Spectral Monitoring Of Nitrogen Nutrition And Recommendation Of Nitrogen Fertilizer In Drip-Irrigated Potatoes

Monitoring the nitrogen nutrition status of crops by using spectral information is an important research content of accurate and automatic management of nitrogen fertilizer in modern agriculture.How to ensure the accuracy of spectral monitoring and establish a reliable nitrogen fertilizer recommendation algorithm has become the key to achieve optimal management of nitrogen fertilizer in crop growth season,especially for intensive production systems with irrigation systems.The rational application of nitrogen fertilizer is not only related to the national food security,but also affects the ecological environment of the region.The objective of this study was to explore the response of 29hyperspectral index formulas and their band positions to estimate plant nitrogen concentration and biomass at the canopy scale based on a five-year field experiment with six potato varieties under drip irrigation at the northern foot of Yinshan Mountain in Inner Mongolia.The selected optimal spectral indices were used as the input variables of four machine learning algorithms,including random forest,partial least squares regression,support vector regression and article neural network,to establish an accurate and applicable monitoring model for potato aboveground plant nitrogen concentration and biomass in the key fertilization period of drip irrigation.Subsequently,these models were coupled with potato critical nitrogen concentration dilution curve,nitrogen balance principle and nitrogen absorption law.Under the concept of total amount control,stage distribution and process regulation,a model that can realize multiple recommendation of drip irrigation crop growth season was proposed,and a spectrum-based drip irrigation potato nitrogen recommendation algorithm was integrated.The invention provides an effective technical support for the non-destructive monitoring of nitrogen nutrition in the growth season of the drip irrigation potato and the integrated optimal management of water and nitrogen,and also provides a new method for the development of a new generation of crop nitrogen nutrition portable sensor and the analysis of space remote sensing information.The main results are as follows:1.There were significant differences in the spectral sensitive bands of nitrogen concentration and biomass in the aboveground of potato.Based on the spectral index,the sensitive bands of plant nitrogen concentration were mainly distributed in 340-400 nm ultraviolet light and 450-600 nm visible light,while the sensitive bands of biomass were mainly distributed in 900-1000 nm near-infrared region.2.The band position and the formulas significantly affect the performance of the spectral index.Compared with the published spectral indices,the optimization of bands and the selection of formulas improved the explanation ability of potato plant nitrogen concentration by 16%-71%and 3%-18%,respectively,and the explanation ability of potato aboveground biomass by 4%-73%and 5%-41%,respectively.Opt-CCCI was the most promising optimized spectral index for simple parametric regression models based on optimized spectral indices,explaining 67%and 68%of the variation in potato plant nitrogen concentration and biomass,respectively,on an independent validation dataset from a farmer’s field.The RMSE and RE of plant nitrogen concentration were 0.36%and 10.87%,respectively,and the aboveground biomass were 421.5 kg hm^-2and 21.6%,respectively.3.The type and quality of machine learning input variables significantly affect the performance of machine learning models,and optimized spectral index was a universal input variable for machine learning algorithms.Using multiple optimized spectral indices as the input variables of the random forest algorithm can further improve the prediction accuracy and robustness of the simple parametric regression model for the potato aboveground plant nitrogen concentration and biomass.The R²,RMSE and RE were 0.76,0.31%and 9.35%,respectively,in plant nitrogen concentration validation data,and 0.80,254.5 kg hm^-2and 13.5%,respectively,in aboveground biomass validation data.4.The proposed spectrum-based nitrogen recommendation algorithm for drip irrigation potato in this study can control the amount of nitrogen fertilizer in a reasonable range during the growing season of drip irrigation potato.Without considering the residual nitrate nitrogen in the soil before sowing,this algorithm can reduce the traditional nitrogen fertilizer application rate of farmers by 34.8%-47.3%and control the nitrogen surplus between-13 and 47 kg N hm^-2while ensuring the yield;When considering soil nitrate nitrogen as a nitrogen input source before sowing,the traditional nitrogen fertilizer application rate of farmers can be reduced by 61.8%-64.0%,and the nitrogen surplus can be controlled at 49-53 kg N hm^-2.Compared with other nitrogen recommendation algorithms,the proposed spectrum-based nitrogen recommendation algorithm for drip-irrigated potatoes in this study takes into account the change of nitrogen flow in soil-crop system,and the application of nitrogen fertilizer trade off the balance and stability of yield,environment and soil nitrogen pool,which is in line with the goal of intensive sustainable development and lays a foundation for intelligent and automatic nitrogen management of drip-irrigated potatoes in modern agriculture.