| dc.description.abstract | 地電阻影像剖面法（Electrical Resistivity Imaging Method, ERI）是一種有效重現地下電性構造的技術，但要取得準確的電阻率模型仍具有挑戰性。本研究估算電阻率不確定性並評估常用陣列的對於模型的重現能力：雙偶極（Dipole-Dipole, DD）、單偶極（Pole-Dipole, PD）、溫奈施朗卜吉(Wenner-Schlumberger, WS)及單極法（Pole-Pole, PP）。本研究運用Python模組程式評估各陣列的性能，並進行聯合反演算，降低單一方法的固有不確定性。本研究運用殘差、重複誤差及潛在雜訊、人造物、反演算誤差和模型精度影響評估地電阻量測及模型不確定性。此外，也使用棋盤格模型測試評估空間分辨率和對電阻率擾動的靈敏度，並根據調查深度 (Depth of Investigation, DOI)、異常分析、統計方差和圖像相關性來檢查陣列成像效率。本研究對三相狀態的飽和沉積物（如固體岩石基質、液態水和空氣）進行了 ERI 和 SRT 提供的岩石物理進行聯合反演。ERI 測量不確定性結果顯示，旱季有約 3.2%、濕季有 0.83% 的資料點會高於重複性誤差的3%臨界值，一般來說，在乾季狀態下的地電阻量測會有較高的接地電阻、重複性誤差及殘餘物差，進而造成較多被排除使用的資料點。DOI指數表示模型重現的真實程度，DOI 的閾值深度會隨測量雜訊增加和測量深度改變而減少。基於實驗中的空洞目標物及模型模擬結果顯示，對於較淺的空洞目標物（2.2公尺深），DD陣列的結果有最高的異常效應（1.46）及電阻率變化（24400 .m），PP陣列則有最低的異常效應(0.6)及電阻率變化（2401.m）。反演模型顯示在更深的模型，其解析度和精準度會下降，從而在電阻率模型解釋上產生分歧。本研究推斷 DD 陣列最適合地下目標研究，PD 及WS陣列同樣足夠作為探索目標結構物，然而PP陣列則最為不適合。研究結果顯示，在所有測試的反演模組中（pyGIMLi、BERT、ResIPy 和 SimPEG），較大的異常目標都得能正確解析，然而，在深度3公尺所設置的目標半徑小於 0.5 公尺時，上述模組並沒有顯示任何目標特徵 。以 pyGIMLi、BERT 和 SimPEG 而言，可以重現大於其深度四分之一的目標直徑，而 ResIPy 可以重現大於其深度三分之一倍的目標直徑。 另外，ERT 和 SRT 的聯合反演則充分重現結構，減少了模型的歧義。總體而言，本研究探討了 ERT 測量和模型不確定性，並概述了能有效成像的陣列和反演模組，顯示使用聯合反演算和約束反演算對單一方法的約束，建議未來的將此方法作為應用。 | zh_TW |
| dc.description.abstract | Although electrical resistivity imaging (ERI) methods are effective for recovering subsurface structures, obtaining an accurate resistivity model is challenging. This study evaluates the recoverability and resistivity uncertainty of commonly used imaging arrays such as dipole-dipole (DD), pole-dipole (PD), Wenner-Schlumberger (WS), and pole-pole (PP). The study also evaluates the performance of free Python-based inversion software and employs joint inversion to reduce the inherent uncertainty of the single method. The resistivity measurement and model uncertainties are investigated using reciprocal error, repeatability error, potential noise, artificial effect, inversion data misfit, and model accuracy. Checkerboard tests are also used to evaluate spatial resolution and sensitivity to resistivity perturbations. The depth of investigation (DOI), anomaly analysis, statistical variances, and image correlations are used to evaluate array imaging efficiencies. The study combines ERI and seismic refraction tomography (SRT) inversion for saturated sediments with three-phase volumetric fractions, such as solid rock matrix, liquid water, and air. According to the ERI measurement uncertainty result, approximately 3.2%t of the dry season datasets and 0.83% of the wet season datasets are above the 3% repeatability error cut-off values. In dry conditions, resistivity measurements generally show high contact resistance, repeatability, and reciprocal errors, resulting in significant data discarding. The DOI threshold depth decreases as measurement noise and survey depths increase. The DD array recovered the highest anomaly index (1.46) and variance (24400 .m) in resistivity data, whereas the PP array recovered the lowest anomaly index (0.60) and variance (2401 .m) for the shallowest target cavity set at 2.2 m depth. At deeper depths, the inverted models exhibit a reduction in model resolution and accuracy, resulting in ambiguity in resistivity model interpretations. Our study shows that the DD array is best suited for subsurface target research. The PD and WS arrays are adequate for surveying target structures, while the PP array is the least suitable. The study results show that while all of the tested inversion freeware packages (pyGIMLi, BERT, ResIPy, and SimPEG) correctly resolve larger anomaly targets, they do not display any signatures for target radius less than 0.5 m set at 3 m depth. ResIPy can reproduce a target diameter greater than one-fourth of its depth, while pyGIMLi, BERT, and SimPEG can recover a target diameter greater than one-fourth of its depth. Furthermore, the joint inversion of ERI and SRT recovered the layer structures adequately, reducing model ambiguity. Overall, this study addresses ERI measurement and model uncertainties, describes effective imaging arrays and inversion freeware packages, and demonstrates single method constraints using joint inversion and constrained inversion, recommending them for future applications. | en_US |