圖書典藏及影音數位平台

台灣地區因水庫集水區地形陡峭、地震頻繁，近年來由於人口與經濟快速成長，以及集水區土地之過度開發利用，且颱風季節常常發生豪大雨，容易造成表土嚴重沖蝕，其夾帶大量的農藥與肥料(營養源)進入水庫水體，致使水體中之養分如氮、磷等濃度過高，加速了湖泊優養化現象，破壞水生態體系，使水質惡化因而影響了水體之正常用途，若能近即時以影像觀測與模式預測，有利決策執行單位進行處理。目前曾文水庫共有六個水質監測站，現場採樣每季（3個月）辦理一次，一年約有4筆水質監測紀錄（2月、5月、8月與11月），採樣後的水體需專人員檢驗，且樣本檢驗需要耗費許多時間，因為受到時間與成本的限制，所以採樣的樣本數通常不多，僅能獲取點狀式的資料，無法代表整個水庫的水質狀況。然而，近年來遙測技術不斷發展，使用寬波段反射及其它特性與水體之間的關聯性，已廣泛的應用於各類水體污染的監測。遙感探測技術具有許多優點，包括可進行大面積水體快速同步監測，不受時間與地點的限制，且透過不同季節及不同時間的連續監測，並重複成像，便於瞭解和掌握水體水質的變化趨勢。遙測技術的出現及其在水質監測領域的應用，提高了現代區域性水質監測的科學合理性。 本研究第一年利用各類衛星影像，如Formosat-2與SPOT-5/6或Landsat-7或Aster等衛星進行監測，水庫範圍一張衛星影像即可覆蓋，故期望於每季現地採樣時間同時，可配合獲取相關各類衛星影像資料，以評估遙測對水庫水質監測之適用性與準確度，建構豐估期之水庫水質參數與各類衛星波段間的關連性，結果可建議不同水質搭配適合的衛星影像（精確），以及建立各類衛星之通式模式，可利用各類衛星影像快速監測水庫水質狀況（略估），並探討衛星影像與現地水質資料日期差異的最大容許度，目標為探討水庫水質的空間變異程度。

The water quality of a reservoir is an important factor in evaluating and analyzing the reservoir’s status and providing a reference for decision making. The quarterly and monthly water quality monitoring data from EPA and South Region Water Resources Office provide the basic water status as well as long-term and seasonal changes of a reservoir; however, the data merely provide limited spatial information. Spatial information on water quality can be utilized to confirm the source of pollution, help determine pollution hotspots, and fully evaluate the impact of natural disasters and pollution accidents. Hence, such information is indispensable for supporting management decisions on water quality. To help reservoir management units effectively master spatial information on water quality to make decisions, high temporal resolution telemetric image is employed in this study to analyze the water quality of a reservoir and establish a water quality management aided decision-making system. In the first year of the study, the researchers mainly collect satellite imagery and water quality monitoring data from 2004, establish the water quality image process flow and a water quality regression model, evaluate the error of telemetric water quality, and produce 560 water quality images of Tsengwen Reservoir to be used in evaluating the feasibility of satellite telemetric imagery for water quality monitoring and management. The revisit rate for Tsengwen Reservoir in the past 10 years and FORMOSAT-2 imagery with the highest autonomy are utilized as data sources. After dislocation matching, ortho-rectification, geometric registration, and relative radiation correction of imagery are completed, sampling points of water quality and site monitoring data are selected to establish linear and non-linear water quality regression models. Test results show that the spectral values of imagery measured and corrected through the reflectivity of site radiation are better than original data. The regression model based on water quality data a day away from the day the imagery is taken has a higher coefficient of determination (R2). If water quality is divided into high-water period and low-water period to establish the model, the error can be further reduced. The non-linear water quality regression model established based on an artificial neutral network is more accurate than the linear water quality regression model. Water quality imagery is generated by means of the optimal regression model. Preliminary results show that the spatial distribution of water quality imagery can not only reflect increased turbidity, sharply reduced transparency, increased nutritive salt load that causes an increase in chlorophyll content, and rainfall erosion that cause an increase in total phosphorus content after a typhoon event, but also shows reasonable distribution of high-content hotspots. For example, the turbidity of the upstream section, total phosphorus content, and content of suspended matter are high. The content of chlorophyll in the inner area of the reservoir is higher than that in the outer reservoir area. In addition, it shows the transport and distribution of pollutants to the dam area. This study adopts multiple observation data (imagery and water quality data) to establish the regression model. However, to apply the empirical model, the future management unit should add spectral measurement in routine water quality sampling to continuously accumulate data and improve the reliability and detection limit of the regression model. A semi-automatic water quality image module is completed. A suggestion is to establish a history water quality image analysis and management system to enhance the management of the catchment area.