A time series based machine learning strategy for wastewater-based forecasting and nowcasting of COVID-19 dynamics

Challenges on the implementation of wastewater-based epidemiology as a prediction tool: the paradigm of SARS-CoV-2

Wastewater Based Epidemiology (WBE) has been identified as a tool for monitoring and predicting patterns of SARS-CoV-2 in communities. Several factors may lead to a day-to-day variation in the measurement of viral genetic material. Wastewater samples are systematically collected from the two major wastewater treatment plants in Crete, Greece. Physico-chemical factors were tested, viral concentration was determined by RT-real time PCR and the results were normalized. The influence of restriction measures, rain and physico-chemical agents was addressed. Statistics together with machine learning (ML) were applied to predict human cases. 781 samples were analyzed. RNA concentration was reduced during lockdown and was impacted by rain. Fluctuations in pH and total solids' concentrations were associated with changes in viral load. Conductivity was mainly related to chloride ions. In Heraklion, wastewater viral load preceded human cases by three days on average. Cross- correlation estimates did not perform likewise in Chania. According to ML, the ratio of sewage RNA measurements to reported cases decreased in comparison to the first wave, due to different variants, climatological parameters, testing rate and behaviors related to seeking healthcare. The model developed showed a close approximation between recorded and predicted cases. Parameters such as total solids, pH, conductivity, rain and inhibitors can significantly impact the recovery of viral RNA. The correlation between viral load in wastewater and human cases is not straightforward. The application of ML may fill some but not every gap. Existing models cannot be directly applied to different Wastewater Treatment Plants or countries.

Interpretation of COVID-19 Epidemiological Trends in Mexico Through Wastewater Surveillance Using Simple Machine Learning Algorithms for Rapid Decision-Making

Detection and quantification of disease-related biomarkers in wastewater samples, denominated Wastewater-based Surveillance (WBS), has proven a valuable strategy for studying the prevalence of infectious diseases within populations in a time- and resource-efficient manner, as wastewater samples are representative of all cases within the catchment area, whether they are clinically reported or not. However, analysis and interpretation of WBS datasets for decision-making during public health emergencies, such as the COVID-19 pandemic, remains an area of opportunity. In this article, a database obtained from wastewater sampling at wastewater treatment plants (WWTPs) and university campuses in Monterrey and Mexico City between 2021 and 2022 was used to train simple clustering- and regression-based risk assessment models to allow for informed prevention and control measures in high-affluence facilities, even if working with low-dimensionality datasets and a limited number of observations. When dividing weekly data points based on whether the seven-day average daily new COVID-19 cases were above a certain threshold, the resulting clustering model could differentiate between weeks with surges in clinical reports and periods between them with an 87.9% accuracy rate. Moreover, the clustering model provided satisfactory forecasts one week (80.4% accuracy) and two weeks (81.8%) into the future. However, the prediction of the weekly average of new daily cases was limited (R2 = 0.80, MAPE = 72.6%), likely because of insufficient dimensionality in the database. Overall, while simple, WBS-supported models can provide relevant insights for decision-makers during epidemiological outbreaks, regression algorithms for prediction using low-dimensionality datasets can still be improved.

A new strong convective precipitation forecasting method based on attention mechanism and spatio-temporal reasoning

Radar observation variables reflect the precipitation amount of strong convective precipitation processes, which accurate forecast is an important difficulty in weather forecasting. Current forecasting methods are mostly based on radar echo extrapolation, which has the insufficiency of input information and the ineffectiveness of model architecture. This paper presents a Bidirectional Long Short-Term Memory forecasting method for strong convective precipitation based on the attention mechanism and residual neural network (ResNet-Attention-BiLSTM). First, this paper uses ResNet to effectively extract the key information of extreme weather and solves the problem of regression to the mean of the prediction model by learning the residuals of the radar observation data. Second, this paper uses the attention mechanism to adaptively weight the fusion of the features to enhance the extraction of the important features of the precipitation image data. On this basis, this paper presents a novel spatio-temporal reasoning method for radar observations and establishes a precipitation forecasting model, which captures the past and future time-order relationship of the sequence data. Finally, this paper conducts experiments based on the real collected data of a strong convective precipitation process and compares its performance with the existing models, the mean absolute percentage error of this model was reduced by 15.94% (1 km), 18.72% (3 km), and 14.91% (7 km), and the coefficient of determination ( R 2 ) was increased by 10.89% (1 km), 9.61% (3 km), and 9.29% (7 km), which proves the state of the art and effectiveness of this forecasting model.

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.

Integrating socio-economic vulnerability factors improves neighborhood-scale wastewater-based epidemiology for public health applications

Wastewater Based Epidemiology (WBE) of COVID-19 is a low-cost, non-invasive, and inclusive early warning tool for disease spread. Previously studied WBE focused on sampling at wastewater treatment plant scale, limiting the level at which demographic and geographic variations in disease dynamics can be incorporated into the analysis of certain neighborhoods. This study demonstrates the integration of demographic mapping to improve the WBE of COVID-19 and associated post-COVID disease prediction (here kidney disease) at the neighborhood level using machine learning. WBE was conducted at six neighborhoods in Seattle during October 2020 - February 2022. Wastewater processing and RT-qPCR were performed to obtain SARS-CoV-2 RNA concentration. Census data, clinical data of COVID-19, as well as patient data of acute kidney injury (AKI) cases reported during the study period were collected and the distribution across the city was studied using Geographic Information System (GIS) mapping. Further, we analyzed the data set to better understand socioeconomic impacts on disease prevalence of COVID-19 and AKI per neighborhood. The heterogeneity of eleven demographic factors (such as education and age among others) was observed within neighborhoods across the city of Seattle. Dynamics of COVID-19 clinical cases and wastewater SARS-CoV-2 varied across neighborhood with different levels of demographics. Machine learning models trained with data from the earlier stages of the pandemic were able to predict both COVID-19 and AKI incidence in the later stages of the pandemic (Spearman correlation coefficient of 0·546 - 0·904), with the most predictive model trained on the combination of wastewater data and demographics. The integration of demographics strengthened machine learning models' capabilities to predict prevalence of COVID-19, and of AKI as a marker for post-COVID sequelae. Demographic-based WBE presents an effective tool to monitor and manage public health beyond COVID-19 at the neighborhood level.

Online trend estimation and detection of trend deviations in sub-sewershed time series of SARS-CoV-2 RNA measured in wastewater

Wastewater surveillance has proven a cost-effective key public health tool to understand a wide range of community health diseases and has been a strong source of information on community levels and spread for health departments throughout the SARS- CoV-2 pandemic. Studies spanning the globe demonstrate the strong association between virus levels observed in wastewater and quality clinical case information of the population served by the sewershed. Few of these studies incorporate the temporal dependence present in sampling over time, which can lead to estimation issues which in turn impact conclusions. We contribute to the literature for this important public health science by putting forward time series methods coupled with statistical process control that (1) capture the evolving trend of a disease in the population; (2) separate the uncertainty in the population disease trend from the uncertainty due to sampling and measurement; and (3) support comparison of sub-sewershed population disease dynamics with those of the population represented by the larger downstream treatment plant. Our statistical methods incorporate the fact that measurements are over time, ensuring correct statistical conclusions. We provide a retrospective example of how sub-sewersheds virus levels compare to the upstream wastewater treatment plant virus levels. An on-line algorithm supports real-time statistical assessment of deviations of virus level in a population represented by a sub-sewershed to the virus level in the corresponding larger downstream wastewater treatment plant. This information supports public health decisions by spotlighting segments of the population where outbreaks may be occurring.

An interpretable time series machine learning method for varying forecast and nowcast lengths in wastewater-based epidemiology

Wastewater-based epidemiology has emerged as a viable tool for monitoring disease prevalence in a population. This paper details a time series machine learning (TSML) method for predicting COVID-19 cases from wastewater and environmental variables. The TSML method utilizes a number of techniques to create an interpretable, hypothesis-driven framework for machine learning that can handle different nowcast and forecast lengths. Some of the techniques employed include:•Feature engineering to construct interpretable features, like site-specific lead times, hypothesized to be potential predictors of COVID-19 cases.•Feature selection to identify features with the best predictive performance for the tasks of nowcasting and forecasting.•Prequential evaluation to prevent data leakage while evaluating the performance of the machine learning algorithm.