Evaluating the performance of PC-ANN for the estimation of rice nitrogen concentration from canopy hyper-spectral reflectance

With the increasing emphasis on environmental issue, ensuring high quality of crops yield, meanwhile, preventing or reducing environmental pollution is the most important problem to need to be solved by governments, experts, environmental workers and producers. Therefore, effective nitrogen management, rational fertilizer use, and accurately, rapidly and economically determining nitrogen regime and requirement of crops and improving N-fertilizer use efficiency is of great economic and ecological significance.

Remote sensing technology provides a convenient diversified tool to obtain biochemical component contents at different scales. With the development of hyper-spectral technology, the study of extract quantitative information of plant biochemical components is quickly developed; meanwhile modern computer technology provides powerful computing and processing capacity of data. These technologies enrich substantially the data processing methods of the biochemical component by remote sensing information extraction. Therefore, data mining technology occurs and become the hot spot of quantitative remote sensing of vegetation biochemical.

Dr. Yi Qiuxiang, from Xinjiang Institute of Ecology and Geography, CAS, discussed the accuracy of remote sensing techniques to evaluate the nitrogen concentration of rice canopy combining the properties of artificial neural network (ANN) with principal component analysis (PCA). In the study, hyper-spectral reflectance data of the rice canopy through rice whole growth stages were acquired through different fertilizer level experiments. Comparisons of prediction power of two statistical methods (linear regression technique (LR) and ANN), for rice nitrogen estimation were performed using nitrogen sensitive hyper-spectral reflectance and principal component scores. The performance of models was measured by the root mean square error (RMSE) and the relative error of prediction (REP). The results indicated a very good agreement between the observed and predicated N with all model methods, which is especially true for the PC-ANN model. Compared to the LR algorithm, the ANN increased accuracy by lowering the RMSE by 17.6% and 25.8% for models based on spectral reflectance and PCs, respectively. The study result is published by the journal of International Journal of Remote Sensing.