Search

• A locally modelling approach was proposed to build spectral model for estimating leaf nutrients at regional scale. • This approach considered the varied relations between leaf nutrients and leaf spectra over space. • This approach was applied to a case study in Hainan Island to estimate leaf phosphorus concentration of rubber trees. • The proposed approach outperformed the commonly used globally modelling method. The globally modelling method was commonly used to build spectral models for estimating leaf nutrients of plants. However, predictive ability of such models was rather limited at regional scale because of the unreasonable assumption inherent in the globally modelling method which assumed that relationships between leaf nutrients and leaf spectra were stable over space. In fact, the leaf nutrients-spectra relationships generally varied across extensive area due to influences of various environmental factors. In order to accommodate the varied leaf nutrients-spectra relationships at regional scale, the locally modelling approach was proposed to develop spectral models for predicting leaf nutrients. The locally modelling approach assumed that the leaf nutrients-spectra relationships were stable at local spaces within which environmental conditions were relatively uniform. According to this basic idea, the proposed approach used environmental factor as grouping variable to classify leaf samples into several homogenous sub-datasets. Within these sub-datasets, the leaf nutrients-spectra relationships would be more stable. Then, linear or non-linear method were employed to fit model for each sub-dataset. In order to verify the effectiveness of this method, the locally modelling approach was applied to a case study to estimate leaf phosphorus concentration (LPC) of rubber trees (Hevea brasiliensis) in Hainan Island, China. In this case study, three scenarios were set up. In the first scenario, the local model was built using the partial least squares regression (PLSR) with full wavelengths as input variables. In the second scenario, the local model was also developed using the PLSR but with manually selected wavelengths as input variables. In the third scenario, the local model was constructed using the artificial neural networks with feature bands as input variables. Performances of these local models were compared to that of the corresponding global models. Results showed that for the three scenarios local models always yielded higher prediction accuracies than that of the global models in the test dataset. For the first scenario, values of coefficient of determination (r2) and root mean square error (RMSE) ranged from 0.652 to 0.735 and from 0.030% to 0.034%, respectively, for the local model, whereas those varied from 0.458 to 0.637 and from 0.037% to 0.041%, respectively, for the global model. For the second scenario, values of r2 and RMSE ranging from 0.565 to 0.734 and from 0.031% to 0.046%, respectively, for the local model, while those varied from 0.275 to 0.559 and from 0.038% to 0.053%, respectively, for the global model. For the third scenario, values of r2 and RMSE ranging from 0.678 to 0.863 and from 0.021% to 0.033%, respectively, for the local model, while those varied from 0.441 to 0.726 and from 0.030% to 0.043%, respectively, for the global model. This result demonstrates that the locally modelling approach was more effective in estimating LPC of rubber trees than the globally modelling method at the regional scale. In the future, more environmental factors should be considered and their combined effects on leaf sample classification need to be investigated. [ABSTRACT FROM AUTHOR]

• CARS-SPA was used to extract feature bands for estimating leaf N of rubber trees. • Sixty bands were selected from the pretreated first derivative spectra (2150 bands). • Except the N related bands, the ones related to starch, cellulose, and lignin were also important. • ANN model with feature bands is accurate and stable in estimating leaf N of rubber trees. Leaf nitrogen is an indispensable parameter for simulating biogeochemical cycling in ecosystems. However, detailed information about leaf nitrogen of rubber trees (Hevea brasiliensis) in the tropical regions is rare. Aims of the current study were to determine the feature bands of foliar nitrogen of rubber trees and to examine the ability of these bands to estimate leaf nitrogen concentration (LNC). In this work, a hybrid variable selection method namely competitive adaptive reweighted sampling (CARS) combined with successive projections algorithm (SPA) (CARS-SPA) was used to extract feature bands for predicting LNC of rubber trees. Two hundred leaf samples were gathered from five fields through April to November (tapping season) in 2014. These samples were divided into calibration dataset (n = 140) and prediction dataset (n = 60) using the Kennard-Stone algorithm. Sixty bands were determined from the first derivative spectrum using the CARS-SPA. Among these bands, not only the commonly used absorption features of chlorophyll, protein, and the red edge position were included, but also those of starch, cellulose, and lignin were selected. All the 60 bands were used as input variables for the partial least squares regression (PLSR) and artificial neural networks (ANN) models to estimate LNC of rubber trees. The determination coefficient of calibration (r c 2) and prediction (r p 2), and root mean square error of calibration (RMSEC) and prediction (RMSEP), as well as normalized RMSEC (nRMSEC) and normalized RMSEP (nRMSEP) were employed to evaluate performances of these models. Results indicated that CARS-SPA-ANN (r c 2 = 0.82, RMSEC = 0.24%, nRMSEC = 7.76%; r p 2 = 0.78, RMSEP = 0.22%, nRMSEP = 6.73%) outperformed the other selected models except CARS-PLSR (r c 2 = 0.91, RMSEC = 0.17%, nRMSEC = 5.46%; r p 2 = 0.80, RMSEP = 0.21%, nRMSECP = 6.44%). However, CARS-SPA-ANN contained much less bands and was more stable than CARS-PLSR. In conclusion, hyperspectral feature bands in combination with ANN could accurately and robustly estimate LNC of rubber trees. [ABSTRACT FROM AUTHOR]