Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/27812
標題: 利用多頻譜觀測儀觀測地表水體濁度及總固體濃度
Estimation of total solids and turbidity of surface water using Hyperspectral sensing in Taiwan
作者: 黃家慶
Huang, Chia-Ching
關鍵字: Hyperspectral;多頻譜;Multiple-regression;Neural Network;Total solids;Turbidity;複迴歸;類神經網路;總固體
出版社: 土壤環境科學系所
摘要: 
This study is to establish empirical relationship between spectral reflectance and TS and turbidity from hyperspectral measurements. Several factors can affect the spectral reflectance characteristics; these include mainly suspended sediments, phytoplankton biomass (e.g., chlorophyll-a), and dissolved organic carbon. The objective of this research is to explore reliable and fast methods for water quality monitoring in terms of total solids in rivers of central Taiwan. Reflective radiance was made simultaneously with total solids and turbidity measurements of freshwater rivers in central Taiwan. The optimal wavelength is identified when correlation coefficients between the reflective radiance and TS or turbidity are maximum using both the stepwise approach, i.e., multiple-regression and neural network techniques. The aim of was to use neural network and multiple-regression methods in conjunction with field water quality monitoring using remote sensing system as a means to predict water quality changes. It was to identify conditions under which neural network or multiple-regression can be better applied to water quality monitoring .Different sample number affect the performance of both the neural network and the multiple-regression methods. We find that if sample number was less than 25, neural network is better than multiple-regression. The study was also to test the feasibility of applying satellite generated data to river water quality monitoring in Taiwan. Results showed if only the spectral resolution of the satellite was of concern, MODIS would be the best satellite sensor for water quality analysis.
The results show that water quality variables correlate well with those predicted with both the neural network and the multiple-regression approaches. The degree of correlation between the reflective radiance and turbidity is better than that with total solids. Particle size and type of minerals in the suspended solids appear to be important factors controlling the correlation coefficients.

本研究目的在利用遙測資料建立台灣地面水體水質推估模式,在水體水質之觀測上,由於水體本身對電磁波譜反射率極低,因此取決於水體反射率大小的主要是水體中的物質如葉綠素、有色可溶性有機物質(coloured dissolved organic matter) (CDOM)以及總固體(Lahet et al., 2001; Pattiaratchi et al., 1994)。因此本研究目標將以台灣河川為主要觀測對象,使用多頻譜觀測儀來分析河川水體中總固體及濁度反射波譜,以建立反射波譜與水體中濁度及總固體之推估模式。在遙測資料之處理及模式之建立,本研究將比較前人文獻結果、電磁波譜反射率曲線圖、統計複迴歸分析及類神經網路瞭解各分析方法之優缺點及適用範圍。
依據分析結果,具有觀測水體中總固體及濁度之波長範圍為400~920 nm。其中又以700~900 nm 較具代表性,使用複迴歸及類神經網路分析水體水質,濁度結果推估之R2分別為0.88及0.87,而總固體則為0.63及0.66。此外比較類神經網路與複迴歸分析,此兩種模式雖然都具有不錯之推估效果,但若改變其樣品數,則可發現在較少的樣品數(<25)時以類神經網路的印證效果為佳,而若是可獲得較多的樣品時,則以複迴歸分析較為準確。在衛星推估上,以地面高光譜資料模擬衛星各波段並以總固體及類神經網路推估,類神經網路在模擬衛星的推估上優於複迴歸分析,尤其在較低波譜解析力之衛星上如Landsat TM更為明顯。此外具較高波譜解析力之衛星如MODIS其推估效果也較低波譜解析力之衛星來的好。
若探討總固體R2較濁度低之原因,推測可能濁度為表示光入射水體後被散射的電磁波強度,其反映水體整體之光散射狀態,而總固體僅為影響水體光散射的因子之一,故以後作為後續遙測水質研究時,其土壤粒徑大小與礦物性質都需詳加探討。
URI: http://hdl.handle.net/11455/27812
Appears in Collections:土壤環境科學系

Show full item record
 

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.