Estimation of forest canopy nitrogen content based on remote sensing

Abstract

Hypespectral data was used to estimate leaf and canopy nitrogen content. Erf-BP, an improved model based on the Gaussian error function of BP neural network, was used to develop remote sensing models for estimating leaf nitrogen content. Then the scaling conversion function during downscales from canopy to leaf spectral was derived according to principles of geometric optics model. These relations were used during downscales from the canopy reflectance of Hyperion image to leaf spectral for leaf nitrogen content estimation. Finally, forest structural parameter leaf area index (LAI) was used to obtain canopy nitrogen content from leaf level. The results showed that the best Erf-BP. neural network model with testing accuracy of 76.8597% includes 8 neurons in hidden layer. Using scaling conversion function to estimate canopy spectra at 670nm and 865nm, correlations (R-2) between modeling spectra and measurements were 0.5203 and 0.4117 respectively. Correlation coefficient between estimated leaf nitrogen content and measurements was 0.7019. This method provides a good reference for more rapid and accurate estimation of leaf and canopy nitrogen.