DSpace community: 遙測科技碩士學位學程

基於數位孿生基礎建設耦合數值天氣預報模式及都市淹水模式建立都市淹水預報;A Digital Twin Urban Flood Forecasting System Coupling a Numerical Weather Prediction Model and an Urban Flood Model

title: 基於數位孿生基礎建設耦合數值天氣預報模式及都市淹水模式建立都市淹水預報;A Digital Twin Urban Flood Forecasting System Coupling a Numerical Weather Prediction Model and an Urban Flood Model abstract: 在極端氣候加劇與都市化快速擴張的雙重衝擊下，台灣都市地區淹水風險日益嚴峻。傳統災防系統普遍存在資料異質、採樣密度不均與應變整合機制缺乏等問題，難以即時應對複雜的降雨與城市環境變化。為強化都市防災韌性，本研究導入數位孿生（Digital Twin）技術，發展一套具備高解析度、即時性與決策支援功能的都市淹水預報與導航系統。研究首先建構一套模組化、可擴充與雲端運算導向的數位孿生基礎建設，實現異質物聯網感測資料處理與儲存、以及進入運算流的全自動流程。接著基於此建設耦合數值天氣預報模式及都市淹水模式，整合衛星資料、IoT感測資料，產製10公尺解析度72小時逐時都市淹水預報。在應用層面，本研究進一步發展淹水導航服務，結合OpenStreetMap道路資料與淹水預報成果，建立可反映淹水災情的可通行路網，並實作基於Dijkstra演算法之避災路徑規劃功能，以利防災應變。本研究以台南市善化區為研究區域，應用本系統於2021年強降雨及2024年凱米颱風之案例，結果顯示其具備基本淹水預測能力，並能即時提供避災導航資訊，驗證了數位孿生技術於智慧城市虛實整合應用之潛力。;Due to intensifying impacts of extreme weather and rapid urbanization, the flood risk in urban areas of Taiwan has become increasingly severe. Traditional disaster prevention systems often suffer from issues such as heterogeneous data, uneven sampling density, and lack of integrated emergency response mechanisms, making it difficult to effectively respond to complex rainfall events and urban environmental changes. To enhance urban disaster resilience, this study introduces a Digital Twin framework to develop a high-resolution, efficient, and decision-support-capable urban flood forecasting and navigation system. First, this study constructs a modular, scalable, and cloud-based digital twin system, enabling automated workflows for processing and storing heterogeneous IoT sensor data into application streams. Based on this system, this study couples a numerical weather prediction (NWP) model with an urban flood model, integrating satellite data and IoT sensor data to produce 72-hour flood forecasts at 10-meter and hourly resolution. In terms of application, the study further develops a flood navigation service by combining OpenStreetMap road data with flood forecasting results to create a flood-aware, drivable road network. It also implements an evacuation route planning feature based on the Dijkstra algorithm to support emergency response.The study selected Shanhua District in Tainan City as the study area and applied the system to cases of heavy rainfall in 2021 and Typhoon Gaemi in 2024. Results indicate that the system exhibits fundamental flood forecasting capabilities and can provide efficient disaster navigation information, demonstrating the potential of digital twin technology for cyber-physical integration in smart cities.

Investigation of Equatorial Plasma Bubble Observations in the Taiwan-Philippine Region Using GNSS Receiving Network and Hualien VIPIR Station (2023-2024)

title: Investigation of Equatorial Plasma Bubble Observations in the Taiwan-Philippine Region Using GNSS Receiving Network and Hualien VIPIR Station (2023-2024) abstract: This study investigated the occurrence frequency, perturbation intensity, and spatial morphology of equatorial plasma bubbles (EPBs) in Taiwan from 2023 to 2024, during which solar activity increased significantly. Through the unified Rate of Total Electron Content Index (ROTI) threshold value (>0.9 TECU/min), this study found a significant increase in EPB events, from 74 in 2023 to 128 in 2024, which is closely related to the increase in solar flux and the increase in sunspot count. The intensity of flickering is measured by the VS4 index and peaks during the spring and autumn equinoxes, but significant activity is also observed during the summer solstices such as May and August, indicating that EPB formation can be extended to atypical seasons during the solar maximum. Latitude analysis shows that higher VS4 values correspond to a larger EPB extension range, up to about 28 degrees of latitude. The auxiliary data provided by the VIPIR ionospheric detector further validated the changes in the underlying ionosphere during the EPB event, with significant increases in h′F2 and foF2 values, consistent with the VS4 high-value event. The findings highlight the impact of solar driving forces on EPB behavior and highlight the importance of integrated diagnostic techniques in low-latitude GNSS vulnerability assessment.;This study explores the occurrence, intensity, and spatial morphology of Equatorial Plasma Bubbles (EPBs) over Taiwan during 2023–2024, a period marked by elevated solar activity. Using a consistent ROTI threshold (>0.9 TECU/min), EPB detection revealed a substantial increase in events—from 74 in 2023 to 128 in 2024—closely tied to rising solar flux and sunspot numbers. Scintillation intensity, measured via the VS4 index, peaked during equinoctial months, but solstitial periods like May and August also showed notable activity, indicating solar maxima broaden EPB development beyond typical seasonal norms. Latitudinal analysis showed stronger VS4 values corresponded to greater EPB spread, up to ~28°. Supporting data from VIPIR ionosonde observations confirmed bottom-side ionospheric responses, with elevated h’F2 and foF2 values matching intense EPB episodes. These findings highlight the influence of solar forcing on EPB behavior and underscore the importance of integrated diagnostics for GNSS vulnerability assessment in equatorial regions.

坦尚尼亞姆貝亞和松圭地區環境因素影響土地覆蓋變化的地理空間分析（2018-2022年）;Geospatial Analysis of Environmental Factors Influencing Land Cover Change in Mbeya and Songwe, Tanzania (2018-2022)

title: 坦尚尼亞姆貝亞和松圭地區環境因素影響土地覆蓋變化的地理空間分析（2018-2022年）;Geospatial Analysis of Environmental Factors Influencing Land Cover Change in Mbeya and Songwe, Tanzania (2018-2022) abstract: 摘要土地覆蓋變遷為人類對地球系統最主要的改變形式之一，對環境退化具有顯著影響。本研究針對坦尚尼亞南部高地的姆貝亞（Mbeya）與松圭（Songwe）地區，進行2018年至2022年間多時期土地覆蓋變遷動態分析。研究量化了環境衝擊，評估地表溫度（Land Surface Temperature, LST）與正規化植生指數（Normalized Difference Vegetation Index, NDVI）之間的關係，並辨識出主要的環境驅動因子。本研究運用隨機森林分類法處理多時期Sentinel-2（10公尺解析度）衛星影像，產製土地覆蓋分類圖，其總體分類精度達86.35%，Kappa係數為83.13%。研究結果顯示當地景觀結構發生明顯變化，其中農業用地擴張為主要變遷因素，增加幅度達41.39%；相對地，森林覆蓋減少9.08%，草地覆蓋則減少48.68%。進一步分析顯示，森林減少面積中有64.1%轉為農地，77.8%轉為灌木地。 NDVI分析顯示顯著的空間與時間變異，其中高地森林地區NDVI值最高（0.57–0.92），而農業與都市地區則相對偏低（0.02–0.44）。LST分析則揭示各類土地覆蓋類型具有明顯的熱特徵變化，高地森林地區溫度最低，約為11.7°C，而裸地與都市區域最高可達44.5°C。研究亦發現明顯的冷卻效應，高植被區溫度比建成區低3–5°C；反之，都市發展則造成熱島效應，使溫度高出鄰近植被區2–4°C。NDVI與LST之間呈現高度負相關（r = -0.738, p < 0.001），顯示植被在熱調節中扮演關鍵角色，其中每增加0.1的NDVI，約可降低地表溫度1.5°C。多元線性回歸模型顯示具中度相關性（R² = 0.5584–0.5867），可解釋55–59%的熱–植生關係變異。環境因子分析結果指出，植被密度為主要的熱調節因子，其次為土地覆蓋類型、季節性氣候模式、地形條件、水資源、人為管理措施與土壤性質等。本研究指出，NDVI高於0.6的高地山地森林具備明顯的蒸散作用與樹冠遮蔽功能，為重要的熱庇護區；相對地，農業強化與都市擴張則加劇熱壓力。此外，植被與溫度之間的關係於不同年度間表現出穩定性，年度變異小於1%，顯示本研究結果具潛力應用於長期環境規劃與氣候調適策略。關鍵詞：土地覆蓋變遷、NDVI、地表溫度、遙測、多元線性回歸 ;Abstract Land cover change represents a primary human alteration to the Earth system, substantially contributing to environmental degradation. This research analyzed the dynamics of multi-pretemporal land cover changes in the Mbeya and Songwe regions of Tanzania′s Southern Highlands between 2018 and 2022. The study quantified environmental impacts, measured the relationships between land surface temperature (LST) and normalized difference vegetation index (NDVI), and identified significant environmental drivers of these processes. Utilizing random forest classification, multi-temporal Sentinel-2 (10m) satellite data yielded land cover maps with an overall accuracy of 86.35% and a Kappa coefficient of 83.13%. The observations indicated landscape changes, with agricultural expansion identified as the principal factor, increasing by 41.39% while forest cover decreased by 9.08% and grasslands diminished by 48.68%. Analysis of the transition matrix indicated that 64.1% of forest reduction was converted to cropland, while 77.8% transitioned to shrubland. The analysis of the Normalized Difference Vegetation Index (NDVI) revealed significant spatial and temporal variations, with highland forested regions exhibiting the highest values (0.57-0.92) compared to lower values in agricultural and urban areas (0.02-0.44). Land Surface Temperature (LST) patterns exhibited clear thermal signatures across various land cover types, with temperatures ranging from 11.7°C in highland forested areas to 44.5°C in bare land and urban environments. The most substantial cooling effect was demonstrated, with temperatures maintained at 3-5°C lower than built-up areas, whereas urban development resulted in heat islands that were 2-4°C warmer than adjacent vegetated regions. A strong negative correlation (r = -0.738, p < 0.001) between NDVI and LST indicates the significant role of vegetation in thermal regulation, with each 0.1 increase in NDVI associated with an approximate 1.5°C decrease in surface temperature. Multiple linear regression models demonstrated moderate predictive capability (R² = 0.5584-0.5867), accounting for 55-59% of the variance in thermal-vegetation relationships. The analysis of environmental factors identified vegetation density as the principal thermal regulator, followed by land cover type, seasonal climate patterns, topographic controls, water availability, human management practices, and soil properties. The study indicates that highland montane forests with NDVI values above 0.6 serve as essential thermal refugia due to evapotranspiration and canopy shading. In contrast, agricultural intensification and urban expansion contribute to considerable thermal stress. The temporal stability of vegetation-temperature relationships, with minimal year effects accounting for less than 1% variance, indicates dependable patterns for long-term environmental planning and climate adaptation strategies. Key words: Land cover change, NDVI, land surface temperature, remote sensing, and multiple linear regression

高光譜影像分類新興深度學習變換器架構之比較：以CTMixer、MAEST與SSTN為例;Comparison of Novel Transformer-based Deep Learning Architecture for Hyperspectral Image Classification: A Case Study of CTMixer, MAEST and SSTN

title: 高光譜影像分類新興深度學習變換器架構之比較：以CTMixer、MAEST與SSTN為例;Comparison of Novel Transformer-based Deep Learning Architecture for Hyperspectral Image Classification: A Case Study of CTMixer, MAEST and SSTN abstract: 隨著遙測（Remote Sensing, RS）與深度學習（Deep Learning, DL）技術的迅速發展，加速了地表物質分類與識別方法的演進與實務應用。高光譜影像（Hyper-Spectral Image, HSI）因具備豐富的光譜與空間資訊，廣泛應用於農業監測、地質探勘及地物分類等領域。然而，其高維度特性與地物類別間的光譜相似性，對於分類任務構成極大挑戰。近年來，深度學習技術迅速發展，特別是結合變換器 (Transformer) 架構的模型在 HSI 分類上展現高度潛力。鑑於這些方法不僅應用於不同數據集，所採用的訓練參數（如訓練週期與學習率）亦有所差異，因此有必要針對其分類表現與運算效率進行更全面的分析與比較。本研究選擇三種近年提出的 Transformer-based 架構進行比較：融合卷積神經網路(CNN) 與 Transformer 結構的 CTMixer、採用遮罩自編碼器設計的 MAEST、以及採用改良式 Swin Transformer 的 SSTN。透過統一的實驗設計與兩組訓練參數配置（CFG1 與 CFG2），分別於 Indian Pines、Pavia University 與 Houston 2013 三組公開數據集上進行分類實驗。研究評估指標涵蓋整體準確率、平均準確率、Kappa 係數、類別分類表現、模型推論時間，並輔以分類地圖進行視覺化分析，以檢視模型在空間邊界識別與分布一致性方面的表現。研究結果顯示，SSTN 在分類準確性與穩定性方面表現最佳，CTMixer 對參數設定較為敏感，適合應用於結構清晰的場景；而MAEST 則具備良好的推論效率，但在光譜相似與樣本不均的情境時，分類表現相對較弱。分類地圖的視覺化進一步揭示各模型在邊界清晰度與區域連續性上的差異：SSTN 可有效維持空間一致性，而 MAEST 在大面積地物的分類上則較易產生破碎現象。綜合比較指出，各架構各具優勢與適用場景，研究結果可作為未來高光譜影像分類任務中，模型選擇與架構設計之有力參考。;In recent years, deep learning techniques have advanced rapidly, with Transformer-based models in particular demonstrating high potential in hyperspectral image classification. Given that these methods are not only applied to different datasets but also adopt varying training parameters (such as training epochs and learning rates), a more comprehensive analysis and comparison of their classification performance and computational efficiency is necessary. This study compares three recently proposed Transformer-based architectures: CTMixer, which integrates Convolutional Neural Network (CNN) and Transformer structures; MAEST, which employs a masked autoencoder design; and SSTN, which adopts a modified Swin Transformer. Through a unified experimental setup and two training parameter configurations (CFG1 and CFG2), classification experiments are conducted on three publicly available datasets: Indian Pines, Pavia University, and Houston 2013. Evaluation metrics include overall accuracy, average accuracy, kappa coefficient, class-wise classification performance, and model inference time. In addition, classification maps are visualized to examine the models′ ability in boundary recognition and spatial distribution consistency. The experimental results show that SSTN achieves the best performance in terms of classification accuracy and stability. CTMixer is more sensitive to parameter settings and is better suited for scenes with clear spatial structures. MAEST demonstrates good inference efficiency but exhibits weaker classification performance in scenarios with high spectral similarity and sample imbalance. The visualization of classification maps further reveals differences in boundary clarity and regional continuity among the models: SSTN effectively maintains spatial consistency, whereas MAEST tends to produce fragmented regions when classifying large-area objects. In summary, each architecture has its strengths and suitable application scenarios. The results of this study can serve as a valuable reference for model selection and architectural design in future HSI classification tasks.