2.39,and RPD of 2.32 in the validation set,and the model prediction ability was very good.