Enhancing water quality prediction for fluctuating missing data scenarios: A dynamic Bayesian network-based processing system to monitor cyanobacteria proliferation.

Tackling the impact of missing data in water management is crucial to ensure the reliability of scientific research that informs decision-making processes in public health. The goal of this study is to ascertain the root causes associated with cyanobacteria proliferation under major missing data scenarios. For this purpose, a dynamic missing data management methodology is proposed using Bayesian Machine Learning for accurate surface water quality prediction of a river from Limia basin (Spain). The methodology used entails a sequence of analytical steps, starting with data pre-processing, followed by the selection of a reliable dynamic Bayesian missing value prediction system, leading finally to a supervised analysis of the behavioral patterns exhibited by cyanobacteria. For that, a total of 2,118,844 data points were used, with 205,316 (9.69 %) missing values identified. The machine learning testing showed the iterative structural expectation maximization (SEM) as the best performing algorithm, above the dynamic imputation (DI) and entropy-based dynamic imputation methods (EBDI), enhancing in some cases the accuracy of imputations by approximately 50 % in R2, RMSE, NRMSE, and logarithmic loss values. These findings can impact how data on water quality is being processed and studied, thus, opening the door for more reliable water management strategies that better inform public health decisions.

A coupled multivariate statistics, geostatistical and machine-learning approach to address soil pollution in a prototypical Hg-mining site in a natural reserve.

The impact of mining activities on the environment is vast. In this regard, many mines were operating well before the introduction of environmental law. This is particularly true of cinnabar mines, whose activity has declined for decades due to growing public concern regarding Hg high toxicity. Here we present the exemplary case study of an abandoned Hg mine located in the Somiedo Natural Reserve (Spain). Until its closure in the 1970s, this mine operated under no environmental regulations, its tailings dumped in two spoil heaps, one of them located uphill and the other in the surroundings of the village of Caunedo. This study attempts to outline the degree to which soil and other environmental compartments have been affected by the two heaps. To this end, we used a novel combination of multivariate statistical, geostatistical and machine-learning methodologies. The techniques used included principal component and clustering analysis, Bayesian networks, indicator kriging, and sequential Gaussian simulations. Our results revealed high concentrations of Hg and, secondarily, As in soil but not in water or sediments. The innovative methodology abovementioned allowed us to identify natural and anthropogenic associations between 25 elements and to conclude that soil pollution was attributable mainly to natural weathering of the uphill heap. Moreover, the probability of surpassing the threshold limits and the local backgrounds was found to be high in a large extension of the area. The methodology used herein demonstrated to be effective for addressing complex pollution scenarios and therefore they are applicable to similar cases.

Combining raw and compositional data to determine the spatial patterns of Potentially Toxic Elements in soils.

When considering complex scenarios involving several attributes, such as in environmental characterization, a clearer picture of reality can be achieved through the dimensional reduction of data. In this context, maps facilitate the visualization of spatial patterns of contaminant distribution and the identification of enriched areas. A set, of 15 Potentially Toxic Elements (PTEs) - (As, Ba, Cd, Co, Cr, Cu, Hg, Mo, Ni, Pb, Sb, Se, Tl, V, and Zn), was measured in soil, collected in Langreo's municipality (80km2), Spain. Relative enrichment (RE) is introduced here to refer to the proportion of elements present in a given context. Indeed, a novel approach is provided for research into PTE fate. This method involves studying the variability of PTE proportions throughout the study area, thereby allowing the identification of dissemination trends. Traditional geostatistical approaches commonly use raw data (concentrations) accepting that the elements analyzed make up the entirety of the soil. However, in geochemical studies the analyzed elements are just a fraction of the total soil composition. Therefore, considering compositional data is pivotal. The spatial characterization of PTEs considering raw and compositional data together allowed a broad discussion about, not only the PTEs concentration's distribution but also to reckon possible trends of relative enrichment (RE). Transformations to open closed data are widely used for this purpose. Spatial patterns have an indubitable interest. In this study, the Centered Log-ratio transformation (clr) was used, followed by its back-transformation, to build a set of compositional data that, combined with raw data, allowed to establish the sources of the PTEs and trends of spatial dissemination. Based on the obtained findings it was possible to conclude that the Langreo area is deeply affected by its industrial and mining legacy. City center is highly enriched in Pb and Hg and As shows enrichment in a northwesterly direction.

Developing a new Bayesian Risk Index for risk evaluation of soil contamination.

Industrial and agricultural activities heavily constrain soil quality. Potentially Toxic Elements (PTEs) are a threat to public health and the environment alike. In this regard, the identification of areas that require remediation is crucial. In the herein research a geochemical dataset (230 samples) comprising 14 elements (Cu, Pb, Zn, Ag, Ni, Mn, Fe, As, Cd, V, Cr, Ti, Al and S) was gathered throughout eight different zones distinguished by their main activity, namely, recreational, agriculture/livestock and heavy industry in the Avilés Estuary (North of Spain). Then a stratified systematic sampling method was used at short, medium, and long distances from each zone to obtain a representative picture of the total variability of the selected attributes. The information was then combined in four risk classes (Low, Moderate, High, Remediation) following reference values from several sediment quality guidelines (SQGs). A Bayesian analysis, inferred for each zone, allowed the characterization of PTEs correlations, the unsupervised learning network technique proving to be the best fit. Based on the Bayesian network structure obtained, Pb, As and Mn were selected as key contamination parameters. For these 3 elements, the conditional probability obtained was allocated to each observed point, and a simple, direct index (Bayesian Risk Index-BRI) was constructed as a linear rating of the pre-defined risk classes weighted by the previously obtained probability. Finally, the BRI underwent geostatistical modeling. One hundred Sequential Gaussian Simulations (SGS) were computed. The Mean Image and the Standard Deviation maps were obtained, allowing the definition of High/Low risk clusters (Local G clustering) and the computation of spatial uncertainty. High-risk clusters are mainly distributed within the area with the highest altitude (agriculture/livestock) showing an associated low spatial uncertainty, clearly indicating the need for remediation. Atmospheric emissions, mainly derived from the metallurgical industry, contribute to soil contamination by PTEs.