Advancing climate-resilient flood mitigation: Utilizing transformer-LSTM for water level forecasting at pumping stations.

Proactive management of pumping stations using artificial intelligence (AI) technology is vital for effectively mitigating the impacts of flood events caused by climate change. Accurate water level forecasts are pivotal in advancing the intelligent operation of pumping stations. This study proposed a novel Transformer-LSTM model to offer accurate multi-step-ahead forecasts of the flood storage pond (FSP) and river water levels for the Zhongshan pumping station in Taipei, Taiwan. A total of 19,647 ten-minute-based datasets of pumping operation and storm sewer, FSP, and river water levels were collected between 2014 and 2020 and further divided into training (70 %), validation (10 %), and test (20 %) datasets for model construction. The results demonstrate that the proposed model dramatically outperforms benchmark models by producing more accurate and reliable water level forecasts at 10-minute (T + 1) to 60-minute (T + 6) horizons. The proposed model effectively enhances the connections between input factors through the Transformer module and increases the connectivity across consecutive time series using the LSTM module. This study reveals interconnected dynamics among pumping operation and storm sewer, FSP, and river water levels, enhancing flood management. Understanding these dynamics is crucial for effective execution of management strategies and infrastructure revitalization against climate impacts. The Transformer-LSTM model's forecasts encourage water practices, resilience, and disaster risk reduction for extreme weather events.

Watershed groundwater level multistep ahead forecasts by fusing convolutional-based autoencoder and LSTM models.

The development of deep learning-based groundwater level forecast models can tackle the challenge of high dimensional groundwater dynamics, predict groundwater variation trends accurately, and manage groundwater resources effectively, thereby contributing to sustainable water resources management. This study proposed a novel ConvAE-LSTM model, which fused a Convolutional-based Autoencoder model (ConvAE) and a Long Short-Term Memory Neural Network model (LSTM), to provide accurate spatiotemporal groundwater level forecasts over the next three months. The HBV-light and LSTM models are chosen as benchmarks. An ensemble of point data and the corresponding derived images concerning the past (observations) and the future (forecasts from a conceptual model) of groundwater levels at 33 groundwater wells in Jhuoshuei River basin of Taiwan between 2000 and 2019 constituted the case study. The findings showcase the effectiveness of the ConvAE-LSTM model in extracting crucial features from both point and imagery datasets. This model successfully establishes spatiotemporal dependencies between regional images and groundwater level data over diverse time frames, leading to accurate multi-step-ahead forecasts of groundwater levels. Notably, the ConvAE-LSTM model exhibits a substantial improvement, with the R-squared values showing an increase of more than 18%, 22%, and 49% for the R1, R2, and R3 regions, respectively, compared to the HBV-light model. Additionally, it outperforms the LSTM model in this regard. This study represents a noteworthy milestone in environmental modeling, offering key insights for designing sustainable groundwater management strategies to ensure the long-term availability of this vital resource.

Deep neural networks for spatiotemporal PM2.5 forecasts based on atmospheric chemical transport model output and monitoring data.

Reliable long-horizon PM2.5 forecasts are crucial and beneficial for health protection through early warning against air pollution. However, the dynamic nature of air quality makes PM2.5 forecasts at long horizons very challenging. This study proposed a novel machine learning-based model (MCNN-BP) that fused multiple convolutional neural networks (MCNN) with a back-propagation neural network (BPNN) for making spatiotemporal PM2.5 forecasts for the next 72 h at 74 stations covering the whole Taiwan simultaneously. Model configuration involved an ensemble of massive hourly air quality and meteorological monitoring datasets and the existing publicly-available PM2.5 simulated (forecasted) datasets from an atmospheric chemical transport (ACT) model. The proposed methodology collaboratively constructed two CNNs to mine the observed data (the past) and the forecasted data from ACT (the future) separately. The results showed that the MCNN-BP model could significantly improve the accuracy of spatiotemporal PM2.5 forecasts and substantially reduce the forecast biases of the ACT model. We demonstrated that the proposed MCNN-BP model with effective feature extraction and good denoising ability could overcome the curse of dimensionality and offer satisfactory regional long-horizon PM2.5 forecasts. Moreover, the MCNN-BP model has considerably shorter computational time (5 min) and lower computational load than the compute-intensive ACT model. The proposed approach hits a milestone in multi-site and multi-horizon forecasting, which significantly contributes to early warning against regional air pollution.

Real-time image-based air quality estimation by deep learning neural networks.

Air quality profoundly impacts public health and environmental equity. Efficient and inexpensive air quality monitoring instruments could be greatly beneficial for human health and air pollution control. This study proposes an image-based deep learning model (CNN-RC) that integrates a convolutional neural network (CNN) and a regression classifier (RC) to estimate air quality at areas of interest through feature extraction from photos and feature classification into air quality levels. The models were trained and tested on datasets with different combinations of the current image, the baseline image, and HSV (hue, saturation, value) statistics for increasing model reliability and estimation accuracy. A total of 3549 hourly air quality datasets (including photos, PM2.5, PM10, and the air quality index (AQI)) collected at the Linyuan air quality monitoring station of Kaohsiung City in Taiwan constituted the case study. The main breakthrough of this study is to timely produce an accurate image-based estimation of several pollutants simultaneously by using only one single deep learning model. The test results show that estimation accuracy in terms of R2 for PM2.5, PM10, and AQI based on daytime (nighttime) images reaches 76% (83%), 84% (84%), and 76% (74%), respectively, which demonstrates the great capability of our method. The proposed model offers a promising solution for rapid and reliable multi-pollutant estimation and classification based solely on captured images. This readily scalable measurement approach could address major gaps between air quality data acquired from expensive instruments worldwide.

Explore Regional PM2.5 Features and Compositions Causing Health Effects in Taiwan.

Chemical compositions of atmospheric fine particles like PM2.5 prove harmful to human health, particularly to cardiopulmonary functions. Multifaceted health effects of PM2.5 have raised broader, stronger concerns in recent years, calling for comprehensive environmental health-risk assessments to offer new insights into air-pollution control. However, there have been few studies adopting local air-quality-monitoring datasets or local coefficients related to PM2.5 health-risk assessment. This study aims to assess health effects caused by PM2.5 concentrations and metal toxicity using epidemiological and toxicological methods based on long-term (2007-2017) hourly monitoring datasets of PM2.5 concentrations in four cities of Taiwan. The results indicated that (1) PM2.5 concentrations and hazardous substances varied substantially from region to region, (2) PM2.5 concentrations significantly decreased after 2013, which benefited mainly from two actions against air pollution, i.e., implementing air-pollution-control strategies and raising air-quality standards for certain emission sources, and (3) under the condition of low PM2.5 concentrations, high health risks occurred in eastern Taiwan on account of toxic substances adsorbed on PM2.5 surface. It appears that under the condition of low PM2.5 concentrations, the results of epidemiological and toxicological health-risk assessments may not agree with each other. This raises a warning that air-pollution control needs to consider toxic substances adsorbed in PM2.5 and region-oriented control strategies are desirable. We hope that our findings and the proposed transferable methodology can call on domestic and foreign authorities to review current air-pollution-control policies with an outlook on the toxicity of PM2.5.

Explore spatio-temporal PM2.5 features in northern Taiwan using machine learning techniques.

The complex mixtures of local emission sources and regional transportations of air pollutants make accurate PM2.5 prediction a very challenging yet crucial task, especially under high pollution conditions. A symbolic representation of spatio-temporal PM2.5 features is the key to effective air pollution regulatory plans that notify the public to take necessary precautions against air pollution. The self-organizing map (SOM) can cluster high-dimensional datasets to form a meaningful topological map. This study implements the SOM to effectively extract and clearly distinguish the spatio-temporal features of long-term regional PM2.5 concentrations in a visible two-dimensional topological map. The spatial distribution of the configured topological map spans the long-term datasets of 25 monitoring stations in northern Taiwan using the Kriging method, and the temporal behavior of PM2.5 concentrations at various time scales (i.e., yearly, seasonal, and hourly) are explored in detail. Finally, we establish a machine learning model to predict PM2.5 concentrations for high pollution events. The analytical results indicate that: (1) high population density and heavy traffic load correspond to high PM2.5 concentrations; (2) the change of seasons brings obvious effects on PM2.5 concentration variation; and (3) the key input variables of the prediction model identified by the Gamma Test can improve model's reliability and accuracy for multi-step-ahead PM2.5 prediction. The results demonstrated that machine learning techniques can skillfully summarize and visibly present the clusted spatio-temporal PM2.5 features as well as improve air quality prediction accuracy.

Self-organizing maps of typhoon tracks allow for flood forecasts up to two days in advance.

Typhoons are among the greatest natural hazards along East Asian coasts. Typhoon-related precipitation can produce flooding that is often only predictable a few hours in advance. Here, we present a machine-learning method comparing projected typhoon tracks with past trajectories, then using the information to predict flood hydrographs for a watershed on Taiwan. The hydrographs provide early warning of possible flooding prior to typhoon landfall, and then real-time updates of expected flooding along the typhoon's path. The method associates different types of typhoon tracks with landscape topography and runoff data to estimate the water inflow into a reservoir, allowing prediction of flood hydrographs up to two days in advance with continual updates. Modelling involves identifying typhoon track vectors, clustering vectors using a self-organizing map, extracting flow characteristic curves, and predicting flood hydrographs. This machine learning approach can significantly improve existing flood warning systems and provide early warnings to reservoir management.

Explore a Multivariate Bayesian Uncertainty Processor driven by artificial neural networks for probabilistic PM2.5 forecasting.

Quantifying predictive uncertainty inherent in the nonlinear multivariate dependence structure of multi-step-ahead PM2.5 forecasts is challenging. This study integrates a Multivariate Bayesian Uncertainty Processor (MBUP) and an artificial neural network (ANN) to make accurate probabilistic PM2.5 forecasts. The contributions of the proposed approach are two-fold. First, the MBUP can capture the nonlinear multivariate dependence structure between observed and forecasted data. Second, the MBUP can alleviate predictive uncertainty encountered in PM2.5 forecast models that are configured by ANNs. The reliability of the proposed approach was assessed by a case study on air quality in Taipei City of Taiwan. We consider forecasts of PM2.5 concentrations as a function of meteorological and air quality factors based on long-term (2010-2018) hourly observational datasets. Firstly, the Back Propagation Neural Network (BPNN) and the Adaptive Neural Fuzzy Inference System (ANFIS) were investigated to produce deterministic forecasts. Results revealed that the ANFIS model could learn different air pollutant emission mechanisms (i.e. primary, secondary and natural processes) from the clustering-based fuzzy inference system and produce more accurate deterministic forecasts than the BPNN. The ANFIS model then provided inputs (i.e. point estimates) to probabilistic forecast models. Next, two post-processing techniques (MBUP and the Univariate Bayesian Uncertainty Processor (UBUP)) were separately employed to produce probabilistic forecasts. The Bayesian Uncertainty Processors (BUPs) can model the dependence structure (i.e. posterior density function) between observed and forecasted data using a prior density function and a likelihood density function. Here in BUPs, the Monte Carlo simulation was introduced to create a probabilistic predictive interval of PM2.5 concentrations. The results demonstrated that the MBUP not only outperformed the UBUP but also suitably characterized the complex nonlinear multivariate dependence structure between observations and forecasts. Consequently, the proposed approach could reduce predictive uncertainty while significantly improving model reliability and PM2.5 forecast accuracy for future horizons.

Multi-output support vector machine for regional multi-step-ahead PM2.5 forecasting.

Air quality deteriorates fast under urbanization in recent decades. Reliable and precise regional multi-step-ahead PM2.5 forecasts are crucial and beneficial for mitigating health risks. This work explores a novel framework (MM-SVM) that combines the Multi-output Support Vector Machine (M-SVM) and the Multi-Task Learning (MTL) algorithm for effectively increasing the accuracy of regional multi-step-ahead forecasts through tackling error accumulation and propagation that is commonly encountered in regional forecasting. The Single-output SVM (S-SVM) is implemented as a benchmark. Taipei City of Taiwan is our study area, where three types of air quality monitoring stations are selected to represent areas imposed with high traffic influences, high human activities and commercial trading influences, and less human interventions close to nature situation, respectively. We consider forecasts of PM2.5 concentrations as a function of meteorological and air quality factors based on long-term (2010-2016) observational datasets. Firstly, the Kendall tau coefficient is conducted to extract key spatiotemporal factors from regional meteorological and air quality inputs. Secondly, the M-SVM model is trained by the MTL to capture non-linear relationships and share correlation information across related tasks. Lastly, the MM-SVM model is validated using hourly time series of PM2.5 concentrations as well as meteorological and air quality datasets. Regarding the applicability of regional multi-step-ahead forecasts, the results demonstrate that the MM-SVM model is much more promising than the S-SVM model because only one forecast model (MM-SVM) is required, instead of constructing a site-specific S-SVM model for each station. Moreover, the forecasts of the MM-SVM are found better consistent with observations than those of any single S-SVM in both training and testing stages. Consequently, the results clearly demonstrate that the MM-SVM model could be recommended as a novel integrative technique for improving the spatiotemporal stability and accuracy of regional multi-step-ahead PM2.5 forecasts.

Conservation of groundwater from over-exploitation-Scientific analyses for groundwater resources management.

Groundwater over-exploitation has produced many critical problems in the southern Taiwan. The accumulated stresses and demands make groundwater management a complex issue that needs innovative scientific analyses for deriving better water management strategies. In this study, we aimed to provide scientific analyses of the groundwater systems in the Pingtung Plain through soft-computing techniques to explore its spatial-temporal and hydro-geological characteristics for the elaboration of future groundwater management plans and in decision-making process. We conducted a study to assess the essential features of the groundwater systems based on the long-term large datasets of regional groundwater levels by using the principal component analysis (PCA), and the self-organizing map (SOM) with regression analysis. The PCA results demonstrated that two leading components could well present the spatial characteristics of the groundwater systems and classify the region into eastern, western and transition zones. The SOM results could visibly explore the behavior of regional groundwater variations in various aquifers and the multi-relations among climate and hydrogeological variables. Results revealed that the potential of groundwater recharge made by precipitation or river flow was higher in the eastern zone than in the western zone. Analysis results further showed an increase of the groundwater levels in the western zone after year 2006, while there were no obvious increases of the groundwater levels in the eastern or transition zones. Based on the investigated characteristics, we suggest that a sound groundwater management plan should consider zonal difference of the groundwater systems to achieve groundwater conservation.

Estimating spatio-temporal dynamics of stream total phosphate concentration by soft computing techniques.

This study attempts to model the spatio-temporal dynamics of total phosphate (TP) concentrations along a river for effective hydro-environmental management. We propose a systematical modeling scheme (SMS), which is an ingenious modeling process equipped with a dynamic neural network and three refined statistical methods, for reliably predicting the TP concentrations along a river simultaneously. Two different types of artificial neural network (BPNN-static neural network; NARX network-dynamic neural network) are constructed in modeling the dynamic system. The Dahan River in Taiwan is used as a study case, where ten-year seasonal water quality data collected at seven monitoring stations along the river are used for model training and validation. Results demonstrate that the NARX network can suitably capture the important dynamic features and remarkably outperforms the BPNN model, and the SMS can effectively identify key input factors, suitably overcome data scarcity, significantly increase model reliability, satisfactorily estimate site-specific TP concentration at seven monitoring stations simultaneously, and adequately reconstruct seasonal TP data into a monthly scale. The proposed SMS can reliably model the dynamic spatio-temporal water pollution variation in a river system for missing, hazardous or costly data of interest.

Reinforced two-step-ahead weight adjustment technique for online training of recurrent neural networks.

A reliable forecast of future events possesses great value. The main purpose of this paper is to propose an innovative learning technique for reinforcing the accuracy of two-step-ahead (2SA) forecasts. The real-time recurrent learning (RTRL) algorithm for recurrent neural networks (RNNs) can effectively model the dynamics of complex processes and has been used successfully in one-step-ahead forecasts for various time series. A reinforced RTRL algorithm for 2SA forecasts using RNNs is proposed in this paper, and its performance is investigated by two famous benchmark time series and a streamflow during flood events in Taiwan. Results demonstrate that the proposed reinforced 2SA RTRL algorithm for RNNs can adequately forecast the benchmark (theoretical) time series, significantly improve the accuracy of flood forecasts, and effectively reduce time-lag effects.

Forecasting of ozone episode days by cost-sensitive neural network methods.

Forecasting the occurrence of ozone episode days can be regarded as an imbalanced dataset classification problem. Since the standard artificial neural network (ANN) methods cannot make accurate predictions of such a problem, two cost-sensitive ANN methods, cost-penalty and moving threshold, were used in this study. The models classify each day as episode or non-episode according to the standard of daily maximum 8 h O(3) concentration. The ozone measurements from six monitoring stations in Taiwan were used for model training and performance evaluation. Two different input datasets, regional and single-site, were generated from raw air quality and meteorological observations. According to the numerical experiments, the predictions based on the regional dataset are much better than those obtained from the single-site dataset. Two cost-sensitive ANN methods were evaluated by receiver operating characteristic (ROC) curves. It was found that the results obtained by the two approaches are similar. If the misclassification costs are known, the cost-sensitive method can minimise the total costs. If the misclassification costs are unknown, the cost-sensitive ANN can obtain a better forecast than the standard ANN method when an appropriate cost ratio is used. For clean areas where episode days are very rare, the forecasts are poor for all methods.