关键词: 树种分类, 支持向量机, 时相选择, C5.0决策树, 高分二号, 随机森林

Abstract: In order to screen a continuous cropping pattern of water and drought replanting rotation, which is sustainable and suitable for rice fields in southern China, we carried out an experiment with five treatments from 2014-2016. The “winter fallowearly ricelate rice” model was control (A), and there were four types of paddy field irrigation and drought multiple cropping rotation modes, including potatomaize ‖ soybeanlate rice→vegetablespeanut ‖ corn late rice→milk vetchearly ricelate rice (B), vegetablespeanut ‖ corn-late rice→milk vetchearly rice-late rice→rapepeanutlate rice (C), milk vetchearly ricelate rice→oil rapepeanutlate rice→potatomaize ‖ soybeanlate rice (D), oil rapepeanutlate rice→potatomaize ‖ soybeanlate rice→vegetablespeanut ‖ cornlate rice (E). We analyzed crop yield with mainly for late rice yield and the changes of soil nutrients under different planting patterns. The results showed that paddyupland multiple cropping rotation systems had advantages in yield, with the yield of late rice being higher than that of winter fallow cropping pattern. Compared with control, the paddyupland multiple cropping rotation systems increased soil organic matter and available nutrient content, being more conducive to organic carbon fixation. Paddyupland multiple cropping rotation patterns are planting patterns of increasing yield and improving soil fertility, with better performance for treatments C and E.

Key words: GF-2, C50 decision tree, random forest, tree species classification, temporal stage selection, support vector machines