English  |  正體中文  |  简体中文  |  全文筆數/總筆數 : 96218/96218 (100%)
造訪人次 : 92462888      線上人數 : 247
RC Version 7.0 © Powered By DSPACE, MIT. Enhanced by NTU Library IR team.
搜尋範圍 查詢小技巧:
  • 您可在西文檢索詞彙前後加上"雙引號",以獲取較精準的檢索結果
  • 若欲以作者姓名搜尋,建議至進階搜尋限定作者欄位,可獲得較完整資料
  • 進階搜尋


    請使用永久網址來引用或連結此文件: https://ir.lib.ncu.edu.tw/handle/987654321/111306


    題名: 屏東平原地下水補注潛勢評估:基於 THMC 模型與機器學習整合之非飽和水力傳導係數預測;Recharge Assessment in Pingtung Plain: Predicting Unsaturated Hydraulic Conductivity via THMC – Machine Learning Integration
    作者: 武金紅;HUONG, VO THI KIM
    貢獻者: 應用地質研究所
    關鍵詞: 非飽和帶;地下水補注潛勢;垂向水力傳導係數;人工神經網路;隨機森林;unsaturated zone;potential recharge;vertical hydraulic conductivity;Artificial Neural Network;Random Forest
    日期: 2026-07-28
    上傳時間: 2026-10-20 17:15:02 (UTC+8)
    出版者: 國立中央大學
    摘要: 地下水補注對於含水層的永續利用至關重要,尤其是在高度依賴地下水資源的地區。在地下水補注過程中,非飽和帶扮演關鍵角色,其中**垂向水力傳導係數(Kzz)**控制入滲水到達地下水位的速率。然而,目前對於沖積系統Kzz 的空間分布仍缺乏系統性的研究。
    本研究結合 THMC 地下水流動模式與機器學習,針對臺灣南部屏東平原沖積層建立 Kzz 空間分布圖,並探討其主要控制因子。THMC 模式利用 2021 至 2024 年間 65 口地下水觀測井之地下水位資料進行校正,建立包含 3,606 個節點之 Kzz 空間分布資料集。接著,以校正所得之 Kzz 作為目標變數,利用 27 項環境與土壤因子建立人工神經網路(Artificial Neural Network, ANN)與隨機森林(Random Forest, RF)模型。這些輸入因子可分為五大類,包括:空間特徵(3項)、土壤性質(8項)、地形與水文特徵(3項)、遙測特徵(3項)以及土地利用組成(10項)。此外,本研究採用 SHAP(SHapley Additive exPlanations)**分析,以解釋 ANN 模型結果並辨識影響 Kzz 預測的重要因子。
    THMC 模式在 F1 與 F2 含水層之地下水位校正結果,中位數 R² 分別為 0.721 與 0.710,顯示模式具有可接受之模擬能力。校正後之 Kzz 在沿海平原地區(東港、崁頂)介於 4.27 × 10⁻⁴ 至 8.6 × 10⁻³ m/day,而在近山麓沖積扇地區(長治、鹽埔、高樹)則介於 2.0 至 6.89 m/day,反映沉積物由山前向海岸逐漸細粒化之特性。ANN 與 RF 模型於測試資料之 R² 分別達 0.847 與 0.862。SHAP 分析結果顯示,地表高程與距離沖積扇頂點距離為影響 Kzz 最重要的控制因子,其次為礫石含量與黏土含量。
    研究結果顯示,地形位置與沉積物組成共同控制屏東平原 Kzz 的空間分布特性。地下水補注潛勢在東北側近山麓沖積扇區域最高,而於沖積扇末端及沿海地區最低。本研究證明,在現地量測資料有限的情況下,機器學習可有效預測沖積系統中的 Kzz。然而,在建立機器學習模型之前,仍須充分了解研究區之水文地質特性與環境背景,以確認所選輸入因子具有合理的物理意義與可解釋性。;Groundwater recharge is essential for aquifer sustainability, especially in regions that depend heavily on subsurface water. In the recharge process, the unsaturated zone plays a key role, as vertical hydraulic conductivity (Kzz) controls how fast infiltrated water reaches the water table. However, the spatial distribution of Kzz across alluvial systems has not yet been systematically investigated.
    This study combines a THMC groundwater flow model with machine learning to map Kzz and identify its controlling factors across the Pingtung Plain alluvial deposit, southern Taiwan. The THMC model was calibrated using observed water levels from 65 monitoring wells (2021 to 2024) and produced a spatially distributed Kzz dataset of 3,606 nodes. These calibrated values were then used to train an Artificial Neural Network (ANN) and a Random Forest (RF) using 27 environmental and soil covariates grouped into five categories: spatial features (3), soil properties (8), terrain and hydrological features (3), remote sensing features (3), and land use composition (10). SHAP analysis was applied to interpret ANN results and identify which features most influenced the predicted Kzz.
    The THMC model achieved median R² of 0.721 and 0.710 for the F1 and F2 aquifer units, indicating acceptable performance. Calibrated Kzz ranges from 4.27 × 10⁻⁴ to 8.6 × 10⁻³ m/day in the coastal plain (Donggang, Kanding) to 2.0 to 6.89 m/day in the proximal fan (Changzhi, Yanpu, Gaoshu), reflecting sediment fining from the mountain front toward the coast. The ANN achieved a test R² of 0.847 and the RF achieved 0.862. SHAP results show that land surface elevation and distance to the fan apex are the strongest controlling factors, followed by gravel content and clay content.
    The findings suggest that terrain position and sediment composition together explain the spatial distribution of Kzz across the Pingtung Plain. Recharge potential is highest in the proximal fan zone in the northeast and lowest in the distal fan and coastal areas. This study demonstrates that machine learning is an effective approach for predicting Kzz in alluvial systems with limited field data. However, prior understanding of the hydrogeological setting and environmental context remains necessary to evaluate whether the selected input features are physically reasonable and interpretable.
    顯示於類別:[應用地質研究所] 博碩士論文

    文件中的檔案:

    檔案 描述 大小格式瀏覽次數
    index.html0KbHTML9檢視/開啟


    在NCUIR中所有的資料項目都受到原著作權保護.

    社群 sharing

    ::: Copyright National Central University. | 國立中央大學圖書館版權所有 | 收藏本站 | 設為首頁 | 最佳瀏覽畫面: 1024*768 | 建站日期:8-24-2009 :::
    DSpace Software Copyright © 2002-2004  MIT &  Hewlett-Packard  /   Enhanced by   NTU Library IR team Copyright ©   - 隱私權政策聲明