Machine Learning-Aided Causal Inference Framework for Environmental Data Analysis: A COVID-19 Case Study

Estimating the causal impact of non-pharmaceutical interventions on COVID-19 spread in seven EU countries via machine learning

During the COVID-19 pandemic, Non-Pharmaceutical Interventions (NPIs) were imposed all over Europe with the intent to reduce infection spread. However, reports on the effectiveness of those measures across different European countries are inconclusive up to now. Moreover, attempts to predict the effect of NPIs in a prospective and dynamical manner with the aim to support decision makers in future global health emergencies are largely lacking. Here, we explore causal machine learning to isolate causal effects of NPIs in observational public health data from seven EU countries, taking into account specific challenges like their sequential nature, effect heterogeneity, time-dependent confounding and lack of robustness due to violated assumptions. In a pseudo-prospective scenario planning analysis, we investigate which recommendations our model would have made during the second wave of the pandemic in Germany, demonstrating its capacity to generalize to the near future and identifying effective NPIs. In retrospect, our approach indicates that a wide range of response measures curbed COVID-19 across countries, especially in the early phases of the pandemic. Interestingly, this includes controversial interventions like strict school and border closures, but also recommendation-based policies in Sweden. Finally, we discuss important data- and modeling-related considerations that may optimize causal effect estimation in future pandemics.

Causal-inference machine learning reveals the drivers of China's 2022 ozone rebound

Ground-level ozone concentrations rebounded significantly across China in 2022, challenging air quality management and public health. Identifying the drivers of this rebound is crucial for designing effective mitigation strategies. Commonly used methods, such as chemical transport models and machine learning, provide valuable insights but face limitations-chemical transport models are computationally intensive, while machine learning often fails to address confounding factors or establish causality. Here we show that elevated temperatures and increased solar radiation, as primary meteorological drivers, collectively account for 57 % of the total ozone increase, based on an integrated analysis of ground-based monitoring data, satellite observations, and meteorological reanalysis information using explainable machine learning and causal inference techniques. Compared to the year 2021, 90 % of the stations reported an increase in the Formaldehyde to Nitrogen ratio, implying a growing sensitivity of ozone formation to nitrogen oxide levels. These findings highlight the significant causal role of meteorological changes in the ozone rebound, urging the adoption of targeted ozone mitigation strategies under climate warming, particularly through varied regional strategies that consider existing anthropogenic emission levels and the prospective increase in biogenic volatile organic compounds. This identification of causal relationships in air pollution dynamics can support data-driven and accurate decision-making.

Causal prior-embedded physics-informed neural networks and a case study on metformin transport in porous media

This study introduces a novel approach to transport modelling by integrating experimentally derived causal priors into neural networks. We illustrate this paradigm using a case study of metformin, a ubiquitous pharmaceutical emerging pollutant, and its transport behaviour in sandy media. Specifically, data from metformin's sandy column transport experiment was used to estimate unobservable parameters through a physics-based model Hydrus-1D, followed by a data augmentation to produce a more comprehensive dataset. A causal graph incorporating key variables was constructed, aiding in identifying impactful variables and estimating their causal dynamics or "causal prior." The causal priors extracted from the augmented dataset included underexplored system parameters such as the type-1 sorption fraction F, first-order reaction rate coefficient α, and transport system scale. Their moderate impact on the transport process has been quantitatively evaluated (normalized causal effect 0.0423, -0.1447 and -0.0351, respectively) with adequate confounders considered for the first time. The prior was later embedded into multilayer neural networks via two methods: causal weight initialization and causal prior regularization. Based on the results from AutoML hyperparameter tuning experiments, using two embedding methods simultaneously emerged as a more advantageous practice since our proposed causal weight initialization technique can enhance model stability, particularly when used in conjunction with causal prior regularization. amongst those experiments utilizing both techniques, the R-squared values peaked at 0.881. This study demonstrates a balanced approach between expert knowledge and data-driven methods, providing enhanced interpretability in black-box models such as neural networks for environmental modelling.

Machine learning coupled with causal inference to identify COVID-19 related chemicals that pose a high concern to drinking water

Various synthetic substances were utilized in large quantities during the recent coronavirus pandemic, COVID-19. Some of these chemicals could potentially enter drinking water sources. Persistent, mobile, and toxic (PMT) substances have been recognized as a threat to drinking water resources. It has not yet been assessed how many COVID-19 related substances could be considered PMT substances. One reason is the lack of high-quality experimental data for the identification of PMT substances. To solve this problem, we applied a machine learning model to identify the PMT substances among COVID-19 related chemicals. The optimal model achieved an accuracy of 90.6% based on external test data. The model interpretation and causal inference indicated that our approach understood causation between PMT properties and molecular descriptors. Notably, the screening results showed that over 60% of the COVID-19 chemicals considered are candidate PMT substances, which should be prioritized to prevent undue pollution of water resources.

Transport of Microplastic and Dispersed Oil Co-contaminants in the Marine Environment

Microplastics (MPs) and oil pollution are major concerns in oceans. Although their coexistence in oceans and the associated MP-oil-dispersant agglomerates (MODAs) have been reported, limited attention is given to the behavior of the co-contaminants. This study investigated MODA transport in a simulated ocean system and explored related mechanisms under various oil types, salinities, and mineral concentrations. We found that more than 90% of the heavy oil-formed MODAs stayed at the seawater surface, while the light oil-formed MODAs were widely distributed throughout the seawater column. The increased salinity promoted MODAs formed by 7 and 90 μm MPs to transport from the seawater surface to the column. This was elucidated by the Derjaguin-Landau-Verwey-Overbeek theory as more MODAs formed under higher salinities and dispersants kept them stable in the seawater column. Minerals facilitated the sinking of large MP-formed MODAs (e.g., 40 μm) as minerals were adsorbed on the MODA surface, but their impact on small MP-formed MODAs (e.g., 7 μm) was negligible. A MODA-mineral system was proposed to explain their interaction. Rubey's equation was recommended to predict the sinking velocity of MODAs. This study is the first attempt to reveal MODA transport. Findings will contribute to the model development to facilitate their environmental risk evaluation in oceans.

Impacts of climate change and evapotranspiration on shrinkage of Aral Sea

The massive desiccation of the Aral Sea, the fourth largest lake in the world, has led to severe ecological problems, expansion of cropland was thought to be the main factor driving that shrinkage. But this study performed a long-term land cover and use change assessment for Aral Sea Basin (ASB) to show that the cropland has stopped expanding in 2000, of which the cropland in the ASB plain area has decreased significantly (-140 km²/year) from 2001 to 2019. By contrast, this study finds the hydrological cycle in the ASB has intensified through a spatial and temporal scale approach based on Earth observation. Specifically, there is a 7.21 % (+304.56 × 10⁸ m³) increase in annual total precipitation and a 10.13 % (+376.21 × 10⁸ m³) increase in annual total actual evapotranspiration (AET) for the whole ASB during 1980-2019. In particular, the total annual AET in the ASB plain area has increased by 37.81 % (+718.92 × 10⁸ m³), which almost depletes the water that should have flowed into the Aral Sea. Therefore, the Aral Sea shrank by 5625 × 10⁸ m³ (or 42,944.32km²) from 1980 to 2019. Changing climate and increasing AET have accelerated the desiccation of the Aral Sea, and the expansion of cropland is no longer the main factor of that shrinkage. After more water was conserved in the ASB plain area, evapotranspiration plays a more vital role in the Aral Sea shrinkage. Reducing AET and unproductive water losses are key initiatives in future projects to save the Aral Sea. This study explores the causes of Aral Sea shrinkage from an integrated perspective of climate-land-water-ecological change across the ASB, bridging the limitations of previous studies that have focused on Aral Sea waters and subbasins.