日期：11/20(五) 13:30-15:30
地點：國家科學及技術委員會2樓研討室 (台北市大安區和平東路二段106號)
主持人：羅敏輝 教授 (國立臺灣大學大氣科學系，大氣領域召集人)、楊明德 院長 (國立中興大學工學院)
| 時間 | 講題 | 講者 |
|---|---|---|
| 13:30-13:38 | 福衛衛星科學軟體於關鍵水庫安全應用研究 | 翁孟嘉 教授 (國立臺灣大學土木工程學系) |
| 13:38-13:46 | From Simulated Sea Surfaces to Software-Defined Radar Hardware: Building a C-Band Simulation-to-Measurement Chain for Ocean Wind Sensing | 古瓊昇 助理教授 (國立臺北科技大學太空系統工程研究所)、張陽郎 教授 (國立臺北科技大學電機工程系) |
| 13:46-13:54 | SAR成像與光學影像異質融合技術及應用於碳儲量變化之驗證 | 李佩君 教授 (國立臺灣科技大學電子工程系) |
| 13:54-14:02 | 次米級X波段SAR與高解析光學影像融合技術驅動精準作物分類 | 林玉儂 副研究員 (中央研究院地球科學研究所) |
| 14:02-14:10 | 利用多時序與多感測衛星遙測影像融合提升都會區與基礎設施監測能力 | 莊昀叡 教授 (國立臺灣大學土木工程學系) |
| 14:10-14:18 | 基於星上邊緣運算架構之輕量化衛星資料分析技術建立與應用 | 蔡慧萍 副教授 (國立中興大學土木工程學系) |
| 14:18-14:26 | Magnetic Storm Effects Observed by FORMOSAT-7/COSMIC-2 | 林建宏 教授 (國立成功大學地球科學系) |
| 14:26-14:34 | 自主式太空天氣預報系統研發 | 陳佳宏 副教授 (國立成功大學地球科學系) |
| 14:34-14:42 | 透過共同整合多元衛星資料促進福衛產品在大氣對流過程研究之應用 | 陳維婷 教授 (國立臺灣大學大氣科學系暨研究所) |
| 14:42-14:50 | 增進水氣資訊之監測及推估於福衛星系之驗證與未來衛星任務效益評估 (TBC) | 劉千義 副研究員 (中央研究院環境變遷研究中心) |
| 14:50-14:58 | 獵風者衛星GNSS資料加值與應用 | 楊舒芝 特聘教授 (國立中央大學大氣科學學系) |
| 14:58-15:06 | 利用福衛星系加值資料改善低對流層天氣分析與預報 | 陳舒雅 研究員 (國立中央大學全球大氣觀測與資料應用研究中心) |
| 15:06-15:14 | 溫室氣體衛星遙測資料之地面驗證與排放監測應用 | 陳怡均 副研究員 (中央研究院環境變遷研究中心) |
| 15:14-15:30 | 綜合討論 |
主持人簡介
羅敏輝 教授 (國立臺灣大學大氣科學系，大氣領域召集人)
楊明德 院長 (國立中興大學工學院)
楊明德教授現任國立中興大學土木工程學系終身特聘教授兼工學院院長，獲美國俄亥俄州立大學土木及環境工程博士學位。曾任土木工程學系主任、創新產業推廣學院副院長，並主持國科會及農業部智慧農業及無人載具相關研究計畫。專長涵蓋人工智慧、地理資訊系統、遙測及環境監測，曾榮獲科技部傑出研究獎、未來科技獎及國家新創獎等殊榮。
講者簡介
翁孟嘉 教授 (國立臺灣大學土木工程學系)
國立臺灣大學土木工程博士，具土木、大地及結構技師資格，專長岩石力學、工程地質、數值分析及地質災害防治，現任臺灣大學土木工程學系教授，曾主持水利署水庫安全評估計畫及參與相關技術規範修訂。已發表期刊論文90餘篇，獲多項國內外學術獎項。現任中華民國大地工程學會理事長、岩盤工程暨工程地質學會副理事長。
古瓊昇 助理教授 (國立臺北科技大學太空系統工程研究所)
Chiung-Shen Ku received the B.S. degree in Physics from National Central University, Taiwan, in 2005, the M.S. degree in Remote Sensing Science and Technology from National Central University, Taiwan, in 2012, and the Ph.D. degree in Communication Engineering from National Central University, Taiwan, in 2020. He is currently an Assistant Professor with the Institute of Aerospace and Systems Engineering, National Taipei University of Technology, Taiwan. His research interests include synthetic aperture radar system design and imaging algorithms, radar signal processing, microwave remote sensing, wireless communications, microwave circuits, and antenna measurements.
張陽郎 教授 (國立臺北科技大學電機工程系)
李佩君 教授 (國立臺灣科技大學電子工程系)
臺灣科技大學 電子系教授
2026.01-迄今：IEEE CTSoc BoG
2023.01 -2025.12:IEEE CTSoc, WIE Chair/ Region 10 Director
2022.03 – 迄今： IEEE CTSoc Taipei Chapter Chair
2020.10 – 迄今： IET Taipei chapter 理事
2018.06 – 2022.10： 中華民國消費電子學會 理事長
研究專長 / Research Expertise
人工智慧之機器視覺預估、衛星影像、2D/3D 轉換技術、Multi-view video coding、視訊編碼、視訊傳輸、影像處理及應用、FPGA 設計與應用、IC電路設計
林玉儂 副研究員 (中央研究院地球科學研究所)
專注合成孔徑雷達研究近二十年。早期深耕干涉與大氣校正技術，探討地殼變形、地震週期與坡地崩塌前兆；近期研發層析成像技術，精準偵測都會建物變形以支援都更規劃，並推展農林環境情報之鷹用，涵蓋作物分類、物候、與森林生物量監測，以呼應國家太空計畫，回應永續社會之需求。
莊昀叡 教授 (國立臺灣大學土木工程學系)
（簡介待補）
蔡慧萍 副教授 (國立中興大學土木工程學系)
Associate Professor Tsai specializes in remote sensing, image processing, spectral analysis, and spatial change monitoring, with research focusing on environmental and Earth observation applications.
林建宏 教授 (國立成功大學地球科學系)
Scientist for space science and satellite missions. Have been involved in almost all FORMOSAT missions.
陳佳宏 副教授 (國立成功大學地球科學系)
Curriculum Vitae
– Name: Dr. Chia-Hung Chen (Koichi)
– Current Position: Associate Professor, Department of Earth Sciences, National Cheng Kung University, Taiwan
– Education: Ph.D. in Earth and Planetary Sciences, Kyoto University, Japan (2012)
– Research Metrics: Total Citations: 6,017 | H-index: 44 (Google Scholar, https://reurl.cc/DYeyo5)
陳維婷 教授 (國立臺灣大學大氣科學系暨研究所)
Dr. Wei-Ting Chen’s research has centered on organized convection in the tropics and the Asian monsoon region, with emphasis on three tightly connected questions: (1) how loosely scattered convection becomes organized and aggregated over the open ocean and over complex-terrain islands, thereby increasing the likelihood of extreme rainfall; (2) what synoptic-scale environmental conditions promote the initiation and maintenance of organized convection; and (3) how convection feeds back on regional circulation, moisture transport, and energy budgets, yielding characteristics of convection aggregation at specific temporal and spatial scales. The overarching goal of this research is to establish a physically based, cross-scale understanding that strengthens confidence in projections of future changes in weather dominated by organized convection and rainfall extremes under climate change.
To address these important but challenging multi-scale couplings, Dr. Chen has progressively developed an integrated and holistic research strategy that strategically links a hierarchy of modeling tools—large-eddy simulations (LES), cloud-resolving models (CRMs), the superparameterized global model (SPCAM), global storm-resolving models (GSRMs), and conventional global climate models—with multi-platform observations (satellite products and surface networks) and targeted field campaigns. This strategy forms a storyline-centered research framework that integrates observation-based statistical diagnostics, model experiment design and mechanism identification, targeted observational verification, and evaluation and improvement of global models.
劉千義 副研究員 (中央研究院環境變遷研究中心)
楊舒芝 特聘教授 (國立中央大學大氣科學學系)
Shu-Chih Yang received her B.S. and M.S. degrees in Atmospheric Sciences from National Central University (NCU), Taiwan, in 1997 and 1999, respectively, and her Ph.D. in Meteorology from the University of Maryland, USA, in 2006.
She is currently a Distinguished Professor in the Department of Atmospheric Sciences at National Central University and serves as the Director of the Global Atmospheric Observation and Data Application Research Center (GAODARC) at NCU. Her research interests encompass data assimilation, atmospheric predictability, and severe weather prediction, with a strong focus on utilizing atmospheric remote sensing observations, including radar and GNSS-based measurements.
She currently leads a National Science and Technology Council (NSTC) project focusing on the value-added applications of TRITON GNSS-R satellite data.
陳舒雅 研究員 (國立中央大學全球大氣觀測與資料應用研究中心)
陳舒雅博士現任國立中央大學全球大氣觀測與資料應用研究中心研究員。研究專長為全球導航衛星系統(GNSS)掩星觀測、資料同化及數值天氣模擬，致力於GNSS衛星觀測資料在颱風與劇烈天氣預報之應用。長期參與福衛三號、福衛七號及新一代衛星任務之資料分析與應用研究，積極推動國際學術合作與衛星氣象應用發展，並多次榮獲國立中央大學研究傑出獎。
陳怡均 副研究員 (中央研究院環境變遷研究中心)
Yi-Chun Chen is an Associate Research Fellow at the Research Center for Environmental Changes, Academia Sinica, Taiwan. She received her Ph.D. in Environmental Science and Engineering from the California Institute of Technology and previously conducted postdoctoral research at NASA’s Jet Propulsion Laboratory. Her research encompasses satellite-based emission assessment, meteorological influences on regional air quality, aerosol–cloud–climate interactions, and the application of satellite observations and data-driven approaches to air quality assessment and greenhouse gas monitoring.
講題摘要
福衛衛星科學軟體於關鍵水庫安全應用研究
講者：翁孟嘉 教授 (國立臺灣大學土木工程學系)
Reliable assessment of earth-fill dam deformation under reservoir operation remains challenging due to limited monitoring coverage and uncertainties in numerical modeling. This study integrates high-resolution time-series interferometric synthetic aperture radar (InSAR) observations with seepage–mechanical modeling to investigate reservoir-induced deformation of the Renyitan earth-fill dam in southwestern Taiwan. Using 34 TerraSAR-X High-Resolution Spotlight images spanning an annual operation cycle, millimeter-scale vertical deformation was characterized and validated against in-situ leveling measurements. The InSAR results reveal seasonal deformation patterns associated with reservoir water-level fluctuations, including uplift during drawdown and settlement during impoundment, across the dam body and adjacent western ridge. Two-dimensional finite-difference models were developed for five representative dam cross sections to investigate the underlying mechanisms. Conventional simulations assuming steady-state seepage at the full reservoir level followed by rapid drawdown failed to reproduce the observed deformation trends. In contrast, simulations initialized with steady-state seepage at a low reservoir level and subjected to time-dependent water-level fluctuations successfully reproduced the observed deformation patterns. Calibration of foundation soil stiffness using InSAR-derived constraints further improved agreement in deformation magnitude. These findings highlight the critical roles of initial hydraulic conditions and foundation stiffness in reservoir-induced deformation and demonstrate the potential of InSAR-constrained modeling for earth-fill dam safety assessment.
Keywords: Earth-fill dam, Time-series InSAR, Dam deformation
From Simulated Sea Surfaces to Software-Defined Radar Hardware: Building a C-Band Simulation-to-Measurement Chain for Ocean Wind Sensing
講者：古瓊昇 助理教授 (國立臺北科技大學太空系統工程研究所)
Sea-surface wind retrieval from C-band SAR still relies on empirical geophysical model functions fitted to foreign satellite archives. We are building a chain that links sea-surface physics, radar signals and calibrated measurement, and report two of its links. The first is OceanSARSim, which synthesizes sea surfaces from directional wave spectra (Elfouhaily, Pierson–Moskowitz and JONSWAP) under controlled 10-m wind speed (U10), relative wind direction and incidence angle, and generates the corresponding C-band SAR imagery. A 50,000-sample Elfouhaily dataset is used for training, and a paired 5,000-sample JONSWAP set with identical wind and incidence conditions isolates the effect of the assumed spectrum. Four estimators of increasing physical structure (MLP, ResNet-18, and circular and physics-constrained mixture density networks) are used to examine how well U10 and relative wind direction can be identified from a single image.
The second link is the radar front end. On a USRP X310, a custom RFNoC dechirp block performs dechirping and 100× decimation in the FPGA; at 5.3 GHz with 160 MHz bandwidth, targets at 2.5 m and 3.0 m were located within +2.6 cm and +6.9 cm after reference calibration, with millimeter-level repeatability. A ZCU208 RFSoC front end with 300 MHz bandwidth shows that calibrating on a free-space reference target, rather than a cable loopback, is needed to remove a 0.5–0.75 m bias caused by antenna and RF path delays. These results lay the groundwork for an indigenous C-band ocean-sensing SAR capability.
Keywords: C-band SAR; sea-surface wind retrieval; ocean wave spectra; deep learning; FMCW radar
次米級X波段SAR與高解析光學影像融合技術驅動精準作物分類
講者：林玉儂 副研究員 (中央研究院地球科學研究所)
本研究以台灣雲林土庫碎化農地為試驗區，探討商業次米級合成孔徑雷達（SAR）影像應用於作物分類的幾何校正與特徵提取架構。商業SAR星系（如 Umbra）具備極高的時空解析度優勢（每1-2日一張、0.5公尺），但每次拍攝的入射角、斜視角與軌道方向皆不同，與傳統重複軌道、重複姿態SAR系統（如PAZ、TSX等）所拍攝的影像相較，此多幾何拍攝模式除了有顯著的幾何錯位之外，對於同一地面標的也會產生與拍攝幾何有關的輻射差異。針對幾何錯位問題，本研究提出兩步驟影像校準流程：先以角反射器與地面控制點完成基準影像手動幾何校正，再利用 GeFolki 像素追蹤演算法自動對齊其餘影像。經 58 張校準影像評估，此方法可將角反射器中位定位誤差由 10.5 公尺大幅降至 2.4 公尺。分析顯示誤差大小主要取決於入射角（角度愈陡/小，幾何畸變愈明顯），與斜視角及視線方向無顯著相關。在輻射分析上，研究將後向散射係數映射至「入射角—生長期」特徵空間，發現相較於固定 20° 入射角的 PAZ/TSX 資料，Umbra 的多視角動態軌跡對作物結構與介電特性更具敏感度，在 4 至 5 月關鍵生長期更能區分固定視角下難以鑑別的作物（如玉米與花生）。在此基礎之下，本研究針對 7 類作物建立隨機森林基準模型，結果顯示具多角度特徵的 Umbra 資料集分類表現（Macro-F1 0.56、Kappa 0.77）顯著優於單一幾何的 PAZ+TSX（Macro-F1 0.49、Kappa 0.55）；若進一步融合無人機影像，Macro-F1 與 Kappa 可更提升至 0.62 與 0.81。研究證實多幾何次米級 SAR 影像可成功轉化為具高辨識力的特徵，為破碎化農地作物製圖奠定關鍵基礎。
關鍵字：多幾何SAR影像校準、入射角—生長期特徵空間、破碎化農地作物分類
利用多時序與多感測衛星遙測影像融合提升都會區與基礎設施監測能力
講者：莊昀叡 教授 (國立臺灣大學土木工程學系)
本計畫以多尺度合成孔徑雷達（SAR）技術為核心，整合區域地表變形監測、局部高精度相位解纏方法，以及次米級SAR與光學影像融合應用，建立由區域尺度至關鍵基礎設施尺度之監測技術架構。研究內容包含三個子計畫：(1)多時序InSAR三維變形場解算與區域尺度地表變形分析；(2)角反射器輔助之相位解纏與時間序列精度提升方法；(3)次米級SAR與光學影像融合之輸電設施判釋與監測流程建構。本計畫已完成多時期 SAR 影像資料蒐集與前處理，建立研究區域之視線方向變形速度場，並進行升軌降軌資料整合測試。角反射器輔助之pixel-arc分析流程已建構完成，初步驗證其於低相干背景下提升時間序列穩定度之效果。同時進行高解析SAR與光學影像之幾何校正與設施判釋流程，完成平地與山區的實測，分析地形與觀測幾何對輸電塔影像呈現之影響。三項技術模組已逐步建立基礎，後續將進一步強化三維變形解算精度與多源資料整合應用能力，以提升關鍵基礎設施監測之可靠性與實用性。
關鍵字：合成孔徑雷達、三維形變反演、InSAR變形監測、角反射器、干涉網路穩定化、假色次孔徑影像（CSI）
基於星上邊緣運算架構之輕量化衛星資料分析技術建立與應用
講者：蔡慧萍 副教授 (國立中興大學土木工程學系)
This two-year project integrates optical remote sensing for reservoir water quality monitoring, radio frequency interference (RFI) detection in synthetic aperture radar (SAR) imagery, and lightweight artificial intelligence (AI) for onboard edge computing in the FORMOSAT satellite series. First-year work addresses water body extraction, water quality estimation, RFI dataset construction, and post-training quantization (PTQ). Sentinel-2 imagery from 2019–2026 and available FORMOSAT-8 imagery were collected for Tsengwen, Mingde, and Fengshan reservoirs. Five machine learning supervised classifiers were evaluated using reference labels derived from high-resolution FORMOSAT-8 imagery, achieving a Kappa coefficient of 0.861 and an F1-score of 0.896. The Normalized Difference Chlorophyll Index (NDCI) was used to assess eutrophication. Limited temporal correspondence between field measurements and satellite acquisitions constrained regression performance (R² = 0.515), although the approach supported relative water quality classification at Fengshan Reservoir. For SAR image quality control, RFI-affected Sentinel-1 scenes acquired since 2014 were selected to construct a dataset supporting deep learning-based detection. This work addresses limitations of conventional frequency-domain notch filtering, including manual tuning, overfiltering, and secondary sidelobe artifacts. For onboard AI, a workflow incorporating image preprocessing, annotation, hardware-constrained model optimization, floating-point accuracy assessment, and PTQ was established. Quantization calibration used the Neural Network Deployment and Compilation Tool (NNDCT), with model evaluation showing user’s and producer’s accuracies above 80%. Future work will automate water delineation and spectral feature extraction, diversify datasets, and evaluate lightweight deep learning models for near-real-time reservoir monitoring, SAR image quality control, and onboard processing.
Magnetic Storm Effects Observed by FORMOSAT-7/COSMIC-2
講者：林建宏 教授 (國立成功大學地球科學系)
In this study, we present day-to-day variations of the ionosphere driven by the vertical coupling of the atmospheric waves and tides and the magnetic storms using the 3-D electron density outputs from the global ionosphere specification (GIS) built based on COSMIC-2 radio occultation observations and ground-based slant TECs. Previous studies show that ionosphere have the variations being influenced by the upward propagating waves/tides based on averaging several days of satellite observations due to limitation in global coverage of the observations. It might over-estimate or under-estimate the effects as the variations also have obvious longitudinal and local time effects. Average of observations for tidal fitting might amplify/reduce the effect if the sampling is unbalanced. With the global 3-D electron density output in time resolution of an hour, we can re-evaluate the tidal effects due to various lower atmospheric forcing by performing the daily tidal fitting. Additionally, we present the low- and mid-latitude ionosphere coupling during magnetic storm time. One important feature is the super plasma fountain effect that drives equatorial ionization anomaly to mid-latitude and further extend to high latitude through sub-auroral polarization stream (SAPS) electric field and featured by tongue of ionization (TOI) in the ionospheric TEC. COSMIC GIS can provide the vertical structure of the TOI for further study of its interconnection to the high latitude convections. We present case studies to reveal how TOI connects to the convection region using COSMIC GIS.
自主式太空天氣預報系統研發
講者：陳佳宏 副教授 (國立成功大學地球科學系)
Severe space weather events significantly disrupt critical infrastructures, but traditional forecasting offers limited warning and relies on empirical models that often fail during major storms. To address these limitations, we developed an autonomous forecasting system utilizing a multimodal late-fusion deep learning architecture that integrates ACE L1 solar wind data, SDO/AIA 193Å solar coronal imagery, and temporal features. Our model predicts continuous polar convection potential maps by generating 81 spherical harmonic coefficients across eight lead times ranging from +1 to +48 hours.
Preliminary results from our optimized framework indicate that while temporal priors adequately capture routine climatological structures, dynamic multimodal inputs significantly enhance forecasting accuracy during volatile space weather states. Specifically, during periods of high cross-polar cap potential (CPCP≥38.94kV), the full multimodal configuration achieved a CPCP root mean square error of 19.396 kV, substantially outperforming the time-only baseline (21.802 kV). Furthermore, the integration of solar wind streams provides a distinct advantage for short-term (+1hr to +6hr) forecasts. This framework extends warning horizons and delivers dynamically consistent electrodynamic drivers for global space weather resilience.
Keywords: Space weather forecasting, Multimodal deep learning, Electric potential
透過共同整合多元衛星資料促進福衛產品在大氣對流過程研究之應用
講者：陳維婷 教授 (國立臺灣大學大氣科學系暨研究所)
This three-year integrated project has progressed smoothly through its first year. The project develops a unified framework for spatiotemporal integration and collocation of multi-satellite observations to support convection process studies and model evaluation, explicitly incorporating FORMOSAT-7 GNSS radio occultation thermodynamic profiles and TRITON GNSS-reflectometry ocean surface winds together with Himawari, CloudSat, MODIS, GPM-IMERG, and ERA5. The work comprises four coupled components: a main project that develops the AEL-satellite-tools software package and collocation strategies, and three sub-projects that apply the integrated datasets to (i) evaluating East Asian winter cold surges in climate models, (ii) diagnosing organized oceanic convection over the Northwest Pacific using object-based identification/tracking and radiative–convective equilibrium (RCE) comparisons, and (iii) assessing low clouds over Taiwan’s complex terrain through satellite–ground validation and localized cloud detection.
Major first-year milestones have been achieved across all components. (1) The main project completed the core architecture and workflow design of AEL-satellite-tools, establishing standardized preprocessing and quick-look quality control to support multiple satellite products while preserving native data values. (2) Sub-project 1 compiled multi-satellite observations, ERA5, and climate model outputs; completed cold-surge event identification; and produced initial composites that reveal systematic contrasts between SPCAM and CAM in precipitation responses, low-cloud distributions, and liquid water path. (3) Sub-project 2 generated pilot ERA5-grid collocated datasets (column water vapor, Himawari infrared brightness temperature, IMERG precipitation, and sea surface temperature), implemented moist-cluster detection using CWV thresholds, and developed distance-to-edge diagnostics and lifetime statistics for tracked convective systems. In parallel, non-rotating VVM RCE simulations were completed and analyzed with the same object-based framework to diagnose aggregation-related structures and circulation signatures. (4) Sub-project 3 diagnosed Taiwan-specific biases in MODIS single-IR cloud identification, analyzed terrain-modulated cloud regimes in TaiwanVVM winter monsoon cases to guide observing strategies, and established a three-site all-sky camera network with preliminary cloud-fraction retrievals for satellite cross-validation.
To date, the project has produced one peer-reviewed journal publication and three oral presentations at major international conferences. Ongoing work will strengthen routine dataset production and metadata/QC standards, expand supported products and user training, deepen structure-based cold-surge diagnostics, refine object definitions and extend to rotating RCE comparisons, and complete Taiwan-specific cloud-threshold optimization with coordinated satellite–ground validation during intensive observational campaigns.
獵風者衛星GNSS資料加值與應用
講者：楊舒芝 特聘教授 (國立中央大學大氣科學學系)
本計畫期利獵風者衛星GNSS-R觀測，發展訊號處理反演及資料加值之創新應用，並結合如機器學習等進階統計資料科學方法，更細緻的物理法突破前期計畫所遭遇的困難及挑戰。基於獵風者可獲得較高解析度DDM特性，透過了解DDM在颱風高風速下之形態特徵變化以建立可辨識之關聯。並針對海表面白帽特性建立其對GNSS-R風速反演之影響，以優化資料反演及QC流程。除此之外，透過已建立的高解析度颱風風速模擬改進傳統GMF於高風速估算時受限於無觀測資料之困境，並初步證實高解析度的DDM對於高風速特性更為敏感。新版的獵風者風速在20m/s以下準確度有12%之提升，但高風速低估的情況仍顯著！所建立之校正過的ASCAT衛星海表面風速可適用40m/s之高風速範圍，並可作為後續重要的風速參考值或校正模型依據。
本計畫透過優化GNSS-R觀測於高解析度同化系統之使用以改善台灣劇烈天氣預報。首先在本計劃發展的近即時多源衛星融合建立大西洋颶風海表面風速剖線估計模式進一步在演算架構中加入不同資料之觀測誤差考量已優化衛星資料的使用，並於近期完成此模型於太平洋颱風風速剖線及大小估計之應用。針對颱風預報，研究成果顯示結合GNSS掩星觀測及獵風者風速資料對於颱風生成預報有相當正面的影響，透過有效改進颱風路徑誤差改善颱風發展環境，進而對颱風強度亦有改善。為避免GNSS-R高風速低估影響颱風強度預報，本計劃利用新發展的機器學習觀測算符同化DDMA觀測，初步結果顯示同化DDMA可有效調整颱風核區風速，值得持續發展。此外，針對極短期降雨預報，雖然目前獵風者資料量受限，但結合雷達及地面觀測等仍可彌補台灣地基觀測盲區，進一步改進對流尺度降雨預報。
在水文應用方面，我們已建立機器學習模型整合 GNSS-R 衛星資料與現地觀測及氣候條件與地質條件，評估其於乾旱環境中土壤含水量反演的適用性。結果顯示在沙漠與乾旱區域中，多源資料整合為提升土壤含水量反演精度之有效策略。另外，透過發展掩星可降水量產品提升GNSS觀測氣象應用，與海面風場結合可改善台灣近岸低層大氣水氣輻合，並作為其他GNSS氣象資訊模型建立之重要特徵資料。TRITON DDM資料亦應用於探討海表暖水區對流與強風的物理效應，與對流層強對流系統與電離層擾動的耦合。
關鍵字：GNSS-R、海表面風速、土壤溼度
利用福衛星系加值資料改善低對流層天氣分析與預報
講者：陳舒雅 研究員 (國立中央大學全球大氣觀測與資料應用研究中心)
全球導航衛星系統無線電掩星(GNSS-RO)技術已成為現代數值天氣預報之關鍵觀測來源。福衛七號觀測剖線雖具備深入低對流層之探測能力，然受限於大氣超折射、波導效應及訊號追蹤雜訊等影響，導致資料於對流層低層常伴隨顯著負偏差；此現象易使掩星觀測於資料同化之品質控管(QC)階段遭到剔除，因而限制了掩星資料在低層大氣監測與劇烈天氣預報上的應用潛力。本整合型計畫以修正低層偏差、提升低對流層RO資料可用性及其同化效益為核心，建立「低層偏差修正、動態誤差優化、輔助觀測驗證、作業系統評估」之整合架構。計畫由四個子計畫協同執行：子計畫一利用機器學習方式建立偏折角偏差修正模型，優化邊界層頂判定與溫濕度反演品質；子計畫二將導入局地波譜寬(LSW)資訊，於MPAS-GSI系統中建構具環境相依性之動態觀測誤差參數化方法，並配合系集敏感度分析，提升颱風等劇烈天氣個案之同化分析與預報診斷能力；子計畫三透過TROPS系統開發地基GNSS天頂對流層延遲(ZTD)反演技術，提供獨立水氣資訊以支援模型之訓練與驗證；子計畫四則將修正技術導入中央氣象署(CWA)全球作業同化系統，探討其對低層掩星資料同化利用效率之影響，以及評估其對模式預報的改進效益。
關鍵字：福衛七號、低層偏差、動態觀測誤差、天頂對流層延遲、資料同化
溫室氣體衛星遙測資料之地面驗證與排放監測應用
講者：陳怡均 副研究員 (中央研究院環境變遷研究中心)
Satellite observations provide broad spatial coverage of atmospheric CO₂, while ground-based measurements are essential for validating satellite retrievals and supporting regional emission estimation. This study aims to establish an integrated framework linking ground-based observations, satellite validation, and emission estimation to support the reliable application of CO₂ observations for regional emission monitoring in Taiwan.
High-precision measurements of column-averaged dry-air mole fractions of CO₂ (XCO₂) have been conducted using ground-based EM27/SUN Fourier transform spectrometers. These measurements are being compared with Orbiting Carbon Observatory-3 (OCO-3) observations to assess satellite retrieval accuracy. Building on the ground-based and satellite measurements, this study is developing a top-down approach to estimating regional CO₂ emissions using atmospheric transport modeling and the CarbonTracker framework. The resulting estimates will be compared with conventional bottom-up emission inventories to identify potential discrepancies.
This integrated approach is expected to strengthen CO₂ monitoring and provide independent observational constraints on regional emissions, contributing to more reliable emission assessments in Taiwan.
（議程與講題如有異動，以現場公告為準。）