Estimation Of Nitrogen Uptake By Potato Plants Based On Machine Learning Algorithm Combined With Multispectral Satellite Simulation Data

Nitrogen is one of the essential macroelements for potato growth and development,and has a great impact on potato yield and quality.In recent years,the demand for food is increasing with the increase of population,and potato has become the fourth largest food crop after rice,wheat and corn in China.Potato yield can be promoted in appropriate amount of fertilizer.However,unreasonable fertilizer input may lead to increased production costs and unnecessary environmental pollution because farmers are too extensive in conventional fertilization methods,and nitrogen uptake by potato is phased in different growth periods.Therefore,a reasonable fertilization plan is the top priority of modern agricultural development.In recent years,with the development of science and technology,remote sensing monitoring method is widely used in crop nutrient content monitoring based on high efficiency,non-destructive and real-time characteristics.The remote sensing estimation model based on the ground layer is not enough to realize large-scale monitoring for increased planting area,so a new method is provided for large-scale monitoring at the regional scale with the progress of satellite technology.According to ground-measured hyperspectral data,this research simulated sensor data of the typical multispectral satellites Sentinel-2A,GF-2,and Rapideye launched in recent years by resampling.Then the optimal bands of different spectral indices were screened by the band optimization algorithm,and the simulated bands and optimized spectral indices were combined with three machine learning algorithms respectively to establish an estimation model of nitrogen uptake by potato plants.Thus the estimation performance of each sensor in different phenological periods and the estimation ability of different machine learning algorithms were discussed.The optimal spectral index of simulated sentinel-2A satellite sensor is BNI2,the correlation coefficient is 0.51,and the sensitive bands are red edge band 3,near infrared band 1 and near infrared band 2.The optimal spectral index of simulated GF-2 satellite sensor is MCARI and TCARI,the correlation coefficient is 0.45,and the sensitive band is green light band,red light band and near infrared band.The optimal spectral index estimated by simulated Rapideye satellite sensor is MCARI and TCARI,and the correlation coefficient is 0.45.The sensitive band is red light band,red edge band and near infrared band.Then,the estimation models of nitrogen uptake of potato plants in different phenological periods were established by combining simulated band and optimized spectral index with three machine learning algorithms respectively.The results showed that the estimation model based on optimized spectral index was more accurate than that based on simulated band.By comparing the three machine learning algorithms,the modeling accuracy of stochastic forest algorithm is above 0.9,while the modeling accuracy of partial least squares and support vector algorithm is 0.76 and 0.70,respectively.It can be seen that the estimation ability of stochastic forest algorithm on nitrogen uptake is stronger than that of support vector machine and partial least squares algorithm.The accuracy of sentinel-2A satellite sensor is also different,and the estimation ability of Sentinel-2A satellite sensor is significantly stronger than that of GF-2 and Rapideye.By comparing the nitrogen uptake by potato plants in different phenological periods to estimate the model establishment,it is concluded that the sensitive bands of different spectral indices can be accurately screened by the band optimization algorithm to effectively improve the accuracy of the model.The model established by the random forest algorithm in combination with the optimized spectral index calculated by the Sentinel-2A satellite simulation data has high accuracy and can be used to estimate the nitrogen uptake by potato plants;the estimated nitrogen uptake by potato plants before flowering is more accurate than that after flowering.The random forest algorithm can significantly improve the accuracy of the model,so it has a broad prospect in the establishment and application of the potato nitrogen nutrition diagnosis model.