應用遙測技術推估水稻產量之初探

Abstract

利用遙測影像推估田間水稻產量的空間分佈，不僅是實施精準農業的重要基礎，也可供糧政主管單位進行糧食供需的調配。本研究目標主要在分析地面測得之植被反射光譜，建立特徵光譜與水稻產量問的定量關係，作為應用遙測技術推估田間水稻產量空間分布之基礎。研究以台灣地區目前栽培面積較廣之台農67號水稻為研究對象，於農試所嘉義分所之溪口農場內建立水稻樣區，在1999-2001年三年間藉由不等量氮肥施用產生水稻產量間之變異。試驗期間定期測定田間水稻植冠反射光譜，以分析水稻全生育期中反射光譜變動特性，並於水稻收穫時調查水稻之產量，尋求水稻反射光譜特性與產量間的關聯性，以建立利用遙測影像估測水稻產量所需之模式。初步研究結果指出，在抽穗期間水稻植被反射光譜的動態變動最小，因此適合用於進行田間水稻產量空間分布之推估。以植被反射光譜中綠光段（G）、紅光段（R）、和近紅外光段（NIR）計算之波段比值（NIR/R、NIR/G )，可組成推估水稻產量的複回歸模式，並可適用於推估不同年期之水稻產量。以農試所內精準農業農場2001年一期作的資料進行所得模式的應用性檢驗研究，結果指出該模式亦可適用於農試所內精準農業農場，唯可能由於種植密度和產量調查時坪割面積與考種方式的差異，雖然遙測模式預測值與實測值間具有良好的正相關（R2=0.84)，但截距仍約有2 T ha-1，因此該模式目前僅適合用於區劃田間水稻相對產量的空間分布，還不能用於區劃田間實際產量之空間分布情形，仍有待累積更多地點的相關試驗資料以進行模式之修正。 Abilities to estimate rice yields within fields from remote sensing images are not only fundamental to applications of precision agriculture (PA) but also useful to the governmental administrators for food management. Major objectives of this study were to identify spectral characteristics associated with rice yield and to establish their quantitative relationships. Field experiments were conducted at Shi-Ko experimental farm of TART’s Chiayi Branch Station, during 1999-2001. Rice cultivar Tainung 67 (Oryza sativa L.), the major cultivar grown in Taiwan, was used in the study. Varied rice yields were obtained via nitrogen application treatments. Canopy reflectance spectra were measured during rice growth and dynamic changes of characteristic spectrum were analyzed. Relations among rice yields and characteristic spectrum were studied to establish yield estimation models suitable for remote sensing purposes. Preliminary results indicated that the dynamic changes of canopy reflectance spectrum were least during booting stages. Therefore, the canopy reflectance spectrum during this period were selected for model development. A multiple regression model, constituting of band ratio (NIR/R, NIR/G) calculated from green (G), red (R) and near infrared (NIR) portions of canopy reflectance spectrum, was able to estimate rice yields from different years. Validation tests, using data of 1st crop, 2001 collected at PA experimental fields of TART, indicated positive correlation (R20.84) between model prediction and actual measurements. However, an intercept of approximately 2 T ha-1, may be caused by differences in planting density or yield sampling techniques, indicated that the model was suitable only for estimating relative yield distribution within the field instead of actual yield. More data collected at different locations are required to further improve the model.