Prediction of carbon, nitrogen and phosphorus contents of Leymus Chinensis based on soil chemical properties using artificial neural networks.

Ecological stoichiometry is an emerging discipline started in China in recent years. It is the science of studying the balance of energy and elements (i.e. carbon, nitrogen and phosphorus) in ecological processes and ecological interaction, providing an integrative approach to investigate the stoichiometric relationships and rules in the biogeochemical cycling and ecological processes. It has been one of the hotly-discussed issues in ecological research. The contents of carbon, nitrogen, and phosphorus is a core issue in ecological stoichiometry studies. It is necessary to choose a method that can simulate and accurately predict the contents of plant carbon, nitrogen, and phosphorus in order to avoid destructive sampling. There is a complex nonlinear relationship between plant carbon, nitrogen, phosphorus, and soil physical and chemical properties. It is difficult to accurately predict plant carbon, nitrogen, and phosphorus by using traditional methods and models such as linear regression and a BP neural network. As a new artificial neural network model, a RBF (radial basis function) neural network has some advantages of fast learning, getting in the local minimum, and approximating any arbitrary accuracy of the global nonlinear relationship. Therefore, a RBF neural network can show an ability to handle a complex nonlinear relationship. Currently, a RBF neural network is one of the most accepted prediction methods. Taking the prediction of 38 samples as a research sample, this paper established a prediction model based on a RBF Neural network from seven impact indexes including pH, the total soluble salt, total carton, total nitrogen, total phosphorus, available nitrogen, and available phosphorus. Taking the prediction of five samples as a test sample, the results indicated that the relative errors of carbon, nitrogen, and phosphorus contents were only 1.39%, 4.69%, and 7.65%, respectively, and the correlation coefficients were 0.5, 0.93, and 0.94 respectively. Therefore, a RBF neural network had higher prediction accuracy. The statistical results showed that the average contents of carbon, nitrogen, and phosphorus in Leymus chinensis (103 samples) were 411.46, 18.25, and 1.11 mg/g, respectively. They are all lower than the global average contents of carbon, nitrogen, and phosphorus in a terrestrial plant. The values of C/N, C/P, and N/P were 24.70, 429.24, and 17.92, respectively in Leymus chinensis. They were all higher than those in a global terrestrial plant. The N/P was 17.92 in Leymus chinensis. The growth of Leymus chinensis in the research area was limited by phosphorus. [ABSTRACT FROM AUTHOR]