📄 earth science

The Application of the Evidence-Cloud Model On the Water Quality Evaluation

This paper proposes a water quality evaluation method that integrates cloud theory and evidence theory to determine basic probabilities and calculate synthetic probabilities for key indicators, thereby improving predictive accuracy and reliability compared to traditional models.

Original authors: Dan Wang, Xin-Bao Gu

Published 2026-09-29

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Dan Wang, Xin-Bao Gu

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Water is the silent foundation of modern life, flowing through pipes to our taps and seeping through soil to feed crops, yet its purity is rarely a simple matter of sight or smell. In the complex world of environmental science, determining whether groundwater is safe involves navigating a landscape of uncertainty. The data collected from the earth is often fuzzy and random, while the standards for what constitutes "clean" water are rigid lines drawn on a chart. Traditional methods for bridging this gap have struggled; some rely on massive amounts of data that are hard to gather, while others are so conservative that they paint an overly pessimistic picture of water quality. To solve this, scientists have begun combining two powerful ways of thinking: one that handles the fuzziness of real-world measurements, and another that fuses different pieces of evidence to reach a single, reliable conclusion. This approach allows researchers to treat water quality not as a fixed number, but as a spectrum of possibilities that can be weighed and measured with greater precision.

In a recent study focused on the industrial region of Zibo City in Shandong Province, researchers applied this combined method to evaluate the health of local groundwater. The area, characterized by flat terrain and a mix of petrochemical and chemical industries, presented a challenging environment where pollution sources are complex and varied. The team selected five specific indicators to act as the eyes and ears of their assessment: dissolved oxygen, which measures how much air is in the water; the permanganate index, a gauge of organic pollution; total nitrogen and total phosphorus, which track nutrients that can cause algae blooms; and ammonia nitrogen, a common byproduct of waste. They gathered data from ten monitoring wells scattered across the twenty-square-kilometer study area, capturing a snapshot of the water's condition at a specific moment in time.

The researchers did not simply compare these numbers to a pass-or-fail list. Instead, they used a mathematical framework that acknowledges the natural variation in the data. They treated each water quality standard as a "cloud" of possibilities rather than a sharp edge, allowing a measurement that sits near the boundary to be understood in context. This cloud model generated a degree of certainty for each water sample against five different quality grades, ranging from excellent to very poor. Once these individual assessments were made, the team used a fusion technique to blend the five different indicators together. This step was crucial because it allowed the method to weigh the importance of each factor and resolve any conflicts between them, much like a judge weighing testimony from multiple witnesses to reach a verdict. The result was a single, synthesized probability for each location, indicating the most likely water quality grade.

When the team applied this method to the seven cross-sections they analyzed, the results were strikingly accurate. The model correctly identified the water quality at every single site when compared to actual field investigations, achieving a perfect match. In contrast, a more traditional method used for comparison only matched the actual findings eighty percent of the time. The study revealed that while most of the sampled areas had excellent water quality, one specific section showed signs of being merely "good" rather than excellent, primarily due to higher levels of dissolved oxygen fluctuations and nutrient content. The analysis also highlighted which factors mattered most. The researchers found that dissolved oxygen, total nitrogen, and total phosphorus were the most sensitive drivers of water quality changes. This means that small shifts in these three areas have the biggest impact on the overall health of the water, suggesting that monitoring efforts should focus intensely on them.

The study concludes that this new approach offers a more reliable and detailed way to assess groundwater than previous methods. By successfully handling the uncertainty inherent in environmental data, the model provides a clearer picture of where water is safe and where it needs attention. The findings suggest that the groundwater in the studied region is currently in a stable and generally healthy state, with a one hundred percent qualification rate across the tested sections. However, the precision of the method also serves as a warning system; it identifies specific areas and specific chemical factors that require continued vigilance. For water managers and environmental planners, this means they can move beyond guesswork and broad generalizations, using a tool that offers both high accuracy and a deep understanding of the specific forces shaping the water beneath their feet.

Technical Summary: The Application of the Evidence-Cloud Model on Water Quality Evaluation

Problem Statement
Groundwater resources are critical for drinking, industrial, and agricultural use in China, yet their quality assessment is complicated by the fuzziness, randomness, and uncertainty inherent in monitoring data and evaluation criteria. Existing methods for water quality assessment—including neural networks, single-factor index methods, fuzzy comprehensive evaluation, and grey clustering—suffer from specific limitations. Neural networks often require large datasets and suffer from slow convergence or local minima; single-factor methods are overly conservative; fuzzy methods involve complex weight vector calculations with high subjectivity; and grey clustering methods entail massive computational workloads and subjectivity. Consequently, there is a need for a method that can effectively handle the uncertain transformation between qualitative concepts and quantitative descriptions while fusing uncertain information to improve predictive accuracy.

Methodology
The paper proposes a combined evaluation method integrating the Cloud Model and Evidence Theory (specifically the Dempster-Shafer theory), referred to as the "Evidence-Cloud Model." The methodology proceeds through the following steps:

Indicator Selection and Cloud Model Construction: Five key water quality indicators are selected as evidence bodies: Dissolved Oxygen (DO), Permanganate Index, Total Nitrogen (TN), Total Phosphorus (TP), and Ammonia Nitrogen (NH3-N). These are classified into five grades (I to V). Using the cloud model, the study calculates three digital feature parameters (Expectation $Ex$ , Entropy $En$ , and Hyper-entropy $He$ ) for each indicator across the five grades to transform interval standards into cloud droplets.
Membership Degree and Basic Probability Assignment (BPA): Measured values are input into the cloud model to calculate membership degrees for each grade. To satisfy the definition of a Basic Probability Assignment (BPA) required by evidence theory, a certainty degree is introduced to handle uncertainty, converting the membership matrix into a BPA matrix.
Evidence Fusion and Correction: The study employs an improved Dempster-Shafer fusion rule. Before fusion, evidence is screened using a similarity coefficient and a support degree to determine the credibility of each piece of evidence. Conflicting evidence is revised using a correction coefficient based on credibility measures to mitigate the impact of contradictory data.
Synthesis and Defuzzification: The revised evidence bodies are fused using the Dempster combination rule to obtain a synthetic probability for each water quality grade. The final grade is determined by the maximum synthetic probability.
Sensitivity Analysis: A global sensitivity analysis is performed using the Spearman rank correlation coefficient based on Monte Carlo simulation to determine the correlation between input factors and the assessment output, revealing which indicators most significantly influence the results.

Key Contributions

Integration of Theories: The paper successfully combines the cloud model's ability to handle the uncertainty transformation between qualitative and quantitative concepts with the evidence theory's strength in processing "uncertain" and "unknown" information for information fusion.
Improved Accuracy: The proposed method claims to improve predictive accuracy compared to traditional models, specifically the matter-element extension model.
Handling Interval Criteria: The method is designed to effectively predict grade criteria that exist in interval forms, addressing the limitations of methods that rely on single-point values or overly conservative worst-case scenarios.
Quantitative Sensitivity Ranking: The study provides a quantitative ranking of the sensitivity of different water quality indicators, moving beyond simple grading to identify the most critical factors affecting water quality stability.

Results
The model was applied to evaluate water quality at seven cross-sections in Zibo City, Shandong Province, an area characterized by petrochemical and chemical manufacturing industries.

Accuracy: The proposed method achieved a 100% consistency rate with actual investigation results across all seven cross-sections. In contrast, the comparative matter-element extension model achieved only 80% accuracy.
Specific Findings: The assessment determined that cross-sections 1#, 2#, 3#, 5#, 6#, and 7# belong to Grade I (excellent), while cross-section 4# belongs to Grade II (good). The overall qualified rate of water quality across all sections was 100%.
Differentiation Capability: The method demonstrated the ability to differentiate between sections of the same grade based on probability magnitude. For instance, while both 5# and 6# were Grade I, the model assigned a higher synthetic probability (0.9285) to 5# compared to 6# (0.8162), indicating a more robust quality at 5#.
Sensitivity Analysis: The global sensitivity analysis identified DO (X1), TN (X3), and TP (X4) as the three most sensitive factors affecting water quality. The analysis confirmed that increases in these factors (where negative correlation exists) lead to decreased water quality, necessitating targeted monitoring and control.

Significance
The paper concludes that the Evidence-Cloud Model is a feasible and effective approach for water quality assessment. By enhancing predictive efficiency and providing a new perspective for handling the fuzziness and uncertainty of water quality indicators, the method offers a reliable tool for preventing groundwater pollution. The study asserts that the results are entirely consistent with actual field investigations, suggesting that the method can serve as a new approach for future water quality assessments, particularly in complex industrial environments where data uncertainty is high. The authors note that no additional pollution prevention measures are required for the specific sections studied, as the water quality is currently stable and excellent.

More like this