空間尺度及資料解析度：空間外推物種生態樣式之兩難以香桂與蘇鐵蕨為例

Spatial Scale and Data Resolution: The Dilemma of Spatially Extrapolating Ecological Pattern of Species - A Case Study of Cinnamomum subavenium and Brainea insignis

分析物種與環境的關係一直是生態學的中心議題。生態與環境問題，如生物多樣性損失及全球暖化等，是在大區域或於長期時空尺度下運作，但生態研究實測資料通常是於短暫期間且在小區域採集，兩者尺度不相稱。「空間外推」一直是生態學的部份，但20世紀下半變成必要技術且為應用生態學研究焦點。直接作用因子外推大空間尺度生態關係，雖只有粗解析度資料，且預測欠精確，但具普遍性。相對的，間接作用因子外推小空間尺度生態關係，雖有細解析度資料，且預測精確，卻難適用在更廣大地區，此即生態學空間外推的兩難困境。過去數十年，遙測及GIS技術創新，可在更大空間尺度及更細緻層次描繪空間樣式。本研究因取樣設計需要，選香桂與蘇鐵蕨為研究對象。藉由GIS疊合兩物種與地形變數及SPOT-5影像植生指標圖層，協同抉擇樹（DT）及區別分析（DA），預測兩者於惠蓀林場試區空間樣式。建、驗模採三種取樣設計，是分別取自東峰溪與關刀溪流域不同樣本組合而成。準確度評估顯示，DT遠優於DA，而兩者執行效率相當。重要的是DT於首次模擬，大幅縮小實地調查面積，節省可觀經費及人力，故更適用於兩物種適生育地模擬。植生指標改善模式預測能力效用很小，乃因其光譜及空間解析度皆不足，無法分辨散生香桂與蘇鐵蕨。兩種統計法建立「東峰模式」皆未通過關刀驗模組檢測，凸顯僅含地形變數模式無法跨越空間外推無建模樣本區域。未來研究將從高空間、高光譜解析度遙測資料萃取物種光譜資訊作建模用變數，期使模式能普遍應用於大空間尺度。 The analysis of species-environment relationship has always been a central issue in ecology. Ecological and environmental problems, such as global warming and biodiversity loss, operate over very large areas or over extended periods of time, but the field data that characterize ecological research are typically collected over relatively small areas during studies of short duration. The scales with data collection mismatch.”Spatial extrapolation” has always been a part of ecology, but it became a sine qua non and a major research focus in applied ecology in the latter half of the 20th century. The model using direct parameters for extending ecological patterns is more general and applicable over larger areas although it has only coarse resolution and low accuracy data available and thus is inaccurate. On the contrary, the model using indirect parameters can be applied within a limited geographical extent although it has fine resolution and high accuracy data available and thus is accurate. This is the dilemma of spatial extrapolation in ecology. Technological innovations over the last few decades, especially in the fields of remote and GIS, have greatly enhanced scientists' ability to describe patterns over broader spatial scales and at a greater level of detail. Randaishan cinnamons (Cinnamomum subavenium Miq.) and cycad-ferns (Brainea insignis) were chosen as target for this study because their locations and distributions meet the requirements of sampling designs. GIS technique was applied to overlay the sample layers of the two species on the layers of topographic variables and vegetation indices derived from SPOT-5 images for modeling the species' suitable habitat. Decision tree (DT) and discriminant analysis (DA) models were developed to predict and map the species' suitable sites in the study area, and to determine the optimum one in terms of accuracy and efficiency. Three sampling designs derived from different combinations of samples taken from Tong-Feng and Guan-Dau watersheds were used for model development and validation. Accuracy assessment showed that the accuracy of DT was much better than that of DA; and the two models were highly efficient in implementation of model development and validation. More importantly, DT can be applied to predict the species' suitable habitat because they greatly reduced the area of field survey at the first stage. Vegetation indices could not improve the predicting ability of models for the widely distributed species because of SPOT imagery lacking fine spectral resolution and spatial resolution. ”Tong-Feng models” developed from two methods failed to pass validation by Guan-Dau test samples despite passing validation by Tong-Feng test samples. The outcome emphasized that the models only based on topographic variables could not perform spatial extrapolation accurately from a smaller area with training data to a larger area without any training data. Follow-up studies will attempt to extract spectral information associated with species from high spatial, spectral resolution remotely sensed data and use it as variable for model development so that models are more general and applicable over larger areas.

The analysis of species-environment relationship has always been a central issue in ecology. Ecological and environmental problems, such as global warming and biodiversity loss, operate over very large areas or over extended periods of time, but the field data that characterize ecological research are typically collected over relatively small areas during studies of short duration. The scales with data collection mismatch.”Spatial extrapolation” has always been a part of ecology, but it became a sine qua non and a major research focus in applied ecology in the latter half of the 20th century. The model using direct parameters for extending ecological patterns is more general and applicable over larger areas although it has only coarse resolution and low accuracy data available and thus is inaccurate. On the contrary, the model using indirect parameters can be applied within a limited geographical extent although it has fine resolution and high accuracy data available and thus is accurate. This is the dilemma of spatial extrapolation in ecology. Technological innovations over the last few decades, especially in the fields of remote and GIS, have greatly enhanced scientists' ability to describe patterns over broader spatial scales and at a greater level of detail. Randaishan cinnamons (Cinnamomum subavenium Miq.) and cycad-ferns (Brainea insignis) were chosen as target for this study because their locations and distributions meet the requirements of sampling designs. GIS technique was applied to overlay the sample layers of the two species on the layers of topographic variables and vegetation indices derived from SPOT-5 images for modeling the species' suitable habitat. Decision tree (DT) and discriminant analysis (DA) models were developed to predict and map the species' suitable sites in the study area, and to determine the optimum one in terms of accuracy and efficiency. Three sampling designs derived from different combinations of samples taken from Tong-Feng and Guan-Dau watersheds were used for model development and validation. Accuracy assessment showed that the accuracy of DT was much better than that of DA; and the two models were highly efficient in implementation of model development and validation. More importantly, DT can be applied to predict the species' suitable habitat because they greatly reduced the area of field survey at the first stage. Vegetation indices could not improve the predicting ability of models for the widely distributed species because of SPOT imagery lacking fine spectral resolution and spatial resolution. ”Tong-Feng models” developed from two methods failed to pass validation by Guan-Dau test samples despite passing validation by Tong-Feng test samples. The outcome emphasized that the models only based on topographic variables could not perform spatial extrapolation accurately from a smaller area with training data to a larger area without any training data. Follow-up studies will attempt to extract spectral information associated with species from high spatial, spectral resolution remotely sensed data and use it as variable for model development so that models are more general and applicable over larger areas.