應用M5'回歸樹建構台灣玉米花粉飄散模式

Abstract

隨著生物技術的發展，基因改造（genetically modified，GM）作物的種植面積及品項逐年增加。然而，由於在生態環境、消費者食用及經濟上所造成的未知風險，使得GM與非GM作物在田間共存（co-existence）的措施及監測成為許多研究者和政府關心的議題。本研究以紫色玉米模擬GM玉米為花粉貢獻親，白色玉米作為花粉接受親，於行政院農委會農業試驗所與台南區農業改良場朴子分場進行花粉飄散試驗。並以M5'回歸樹、非線性模式（exponential模式、log/log模式和log/square模式）與two-step模式建立台灣花粉飄散模式並進行探討。研究結果顯示，若僅以農業試驗所的田間資料建立M5'回歸樹，其配適結果與非線性模式及two-step模式的表現相似。而若增加解釋變數，以朴子地區的田間資料建構M5'回歸樹的結果顯示能得到更佳的配適。而將農業試驗所和朴子分場的資料合併之後，M5'回歸樹的配適能力則更優於非線性模式和two-step模式。整體而言，由於M5'回歸樹為經驗模式，因此當花粉飄散資料量增加時，並同時參酌更多解釋變數，M5'回歸樹確實能夠充分運用訊息，獲得較佳的結果。 Along with improvements in biotechnology, the acreage and varieties of genetically modified (GM) crops are increasing rapidly. Because of the unknown risks to the ecological environment, consumer consumption, and economic issues, measures and detections of the coexistence between GM and non-GM crops have become concerns of researchers and governments. In our study, the GM crop was simulated by purple-glutinous maize. The non-GM crop was white-glutinous maize. The field experiments were conducted at Potzu Branch Station in Puzih in 2009 and 2010 and at the Taiwan Agricultural Research Institute (TARI) in 2011 to construct pollen-mediated gene flow (PMGF) models suited to the environment in Taiwan (R.O.C). The fitting capabilities of a M5' regression tree, non-linear models (exponential model, log/log model, log/square model), and a two-step model were compared by using the statistical indicators of r and RMSE.According to the findings, the M5' regression tree performs as well as the other models based on statistical indicators when using TARI data. Moreover, the data collected from the Potzu Branch Station show that when the number of explanatory variables is increased, the M5' regression tree performs better than the other models. Furthermore, the combined data collected at TARI and Potzu Branch Station show that the M5' regression tree performs much better than the other models. Finally, because the M5' regression tree is an empirical model, the results indicate excellent fitting capability as the number of explanatory variables and the sample information increase.