People

Dr. Pao-Kuan Wang, Academicians of Academia Sinica-When Run Out of Ideas, Buy Equipment. [ 下載 PDF ]

Claire Lin

Digitalization of Environmental Information

Special Issue Introduction of “Digitalization of Environmental Information” [ 下載 PDF ]

Charles C.-K. Chou

Deep Learning: Applications in the Digitization of Environmental Data [ 下載 PDF ]

Pu-Yun Kow, Yun-Ting Wang, Jun-Jie Lin, Jia-Yi Liou, Yu-Wen Chang, Fi-John Chang

Deep learning accurately captures nonlinear features and sharp changes in data, enabling highprecision predictions. This study applies it to several environmental issues, including air pollution forecasting, sewer water level forecasting in Taipei, and greenhouse microclimate forecasting. For air pollution, historical data from the Environmental Protection Agency station from whole Taiwan, including six pollution and two meteorological factors, were used. The model's attention mechanism effectively solved the vanishing gradient problem, significantly improving 72-hour forecast accuracy, with RMSE errors ranging from 8.5 to 13.2 μg/m3. In sewer water level forecasting for Taipei, a DNN-AE model was employed, providing stable forecasts for 10 to 60 minutes into the future. Among the models, the C-AE model demonstrated structural advantages, as its convolutional layers efficiently extracted temporal features, especially when handling timevarying data. It improved forecasting accuracy, with RMSE errors ranging from 0.21 to 0.51 meters from T+1 to T+6, yielding the most accurate and stable results. For greenhouse microclimate forecasting in Shengang, Changhua, XGBoost was used to select features like TSF for temperature, SWI for shortwave radiation and RH for relative humidity, reflecting the effects of rolling curtains and shading systems. The ANFIS model performed best in temperature forecasting, with an R2 value exceeding 0.8, while CNN excelled in forecasting humidity and light intensity. By integrating these applications, Taiwan’s environmental forecasting technology will further drive the development of digital cities, moving towards a smarter and more advanced future.

Atmospheric Boundary Layer Observation, Modeling and Data Assimilation [ 下載 PDF ]

Fang-Yi Cheng, Sheng-Hsiang Wang, Shu-Chih Yang

The atmospheric boundary layer, located between the Earth's surface and the free atmosphere, is a region influenced by surface forces, with an average thickness of approximately one kilometer. Meteorological conditions within the boundary layer play a critical role in the evolution of shortterm weather systems, long-term climate changes, and the transport and dispersion of air pollutants. Modern observation technologies, such as radiosondes, drones, and micro-pulse lidars, provide detailed data on the vertical structure and meteorological conditions of the boundary layer. In addition to observations, numerical simulations are essential for analyzing the evolution of the boundary layer. Current boundary layer meteorological simulations primarily utilize boundary layer parameterization schemes or large-eddy simulations. Furthermore, data assimilation techniques that integrate boundary layer observational data with meteorological simulations can significantly enhance the accuracy of weather forecasts. This study systematically introduces advancements in observation technologies, numerical simulations, and data assimilation in the study of the atmospheric boundary layer, aiming to deepen our understanding of this region and improve its forecasting capabilities.

Deep-QPF: Deep Learning for Radar-based Rainfall Nowcasting in Taiwan [ 下載 PDF ]

Buo-Fu Chen

Using deep learning to analyze radar and other observational data can enhance realtime (< 3 h) quantitative precipitation forecasting (QPF). Incorporating artificial intelligence (AI) into operational weather forecasting, aligns with the trends of professional specialization, refined forecasting, and real-time operations, thereby strengthening meteorological forecasting capabilities. The Deep-QPF deep learning model predicts rainfall for the next three hours. It effectively integrates heterogeneous data with various characteristics through different approaches, significantly improving rainfall forecasting, particularly for 0-3-hour precipitation forecasts. This paper also analyzes Deep-QPF’s forecasting performance and systematic error, demonstrating that this AI-based forecasting technology suits practical forecasting applications. It can enhance Taiwan's disaster response capabilities against short-duration heavy rainfall events.

Digitization and Visualization of Environmental Information on the Civil IoT Taiwan Data Service Platform [ 下載 PDF ]

Hui-Hung Yu, Wei-Yu Chen, Chen-Kai Sun

The National Center for High-Performance Computing (NCHC) has developed the Civil IoT Taiwan Data Service Platform, which collects and stores sensor data in four key areas: earthquake, water resource, air quality, and disaster prevention. This platform also creates services for the digitization of environmental information to meet users' needs for visual data. The Civil IoT Taiwan Data Service Platform aims to assist both the government and citizens in preparing for future environmental challenges through the digitization of environmental information.　　

Micro Sensor Readout Circuits and Systems [ 下載 PDF ]

, , , ,

This article introduces micro gas sensor readout circuits designed for semiconductor resistive gas sensing materials, including metal oxides, two-dimensional materials, and nanomaterials. Known for their high sensitivity, fast response, and selectivity, these materials enhance detection accuracy and stability by matching the system’s response to different material properties, accurately measuring and correcting resistance changes.

Hyperspectral Sensing in Aagriculture and Water Areas [ 下載 PDF ]

Long-Jeng Lee, Yuan Shen

Since hyperspectral images are in line with the spirit of Industry 4.0, hyperspectral images themselves are big information of data. Researchers can rely on the analysis of these spectral image data to achieve accurate measurement, prediction, and even implementation management. In this article, we take the application of hyperspectral imaging in precision agriculture and water environment observation as examples to illustrate the application of hyperspectral in the ecological environment.

Column

Research on Lithium Battery Explosion- proof Early Warning Detection [ 下載 PDF ]

Qian-Yu Yi, Sheng-Chang Wang, Yu-Jen Hsiao

As the demand for lithium batteries in the global market grows, safety concerns have become a focal point, especially in applications such as electric vehicles and energy storage systems. To address the issue of thermal runaway caused by overcharging, short circuits, and other factors, this paper explores various safety monitoring technologies, including temperature and gas sensors, to monitor changes inside and around the battery, thereby reducing operational risks, extending battery lifespan, and enhancing performance. The study indicates that multi-sensor technology offers high sensitivity and reliability, making it suitable for applications in electric vehicles and energy storage systems, thus providing strong technical support and safety assurance for the sustainable development of the lithium battery market.

Popular Science

Conductor of the Light - Explore the Secrets of Optical Thin Film [ 下載 PDF ]

Po-Li Chen, Tzu-Yu Chuang, Hung-Pin Chen