Scientometric Analysis Maps ML Trends in Water Prediction

By Dr Silpaja Chandrasekar, PhDReviewed by Susha Cheriyedath, M.Sc.Aug 23 2024

In a paper published in the journal Artificial Intelligence in Geosciences, researchers examined 876 articles on water prediction from 2015 to 2022, focusing on hydrological patterns and demand forecasts. Utilizing CiteSpace and bibliometric techniques, they mapped key countries, institutions, authors, and trends in machine learning (ML) and deep learning (DL) for water prediction. The study highlighted significant research domains and emerging areas, offering guidance for future inquiries in the field.

Related Work

Previous works have highlighted the crucial role of water forecasting for sustainable management and its applications in disaster preparedness, agriculture, and resource allocation. Despite significant advancements in ML and DL for water prediction, a comprehensive review of this literature must be available. Challenges persist in integrating diverse data sources and optimizing models for real-world applications. Furthermore, improved visualization tools are required to enhance comprehension of intricate forecasting processes.

CiteSpace Analysis Overview

This study employed CiteSpace software to visualize knowledge graphs and analyze time series forecasting using ML and DL techniques in hydrology. CiteSpace, a Java application, excels at visualizing trends in scientific literature, identifying emerging topics, citation hotspots, and clustering networks based on citation terms. CiteSpace (version R6.1.6) was used to process and visualize data from 876 studies, removing duplicates and configuring parameters to cover the period from 2015 to 2022 with yearly time slices.

The analysis examined article titles, abstracts, author names, and keywords, identifying significant publications, influential authors, and recurring keywords. The Institute for Scientific Information (ISI) Web of Science database reviewed and processed 876 relevant articles on ML, DL, and water forecasting from 2015 to 2022. 

Research Trends Unveiled

The research progress in water forecasting using ML and DL was assessed by analyzing the volume of academic literature over time. The data from 2015 to 2022 revealed two distinct stages: an initial development phase from 2015 to 2018 and a period of rapid growth from 2019 to 2022. Publications increased steadily, reaching a peak of 235 in 2022, although the annual growth rate has gradually declined since 2018.

The collaboration network analysis highlighted key contributors in the field, including countries, institutions, and authors. The top ten countries with the highest publication volumes were the United States, China, Iran, India, England, South Korea, Vietnam, Australia, Canada, and Spain. The United States led publication count and degree centrality, indicating extensive collaborative relationships. China and Iran followed in publication frequency, while Iran and China had significant collaboration ties.

Institutional collaboration was also analyzed, revealing a network of 284 institutions with varying degrees of cooperation. The University of the Chinese Academy of Sciences had the highest publication volume, while Duy Tan University from Vietnam demonstrated the greatest degree of centrality, indicating broad international collaboration.

Other prominent institutions included the University of Tehran, Texas A&M University, and Ton Duc Thang University. The co-citation analysis revealed that the Journal of Hydrology and Science of the Total Environment was the most cited. At the same time, prominent authors significantly advanced flood hazard prediction and related research areas. 

A comprehensive analysis of water forecasting research utilizing ML and DL was conducted by examining the volume of academic literature from 2015 to 2022. The data revealed two key phases: an initial development period from 2015 to 2018 and rapid growth from 2019 to 2022. The number of publications consistently increased, peaking at 235 in 2022.

Despite this growth, the annual increase in publication numbers has gradually slowed since 2018. The analysis used CiteSpace to map out keywords, highlighting the most frequently used terms such as "prediction," "ML," and "climate change," which were integral to understanding research hotspots and trends.

The keyword co-citation network identified critical research areas and provided insight into emerging topics. Keywords with high frequency included "prediction" and "ML," each appearing 128 times. However, keywords like "land use change" and "cellular automata," while less frequent, showed higher importance based on centrality measures. The clustering analysis produced ten distinct research clusters, including topics like "landslide susceptibility," "support vector regression," and "water quality," reflecting the diverse focus areas within water forecasting research.

 The keyword bursts analysis illustrated shifts in research focus over time, with significant bursts observed for terms such as "climate variability" and "flash flood." These bursts indicate periods of heightened interest in specific topics, with "temperature" exhibiting the longest duration of research intensity.

The summary outlines the field's development, identifying leading countries, institutions, and influential literature. This temporal analysis provides a structured overview of the domain's evolution, underscoring the contributions of prominent authors and highlighting the growing significance of various research themes in water forecasting with ML and DL methodologies.

Conclusion

To sum up, the analysis revealed significant growth in water forecasting research between 2019 and 2022, marked by increased publications and citation rates. Leading contributors included the United States, China, Iran, India, and England, although collaboration within the field remained limited.

Despite China's high publication volume, its papers were less cited, likely due to limited international cooperation. Recent studies focused on climate variability, flood risk, and hybrid modeling, indicating that ML and DL methodologies have greatly advanced the field.