臺灣植生指標與氣候因子及耕作制度關係之時空分析

Spatial-temporal analysis of the relationships between vegetation index, climatic factors and cropping system in Taiwan

鄭晢恩(Che-En Cheng)

指導教授 ： 陳虹諺

國立臺灣大學/生物資源暨農學院/農藝學研究所/碩士(2020年)

https://doi.org/10.6342/NTU202001060

摘要

在氣候變遷越來越不受控制的情況下，盡可能地減輕人類對生態環境的干擾並予以保育措施是備受重視的工作。植生變遷直接或間接地反應氣候變遷與人為活動對生態環境的影響。因此利用具有低成本、高覆蓋、高解析等性質的多時期衛星影像可以研究植生改變趨勢，進而反應生態變遷程度。但是氣候環境亦驅使著植物生長反應，並且植生變化是由環境因子組成的非線性系統產物，要分辨是否為人為引起的變化是一項挑戰。本次研究獲取NASA Earth Observation的2001年至2017年間0.05度解析度每月常態化差異植生指標 (NDVI)、TCCIP的降雨量、氣溫網格資料以及MODIS土地覆蓋類型影像，使用IDW將不同解析度與中心點的資料再取樣到相同之座標位置。
研究內容分為趨勢分析與時頻分析兩個主軸。趨勢分析旨在偵測並定量人為引起的植生長期序列平均趨勢變化。首先探討線性迴歸、無母數之Mann-Kendall、Theil-Sen slope等常用趨勢檢定方法的應用，並提出恆定與移動基準線時間整合指標做為具有生物意義的新指標序列，改善序列自相關問題，最後以殘差趨勢 (RESTREND) 方法分離潛在於氣候因子下的植生變化，偵測全臺範圍於2001至2014年間可能由人為引起之綠化/褐化趨勢大小與趨勢等級。由於非平穩與非線性的資料性質，使用希爾伯特黃轉換方法進行時頻分析，透過改善混模問題的CEEMDAN演算法將NDVI、降雨量及氣溫序列分解為不同頻率特徵的分量訊號並轉換為瞬時週期與振幅，探討不同尺度下植生變化與不同耕作制度及氣候因子之間的關聯。
恆定與移動基準線分別適用於組成植被覆蓋的主要物種無改變與有大幅改變的地區。恆定基準衍生殘差趨勢檢定結果顯示臺北–苗栗、南投、嘉南–高屏、花東等地區具有顯著綠化趨勢，並且在桃園、新竹耕地地區有耕地面積增加的文獻紀錄。移動基準衍生殘差趨勢檢定結果顯示中北部、嘉南、花東等耕地地區與中央–阿里山地區存在褐化趨勢，發現部分地區的植物物候模式發生改變，例如作物由稻作轉變為旱作或果樹。此方法可應用於大範圍偵測植生變遷趨勢，有助於決策者鎖定褐化地區並進行後續調查是否有人為開發或不當利用土地導致的植被面積、植物物種減少。
時頻分析結果顯示各地區耕作制度與NDVI的1季到半年的短週期特徵有所關聯，年週期性則反映了氣溫的季節變化。植生指標、氣候資料頻譜的振幅變化可與現實發生的颱風、豪雨、乾旱或暖冬等極端氣候事件造成的農業損害有所對應。分析頻率特徵可瞭解耕地植生變化如何對耕作制度改變或極端氣候事件做出反應，提供這些資訊期望能幫助訂定農業管理決策。

關鍵字

常態化差異植生指標 ； 趨勢分析 ； Mann-Kendall檢定 ； 殘差趨勢 ； 時間整合NDVI ； 時頻分析 ； 希爾伯特黃轉換 ； CEEMDAN ； 耕作制度 ； 距離反比加權

並列摘要

Because of the climate change becoming more and more uncontrollable, it is necessary to minimize the damage to the ecological environment and provide protection. Vegetation changes directly or indirectly reflect the affect of climate change and human activities on the ecological environment. Therefore, the use of multitemporal satellite image with low cost, high coverage rate and fine resolution can study the trend of vegetation changes, and then reflect the ecological changes. However, the climate also drives plant growth responses, and the vegetation changes is the product of a nonlinear system composed of environmental factors. It is a challenge to distinguish the vegetation changes induced by humans or climate. The monthly normalized differential vegetation index (NDVI) with the resolution of 0.05 degrees from 2001 to 2017, which retrieved from NASA Earth Observation, TCCIP rainfall, temperature grided data, and MODIS land cover type image were accessed. The inverse distance weighted method was applied to the resampling of the data with different resolution and centroid of grid.
The study was divided into two main aspects, trend analysis and time-frequency analysis. Trend analysis aims to detect and quantify the average trend of long-term series induced by humans. First, the application of popular trend test methods such as linear regression, non-parametric Mann-Kendall trend test, and Theil-Sen slope estimation were discussed. The time-integrated index of constant and moving bases was proposed as a new proxy with biomass to improve serial correlation. Finally, the RESTREND method was used to identify the potential vegetation changes under the climatic factor, and the degree of the human-induced greening/browning trend in Taiwan from 2001 to 2014 are detected. Due to the non-stationary and non-linear properties of the data, the Hilbert-Huang transform was used for time-frequency analysis. Through the CEEMDAN algorithm to improve the mode mixing problem, the NDVI, rainfall and temperature series were decomposed into component signals of different frequency and transformed into instantaneous frequency and amplitude. Then one explored the relationship between vegetation changes and different cropping systems and climatic factors at different scales.
The constant and moving bases are applicable to regions where the main plant species of vegetation cover are unchanged or have changed significantly. The results of the constant base-derived RESTREND indicate that Taipei-Miaoli, Nantou, Jianan-Gaoping, Huadong Rift Valley regions have significant greening trends, and there are documented records of the increase in cropland areas in Taoyuan and Hsinchu. The moving base-derived test results indicate that there are browning trends in cropland regions such as central plains, hills, and Huadong Rift Valley. It also has been found that the plant phenology pattern has changed in some regions, for example, crop type has changed from paddy rice to dry farming or fruit trees. This method can be applied to decision makers to detect the trend of vegetation changes on a large scale, which can help to lock down browning regions and carry out investigations on whether the plant cover and species are reduced due to anthropogenic development or inappropriate land use.
The results of time-frequency analysis demonstrate that different cropping systems are related to the short periodicity of NDVI from 1 season to half a year, and the annual periodicity reflects the seasonal change of temperature. The fluctuance in the amplitude of vegetation and climate spectrum correspond to the agricultural damage caused by extreme climate events such as typhoons, heavy rain, drought, or warm winter. The analysis of frequency characteristics is conducive for understanding how vegetation changes in cropland have responded to changes in cropping system or extreme climate events, and provide information to help make agricultural management decisions.

並列關鍵字

Normalized differential vegetation index ； Trend analysis ； Mann-Kendall trend test ； RESTREND ； Time-integrated NDVI ； Time-frequency analysis ； Hilbert-Huang transform ； CEEMDAN ； Cropping system ； Inverse distance weighted

參考文獻

Abrams, M., Hook, S., Ramachandran, B. (2002). ASTER user handbook, version 2. Jet Propulsion Laboratory, 4800, 135.

Arun, P. V. (2013). A comparative analysis of different DEM interpolation methods. The Egyptian Journal of Remote Sensing and Space Science, 16(2), 133-139.

Asrar, G. (1989). Estimation of plant-canopy attributes from spectral reflectance measurements. Theory and Application of Optical Remote Sensing, 252-296.

Assessment, M. E. (2005). Ecosystems and human well-being (Vol. 5). Washington, DC: Island Press.

Blain, G. C. (2013). The modified Mann-Kendall test: on the performance of three variance correction approaches. Bragantia, 72(4), 416-425.