利用植被光譜特徵辨識水稻遭受瘤野螟及稻熱病之危害

Abstract

本文研究旨在利用光譜遙測技術偵測近地面水稻植被反射光譜，以篩檢與病蟲害發生有關之光譜特徵，並試以進一步利用統計方法建立光譜特徵與病蟲害感染等級之關係，俾未來應用於水稻精準農業體系之病蟲害管理用途。本文收集遭受不等瘤野螟及稻熱病感染等級之水稻(Oiyza sativa L. cv. Tainung 67)植被光譜(350-2500 nm)，其中感染瘤野螟（Cnaphalocrosis medinalis Guenee)之植被光譜偵測於穀粒充實初期，感染稻熱病(Pyricularia otyzae Cav.)之光譜則記錄於最大分蘗期。近地面光譜遙測乃以田問可攜式高解析輻射光譜分（層）析儀(spectroradiometer, GER-2600，Geophysical & Environmental Research Corp., Millbrook, NY, USA)為之，瘤野螟感染光譜計量測二天，稻熱病感染光譜計量測三天，皆於近午時刻（上午十時三十分至下午一時三十分之問）進行。瘤野螟之感染等級係依照受感染葉片數多寡區分為0（健康株）、1 、3 、5 、7 及9 等六級；稻熱病之感染等級則視受感染葉片之面積百分比區分，由0％（健康株）至25％不等，計有八等級。根據光譜各窄波段反射比與感染等級之相關強度分析(correlation intensity analysis) 結果發現，各窄波段之反射比與感染等級之相關性不一，最大相關係數在感染葉稻熱病光譜出現在1436nm 位置(r=0.982**)，在感染瘤野螟光譜則落於2327 nm 波長(r=0.946**)在光譜指數調查方面，發現WRED/GREEN ratio （紅光波段谷底反射比與綠光波段波峰反射比之比值）、NIR（近紅外光波段波峰反射比）、RED、紅光臨界斜率(red edge slope)及紅光臨界中間值波長（red edge mid point)等五項，均與葉稻熱病感染等級之問呈現顯著相關；而標準差值被指數(norma1ized difference vegetation index, NDVI)、RED/MR ratio及RED/GREEN ratio等三項、則與瘤野螟感染等級相關顯著。又若將植被光譜三處特定窄波段GREEN 、RED 及NIR 納入多變量（元）線性回歸分析(multiple linear regression (IVILR) analysis)，在葉稻熱病部分發現所構建之三元回歸模式Y=-9.391+6.26sRED+ 0.340MR-3.381GREEN具有高達0929 之決定係數（determining factor or coefficient of determination, R2) (P<0.01)。在感染瘤野螟之光譜，包括此三特定窄波段之MIR分析，則最佳二元回歸模式Y=-3.742-3.742RED +8.616GREEN (R2=0.963, P<0.01)已能解釋高達96％以上變異，增加NIR變數並無明顯成效。若由光譜範圍選取多個窄波段進行多變量（元）線性回歸分析，從葉稻熱病試驗之植被光譜選取8 條窄波段之最佳三元回歸模式為Y=-21.401+1.162R620 nm +4.855R1436 nm -3.914R2198 (R2=0.980, P<0.001)，由感染瘤野螟光譜選取6 條窄波段之最佳二元回歸模式為Y=-8.749-2.336R550nm+5.099R691nm (R2=0.989, P<0.002), 其中Y為感染等級Rinm則為特定inm窄波段之反射比。 Reflectance spectra (350-2500 nm) of hvperspectral resolution from rice (Otyza saliva L. cv. Tainting 67) canopy infected with various percentages of leaf blast (Pyricularia oryzae Cay.) and different levels of leaffolder (CnapLalocrosis ,neadinalis) were collected and analyzed using a field portable spectroradiometer. Percent infection from 0% (healthy plants) up to 25% was based on blast disease affected leaf area. Infestation levels by the leaf folder were classified into six levels, ic., 0 (healthy check), 1, 3, 5, 7, and 9, based on the numbers of damaged leaves. By the correlation intensity analysis between reflectance and percent infection or levels of infestation, it showed that the maximum value of correlation coefficient was located at 1436 nm (r 0982) for leaf blast and at 2327 nm (r0.946**) for leaffolder. The spectral indices of RED/GREEN ratio (reflectance ratio of red light minimum to green light peak), NTR (reflectance at near-infrared maximum), RED, red edge slope, and red edge mid point were all positively correlated with percent infection of leaf blast. The normalized difference vegetation index (NDVI), the RED/MR ratio, and the RED/GREEN ratio were curvilinearly correlated with infestation levels of leaffoldet By the multiple linear regression analysis with GREEN, RED and NIR as variables, the model was Y-9.391+6.265RED+ 0.340N1R-3.381GREEN, with R2=0.929 (P<0.01) for disease estimation, and the best 2- variable MLR model for pest evaluation was Y=-3.742-3.742RED +8.616GREEN (R20.963, P<0.001). When 8 optimal narrow bands selected from reflectance spectrum of disease-infected canopy were analyzed, the best 3- variable model was Y=-21.401+ 1.162R620 nm +4.855R1436 nm3914R2198 (R20.980, P<0.001). In pest-infested canopy with six selected narrow bands for analysis, the best 2-variable model was Y=-8.749-2.336R550 nm +5.099R691 nm (R2=0.989, P<0.002).