| Nitrogen(N)is one of the important indexes to maintain the nutrient elements necessary for the growth of apple trees and to evaluate the fertility of the field.The amount of soil nitrogen content directly affects the growth and development of fruit trees and fruit quality.Statistics show that the nitrogen fertilizer applied per unit area in China is about 2.5 times that in the United States.In order to increase yield,fruit farmers blindly fertilize,which not only leads to low nitrogen fertilizer utilization and waste of resources,but also causes land and water pollution,which affects the sustainable use of gardens.Therefore,timely and accurate monitoring of nitrogen content in orchard soil is of great significance for scientific fertilization and accurate management of orchard quality.Therefore,timely and accurate monitoring of nitrogen content in orchard soil is of great significance for scientific fertilization and accurate management of orchard quality.The traditional method of sampling and analysis will damage the roots of fruit trees to some extent,and the analysis cost is high and the time is long,so the application in orchard management is greatly limited.The development of spectral estimation techniques for soil physical and chemical properties has brought new approaches to orchard soil monitoring and management,and research on spectral estimation of total nitrogen(TN)in soil has also achieved many results.However,apple orchards are mostly distributed in mountainous and hilly areas.Under the influence of geography,climate,soil texture and other factors,the soil nitrogen content of apple orchards is low,the spectral signal is weak,it is easy to be interfered or masked,and the effective spectral information is difficult to obtain.In view of these problems,this study took an apple garden in shuangquan town,changqing district,jinan city,shandong province,as the experimental area,and collect soil samples and determine soil TN content 10 days before the four fertilization periods of the young fruit period,fruit expansion period,quality period and confinement period.Use ASD to collect indoor spectral data of soil samples,analyze soil TN spectral characteristics at different fertilization sampling periods,and perform 8 kinds of spectral transformation and continuous wavelet transform processingon soil reflectivity R,and based on the whole wavelength range spectrum characteristic parameter selection,using correlation coefficient method and stepwise regression analysis(SMLR)method to extract the spectral characteristics of soil TN band,The spectral estimation model was constructed based on MLR and MEA-BP,and the spectral estimation process of orchard soil TN was optimized.The main results are as follows:(1)The characteristic spectrum range of total nitrogen in apple orchard soil was determinedBased on the spectral data of different fertilization periods,the soil original spectrum R and its 8 transformation forms were screened by the correlation coefficient method and the multiple step regression(SMLR)method for the characteristic bands of soil TN content.In general,the spectroscopically sensitive areas of soil TN were mainly concentrated in the bands of 500-900 nm,1400-1500 nm,1900-1950 nm and 2200-2400 nm,among which the bands of 562 nm,706nm,808 nm,1933nm and 2345 nm in the whole fertilization period were significantly affected by the content of soil TN;the bands of 541 nm,808nm,1423 nm,1462nm and 2423 nm in young fruit fertilization periodwere significantly affected;the bands of 554 nm,812nm,1931 nm and 2301 nm in fruit expansion fertilization were significantly affected in the swelling stage;the bands of 556 nm,826nm,2041 nm,2313nm and 2450 nm in quality fertilizer period were significantly affected;the bands of 562 nm,809nm,858 nm,1912nm and 2269 nm in lili fertilizer period were significantly affected.(2)The method of selecting soil characteristic spectral index of apple orchard was establishedThree common spectral indexes,RSI,DI and NDSI,were selected,and the correlation between all spectral indexes and soil TN content in the range of 400-2450 nm band reflected by the original spectrum of soil was calculated by using two-two combination method of allband matrix.It was determined that the soil TN content sensitive areas of the spectral index did not change with the sampling period.The sensitive areas of the ratio spectral index(RSI)were located in the bands of 800-900 nm,1900-1950 nm and 2200-2300 nm,and the sensitive areas of the difference spectral index(DI)and the normalized spectral parameter(NDSI)were located in the bands of 1900-1950 nm and 2200-2300 nm.(3)The selection method of characteristic spectral segment of TN content in soil and the modeling effect of spectral index correction model.For different fertilization periods,MLR and mea-bp neural network estimation models of soil TN content are constructed based on characteristic band and characteristic spectral index,and the model prediction effect is better based on the characteristic band selected by smlr method.The addition of spectral index can further improve the modeling accuracy of the model,including young fruit period,expanding fruit period,quality period,monthly sub period and the Unified Soil t throughout the year Cr-sci-mea-bp neural network is the best model to estimate N content.(4)Application of continuous wavelet transform to construct independent and Unified Soil TN spectral estimation model in different fertilization periods of orchard.Based on the original reflectance of soil and its 8 kinds of transformations,the continuous wavelet transform of 10 scales was carried out for 9 kinds of spectra by gaus4 function.The optimal decomposition scale was determined by smlr method and the wavelet coefficients were extracted.The MLR and mea-bp neural network models of soil TN in different fertilization periods and the whole year were constructed respectively based on the wavelet coefficients and characteristic spectral parameters of the optimal scale.The results showed that the coupling of traditional spectral transform and continuous wavelet transform could improve the sensitivity of Soil Spectrum to soil TN,and the accuracy of the model was also improved.Considering the decision coefficient and prediction effect of the modeling,it was determined that the independent and unified estimation model of soil TN content in each fertilization period of the orchard was based on cr-cwt-sci-mea-bp model,and the prediction accuracy of the five models was 0.95,0.96,0.95,0.95 and 0.91 respectively,RMSE was0.0027,0.0023,0.0029,0.0026 and 0.0031,respectively.(5)A spectral estimation procedure for soil TN content was proposedBased on the needs of rapid and nondestructive monitoring of TN content in orchard soil,the key technologies of spectral estimation used in the study were analyzed and compared,forming a complete process.Firstly,the original spectral r of soil was coupled with continuous removal(CR)and continuous wavelet(CWT),and the characteristic wavelet coefficients of the 5th scale were screened by smlr method;then 800-900 nm,1900-1950 nmand 2200-23 were carried out At last,based on the characteristic wavelet coefficient and the characteristic spectral index,the spectral estimation model of soil TN content was constructed by mea-bp neural network.At present,the production management of apple orchards in China is in the alternate stage of new and old production capacity,and some orchards are gradually changing from traditional extensive management to precise and dynamic management.In this study,soil hyperspectral technology was used to explore the corresponding spectral characteristics of TN in orchard soil at different fertilization periods.The estimation model was established,and the unified technical process of rapid estimation of TN in orchard soil at different periods and throughout the year was optimized,which provided technical support for the digital nondestructive diagnosis of TN in apple orchard soil and precise dynamic management of orchard quality in the future. |
PDF Full Text Request