Machine learning for the prediction of L. chinensis carbon, nitrogen and phosphorus contents and understanding of mechanisms underlying grassland degradation.

The grasslands of Western Jilin Province in China have experienced severe degradation during the last 50 years. Radial basis function neural networks (RBFNN) and support vector machines (SVM) were used to predict the carbon, nitrogen, and phosphorus contents of Leymus chinensis (L. chinensis) and explore the degree of grassland degradation using the matter-element extension model. Both RBFNN and SVM demonstrated good prediction accuracy. The results indicated that there was severe degradation, as samples were mainly concentrated in the 3rd and 4th levels. The growth of L. chinensis was shown to be limited by either nitrogen, phosphorus, or both during different stages of degradation. The soil chemistry changed noticeably as degradation aggravated, which represents a destabilization of L. chinensis community homeostasis. Soil salinization aggravates soil nutrient loss and decreases the bioavailability of soil nutrients. This, along with the destabilization of C/N, C/P and N/P ratios, weakens the photosynthetic ability and productivity of L. chinensis. This conclusion was supported by observations that L. chinensis is gradually being replaced by a Chloris virgata, Puccinellia tenuiflora and Suaeda acuminate mixed community.