UAV Multispectral Images-based Nitrogen Status Estimation In Maize

Maize cultivation in Inner Mongolia is important for food security and sustainable development strategies in China.However,a key challenge is to optimize N management to ensure high yield with low environment pollution in maize production.In this study,spring maize was used as the research target in the central and western Inner Mongolia.The UAV multispectral remote sensing images,aboveground biomass（AGB）and maize plant nitrogen concentration（PNC）were obtained through three consecutive years of field trials to establish the critical nitrogen concentration dilution curve for maize and verify its reliability.The correlations were analyzed between spectral index,wavelet features and texture features extracted from multispectral images and maize PNC and AGB,and then selected the main features and corresponding feature fusion as the input variable of Partial Least Squares Regression（PLSR）,Random Forest（RF）and Gaussian Process Regression（GPR）to construct prediction models.Finally,the nitrogen nutrition index（NNI）was calculated to make a diagnosis of nitrogen nutrition in maize plants.The main results were as follows:（1）In the same region,maize varieties with close yield levels can share a common critical N concentration dilution curve.The constructed critical N concentration dilution curve for spring maize in central and western Inner Mongolia was Nc=3.32W^-0.305,and the model validation showed that the model was stable and could effectively diagnose the nitrogen nutrient status of maize plants in central and western Inner Mongolia.The model inferred that the reasonable N application rate was 180-220 kg N hm² for maize in central and western Inner Mongolia.（2）Multi-feature fusion has high potential for application in maize PNC estimation,and spectral indices,wavelet features and texture features are effective features for estimating maize PNC.Modeling with a combination of multi-feature fusion could significantly improve the accuracy of maize PNC estimation.The GPR model has the best generalization ability in validation dataset,the R²,RMSE and RE was 0.76,0.32%and14.11%,respectively.（3）The GPR model had the best performance in maize AGB estimation,explaining75%of the variation in maize AGB with RMSE and RE of 4.29 t/hm² and 10.51%.The integration of features coupled with machine learning algorithms can effectively improve the accuracy of maize AGB estimation and provide a reliable modeling framework for maize AGB monitoring and scientific support for maize field management.（4）The predicted AGB and PNC results obtained by using UAV multispectral remote sensing images were used to calculate NNI,the relationship between predicted and observed NNI were 0.84,0.09 and 6.24%for R²,RMSE and RE,respectively.The NNI can be used for nitrogen nutrition diagnosis in maize fields and can provide a good range of reasonable fertilizer application decisions through the diagnostic visualization.The above results showed that the maize critical N concentration dilution curve model can be used to guide the application of N fertilizer to maize fields in central and western Inner Mongolia.The maize PNC and AGB estimation models established by the spectral indices,wavelet features and texture features extracted from the UAV multispectral remote sensing images in concert with the GPR algorithm have high estimation accuracy.In addition,the NNI calculated based on the indirect method can be used for maize plant nitrogen diagnosis,indicating that the UAV multispectral remote sensing images have the potential to diagnose the nitrogen nutrition of maize plants.The results further provide theoretical basis and technical support for the recommended fertilizer application in maize stages.