Advances in hydrological research in China over the past two decades: Insights from advanced large language model and topic modeling

Highlights

• •Advanced technologies including large language model and topic modeling revealed evolving research patterns and priorities.
• •Hydrology research in China grew 19 papers/year, with increasing author collaboration from 2000 to 2023.
• •SWAT, VIC, and XAJ models are the most popular hydrological models in China’s hydrological research.
• •Research focus shifted from water resources to ecological hydrology, climate change, and carbon dynamics.

Abstract

Hydrology investigates the water cycle and its relationship with the environment and societal activities, and forms the foundation for sustainable water resource management and ecological conservation. A quantitative review of hydrology aids in the scientific prediction of water resource trends, optimizing their allocation, and addressing challenges posed by climate change. However, traditional studies have lacked a systematic quantitative review of Chinese hydrology, particularly at the basin scale and across diverse research topics. Therefore, we employed advanced technologies, including large language models, regular expression techniques, and dynamic topic modeling, to conduct a quantitative analysis of hydrological research in China from 2000 to 2023 based on 289,513 hydrology-related publications. We examined the characteristics, research topics, and spatiotemporal evolution patterns of publications from China’s major basins. The results reveal a significant increase in Chinese hydrology publications (about 19 papers/year) and research collaboration (0.9 authors/decade). The soil and water assessment tool (46.7% usage), variable infiltration capacity model (15.7%), and Xinanjiang model (11.9%) were the most widely used hydrological models in Chinese research. The Yellow River, Yangtze River, and Haihe River basins received the most attention, with key research topics spanning water resources (13.9%), climate change (13.6%), and hydrological modeling (10.8%). A shift from resource-oriented studies to ecological hydrology was observed, with a growing focus on climate change and carbon dynamics. These findings reflect the evolution of Chinese hydrology from traditional water resource management to addressing contemporary environmental challenges, offering valuable insights for future research in China and globally.

Graphical abstract

1. Introduction

As a fundamental Earth science discipline, hydrology plays a crucial role in understanding water cycle processes and managing water resources [1,2]. Over the past three decades, China has experienced unprecedented economic growth and urbanization, accompanied by increasing challenges in water resources management, flooding, and ecological protection [3,4]. This has driven rapid development in Chinese hydrological research, which is marked by evolving research priorities, methodological innovations, and expanding scientific collaboration.

The advancement of hydrological science in China has been characterized by several key developments. First, the emergence of sophisticated hydrological models has significantly enhanced our ability to simulate and predict hydrological processes across various spatial and temporal scales [1,5]. The growing recognition of climate change impacts has shifted research focus from traditional resource-oriented hydrology toward ecological and environmental concerns [6,7]. In addition, the establishment of extensive monitoring networks and the accumulation of long-term observational data have enabled more comprehensive basin-scale studies and data-intensive analyses [8,9].

However, despite these advances, several critical gaps remain in our understanding of the development of hydrological research in China. Previous studies have examined global trends in hydrology [1,10,11] or focused on specific aspects of water resources research, such as ecohydrology, urban hydrology, hydrometeorology, isotopic hydrology, and cryospheric hydrology [[12], [13], [14], [15], [16]]. Nevertheless, there is still a lack of systematic, quantitative analysis of the advancement of Chinese hydrology, particularly in the quantitative analysis of research hotspots in basins. On the other hand, traditional bibliometric approaches often struggle to effectively handle and analyze the vast volume of literature, particularly in identifying specific research domains and extracting geographic information from publications [10,17]. Moreover, existing bibliometric approaches predominantly focus on analyzing publication metadata (such as authorship, citations, and keywords), while neglecting the wealth of information contained within publication abstracts [6,18]. This limitation hampers a comprehensive understanding of the evolution of research topics, methodologies, and geographical focal areas. The inability to effectively analyze this information at scale obstructs our understanding of the research patterns and trends across China’s diverse hydrological regions.

With the advancement of natural language processing (NLP) and artificial intelligence, large language models (LLMs), exemplified by OpenAI’s ChatGPT, have introduced new opportunities for analyzing vast amounts of scientific literature [19,20]. These tools enable efficient extraction and analysis of information from publications, including the identification of study areas, research topics, and methodological approaches [6,18,21]. Topic modeling, another fundamental technique in NLP, facilitates the rapid identification of research topics and their evolution within publications [6,18,21]. When integrated with traditional bibliometric methods, LLMs and topic modeling technologies provide a powerful framework for understanding research evolution and trends.

Therefore, this study aims to conduct a comprehensive quantitative analysis of the progress in Chinese hydrological research from 2000 to 2023. The specific objectives are threefold: (1) to analyze the temporal evolution of publication characteristics, including publication volume, author collaboration patterns, and institutional contributions; (2) to identify and track the development of major research topics and methodological approaches; (3) to examine the spatial distribution and thematic focus of hydrological research across China’s major basins. By achieving these objectives, this study provides valuable insights into the development trajectory of Chinese hydrology, helps identify research gaps and priorities, and offers guidance for future research directions. The findings will benefit both the Chinese and international hydrological communities in understanding research trends and promoting scientific collaboration.

2. Materials and methods

The study can be divided into three main components: data collection and intelligent parsing of basin information, analysis of the publication characteristics, and topic modeling. The research spans the period from January 2000 to October 2023. An overview of the workflow is presented in Fig. 1, which we will now introduce step by step.

Fig. 1.

Study framework. Map is edited on Chinese standard map GS (2019) 1682.

2.1. Data collection and intelligent parsing of basin information

We utilized the Web of Science (WoS) database to gather information about publications related to hydrology in November 2023. The search was conducted using the keywords “hydrolo*” and “water cycle” with the publication timeframe restricted to January 2000 through October 2023. Data collected included publication type, journal name, title, authors, keywords, abstract, affiliations, publication year, and the number of times cited. Duplicate entries were removed based on their WoS IDs, and only articles, reviews, letters, and proceedings papers were retained. Information from a total of 289,513 documents was successfully retrieved.

Basins are the fundamental units of hydrology, and conducting quantitative analyses of various basin studies is crucial for assessing hydrologic progress. Traditional methods struggle to match vast numbers of publications with their corresponding basins. However, with the development of deep learning and artificial intelligence, large language models (LLM) such as ChatGPT and DeepSeek R1 have demonstrated near-human-level text comprehension, offering new possibilities for parsing basin information from extensive literature [19,20]. In this study, we leveraged a framework proposed by Miao et al. [10], combining LLM and geocoding techniques to intelligently parse basin information.

Large Language Model Meta AI (LLaMA) 30B is a pre-trained LLM developed by Meta, part of the LLaMA series [22]. LLaMA 30B is trained on diverse datasets to perform various tasks, including text generation, translation, summarization, and question-answering. To further enhance its performance on specific tasks, such as question answering, the Vicuna dataset and WizardLM method were used to fine-tune the LLaMA 30B model [23,24]. Considering the high hardware demands of LLMs, the generative pre-trained transformers (GPTQ) method was employed to reduce the hardware requirements for model inference and enhance processing speed [25]. The Wizard-Vicuna-30B-Uncensored GPTQ model was developed by applying these techniques, achieving an effective balance between accuracy and efficiency. Thus, we utilized this model as the foundational model for intelligent parsing of basin information, with further details provided in the supplementary materials.

In the field of natural language processing and LLM applications, a prompt refers to the input or instruction provided to the model to generate a desired output [26]. It serves as a context or query that guides the model’s response, influencing the quality, relevance, and accuracy of the generated content based on the prompt’s clarity. In this study, we designed the prompt based on publication titles and abstracts, formatted according to the supplementary materials. By inputting the prompts into the final Wizard-Vicuna-30B-Uncensored GPTQ model, we successfully extracted the research area names for each publication. To enhance the reliability and reproducibility of the results, we have uploaded the Wizard-Vicuna-30B-Uncensored GPTQ model, the parameters used during inference, and the prompt templates to the Zenodo platform (https://doi.org/10.5281/zenodo.15095630).

Then, Google Maps’ geocoding API was utilized to convert the extracted research area names into geographic coordinates (latitude and longitude). In total, geographic locations were identified for 98,311 publications. These locations were then matched with the basin data from the dataset of major river basins in China (http://www.geodata.cn) to determine the corresponding basin for each study. The publications retrieved by the above keywords inevitably contain many studies from related disciplines, such as studies about water environment and hydrochemistry [10]. Given the distinctive characteristics, publication volumes, and influencing factors across disciplines, we focused exclusively on publications closely related to hydrology. Thus, we retained only 25,827 publications published in the journal, which are categorized under the “Water Resources” category in the WoS subfields. Further filtering for studies conducted within China resulted in 4177 hydrological-related publications, with their research areas located within China.

2.2. Analysis of the publication characteristics

We analyzed the annual publication volume and the average number of authors per publication, employing the Theil-Sen slope estimator [27] and the Mann-Kendall test [28] to examine their trends. To assess a journal’s influence in hydrology, we calculated the total citations of all hydrology-related articles in the journal divided by the total number of articles, following the methodology of Miao et al. [10]. To analyze the influence of various research institutions in China within the field of hydrology, we calculated the total number of publications attributed to each institution. Each institution’s count increased by one whenever it contributed to a publication, regardless of its position in the sequences of author and institution, treating all affiliations equally. For instance, if a publication is co-authored by researchers from three institutions, each of these institutions is credited with one additional publication. The total number of publications associated with an institution was regarded as a measure of its impact in hydrology. Notably, only primary affiliations were taken into account; for example, all publications from different colleges within the same university were aggregated under the university’s name. The sole exception was the Chinese Academy of Sciences (CAS), whose individual institutes were treated as independent research entities and counted separately. Furthermore, research institutions were categorized into three groups—CAS, universities, and other research institutions, with the number and proportion of publications calculated for each category.

Co-occurrence analysis of keywords is a widely used bibliometric method that examines the relationships between keywords in academic literature to uncover research themes, hotspots, and developmental trends within a specific field [29]. Van Eck and Waltman [30] introduced the visualization of similarities (VOS) method to quantify similarities between objects and later developed the VOSviewer software [31], providing a powerful tool for co-occurrence analysis. In this study, we employed the VOS method available in VOSviewer to construct a co-occurrence network of keywords, enabling the identification of co-occurrence patterns. In the network visualization, the size of a node reflects the frequency of a keyword’s occurrence, with larger nodes representing higher frequencies. The thickness of the connecting lines between two keywords indicates their co-occurrence frequency, with thicker lines denoting higher frequencies. Keywords are grouped into distinct clusters using the co-occurrence clustering method, and each cluster is represented by a unique color.

Regular expression matching, rooted in pattern recognition, enables the search, matching, or manipulation of specific patterns within text based on predefined expressions (regular expressions) [32]. It is highly flexible and efficient, capable of rapidly extracting required information from vast amounts of text. Therefore, we employed this method to parse potential hydrological models utilized in the publications from abstracts and titles. The regular expressions used for this analysis are detailed in Table S1 of the supplementary materials.

2.3. Topic modeling

To uncover potential thematic structures from extensive publications, we employ topic modeling techniques, which can extract semantic information from extensive text based on statistical methods [6]. Prior to topic modeling, text preprocessing is essential. Abstracts of all publications are vectorized individually, transforming them into lists of terms. Stop words—commonly occurring words with minimal contribution to the semantic meaning of the text, such as articles (e.g., a, an, the), pronouns (e.g., I, we, our), prepositions (e.g., about, as, in), and conjunctions (e.g., and, but, so)—are removed to reduce interference in the modeling process, thereby improving efficiency, accuracy, and interpretability. Accordingly, we excluded the English stop words provided by Natural Language Toolkit (NLTK) [33]. Additionally, we compiled a list of high-frequency yet insignificant terms specific to the hydrology publications (Table S2) and also removed these as stop words. Following this, we extracted bigram phrases based on co-occurrence frequency and applied the scoring method proposed by Mikolov et al. [34]. The bigrams with a score below 100 or occurrences fewer than five times were discarded. Then, we used the spacy en_core_web_trf model to perform lemmatization on the bigram phrases, eliminating the effects of word forms (e.g., singular/plural nouns, verb tenses) on the topic modeling process [35]. Specifically, lemmatization was applied to four parts of speech: nouns, adjectives, verbs, and adverbs.

Once the preprocessing is completed, the focus shifts to the topic modeling. Latent Dirichlet allocation (LDA), proposed by Blei et al. [36], represents a pivotal advancement in the field of topic modeling. By assuming that each document is generated from a mixture of topics, and each topic is characterized by a probability distribution over words, LDA effectively models the relationship between documents and vocabulary. This technique is particularly adept at processing and analyzing large-scale text data, uncovering latent topics within textual content. With the rise of computational power and big data, LDA has found widespread applications in domains such as text classification, information retrieval, and recommendation systems [37,38], establishing itself as one of the core technologies in natural language processing and text mining.

However, LDA is a static topic model, assuming that the topic distribution of each document in a collection remains fixed. In practical applications, academic topics often evolve over time, rendering the LDA model less effective in capturing such dynamic shifts. Dynamic topic model (DTM) improves topic modeling by incorporating temporal factors into the LDA framework [6]. The core idea is that the topic distribution of a document changes over time, with each time point’s distribution being correlated with that of the preceding one [21,39]. Specifically, DTM models the topic distribution for each time period and smooths transitions between different periods, enabling it to capture the dynamic process of topic evolution. DTM allows us to track topic shifts over time, aiding in the understanding of how publications evolve, while also shedding light on the rise and fall of specific themes, offering a better explanation of the evolution of research topics [40]. Therefore, we use the DTM method to extract topics from the abstracts of all publications and analyze the evolution of these topics over time.

We partition the publications according to their publication year and initialize the LDA model using Gensim’s LDA initialization method [41]. The number of latent topics is a crucial parameter in the DTM. To determine the optimal number of latent topics, we employ the coherence score [42] and the elbow method [43]. The search range for the optimal number of topics was set from 4 to 50, with a step size of 2. As shown in Fig. S2 of the supplementary materials, the coherence score reached its peak when the number of topics was 24. Consequently, we selected 24 as the optimal number of latent topics and applied DTM for topic modeling. The DTM model can obtain popularity in various topics over different years. The popularity of each topic at a given time slice is determined by the frequency of its associated keywords within that time slice. A single publication may cover multiple topics, and the total popularity of all topics within a given time slice (one year in this case) sums to one. Topic popularity can help reveal the major research topics in Chinese hydrology and their annual popularity trends. The trend in the popularity of topics over the years is assessed by the Theil-Sen slope estimator [27] and the Mann-Kendall test [28]. The overall popularity of each topic is defined as the average annual popularity.

3. Results

3.1. Characteristics of publications and popularity of hydrologic models

Fig. 2a illustrates the annual trends in publication volume and the average number of authors per paper, represented by the blue and red dashed lines, respectively. The solid lines depict the corresponding Theil-Sen slope estimates and the results of the Mann-Kendall test. It is evident that both publication volume and the average number of authors per paper have exhibited significant upward trends over time (p < 0.001). The publication volume has increased at an average annual rate of approximately 19 papers, while the average number of authors per paper has risen at a rate of 0.9 authors per decade. The sharp increase in publication volume can be attributed to China’s heightened focus on scientific research in recent years, with continuous growth in research investment. By 2017, China became the world’s second-largest spender on research and development, accounting for approximately 21% of global expenditure [44,45]. The rise in the average number of authors reflects the growing collaboration among Chinese hydrologists and an increase in the overall workload in hydrological research [10].

Fig. 2.

(a) the number of publications per year and the average number of authors per article per year; (b) the top 15 journals ranked by influence.

Fig. 2b presents the top 15 most influential journals in the field of Chinese hydrology, with full journal names listed in Table S3 of the supplementary materials. Notably, Water Resources Research emerges as the most influential journal in China’s hydrology field, with an impact factor of 43.6, aligning with previous global findings [10]. Additionally, journals such as Agricultural Water Management (41.6), Environmental Geology (41.4), Water Research (41.2), and Journal of Hydrology (39.0) also have a significant influence on the Chinese hydrological community.

Fig. 3 illustrates the top 15 research institutions in China’s hydrology field by total publication volume. Universities are the mainstay of hydrology research in China. Among them, Hohai University (459), Wuhan University (320), and Beijing Normal University (313) lead the nation in hydrology-related research output. The Chinese Academy of Sciences (CAS) is another important pillar of hydrological research in China, where the Institute of Geographic Sciences and Natural Resources Research (262), Nanjing Institute of Geography and Limnology (184), and Xinjiang Institute of Ecology and Geography (102) are the leading contributors. Additionally, other research institutions have made significant contributions to hydrological studies, particularly the China Institute of Water Resources & Hydropower Research (322), the Yangtze River Water Resources Protection Bureau (118), and the Nanjing Hydraulic Research Institute (117).

Fig. 3.

The top 15 Chinese academic institutions in the field of hydrology by publication volume and its category.

We employed regular expression matching to extract the potential usage frequency of various hydrological models across all publications, and Fig. 4 illustrates the relative popularity of each model (i.e., the percentage of a model’s usage frequency relative to the total frequency). It is evident that the soil and water assessment tool (SWAT), variable infiltration capacity model (VIC), and Xinanjiang model (XAJ Model) are the three most widely adopted in Chinese hydrological research, with SWAT leading at a remarkable 46.7%, significantly outpacing the others. The VIC model and XAJ model’s popularity reached 15.7% and 11.9% respectively, ranking second and third in China. Additionally, models such as SWMM, SHE, and TOPMODEL also exhibit a notable level of usage.

Fig. 4.

Relative popularity of hydrologic models.

3.2. Popularity of topics and their trends

Fig. 5 illustrates the co-occurrence relationships among the keywords in the publications. These keywords can be categorized into three distinct clusters: hydrological modeling (green cluster), climate change (blue cluster), and other hydrology-related studies (red cluster). The model is the core of hydrological modeling and mainly focuses on precipitation, evapotranspiration, optimization, uncertainty, and other aspects. Climate change has emerged as a focal point in Chinese hydrological studies over the past 24 years, addressing key issues such as precipitation, runoff, temperature, variability, and human activities. Additionally, topics like reservoirs, land use, water quality, nitrogen, and groundwater have also garnered significant attention in the same period of Chinese hydrological research.

Fig. 5.

Co-occurrence networks of publication keywords. Node size in the figure represents the occurrence frequency of a keyword, with larger nodes indicating higher frequencies. Line width represents the co-occurrence frequency of two keywords, with thicker lines indicating higher frequencies of co-occurrence. Keywords have been categorized into different clusters using co-occurrence clustering methods, and distinct clusters are represented by different colors.

Furthermore, the topic modeling method was conducted on Chinese hydrology publications, resulting in the extraction of 24 distinct topics. Fig. 6 presents the relative popularity for each topic, with the rectangular area corresponding to each topic indicating the overall popularity of each topic. Notably, water resources (13.9%), climate change (13.6%), and hydrological modeling (10.8%) emerged as the most dominant topics, each with a relative popularity exceeding 10%. Other topics, such as discharge (8.7%), flood (6.8%), and precipitation (5.7%), while slightly less prominent, still surpassed 5%. Additionally, a clear disparity in the distribution of research topics is evident, with the top 7 topics accounting for over 60% of the total popularity.

Fig. 6.

Relative popularity for various topics based on the DTM method. The popularity of each topic is determined by the sum of its popularity across all time periods. To facilitate comparison, the popularity of each topic is normalized into relative popularity by dividing it by the total popularity of all topics. The relative popularity of each topic is represented by the rectangular area corresponding to each topic.

Fig. 7 illustrates the temporal variation in the popularity of 13 topics exhibiting significant trends (p < 0.01). Notable topics such as climate change, hydrological modeling, glaciers, carbon, urbanization, water level, drought, streamflow, and wetland demonstrate a clear upward trend, with climate change and carbon showing particularly striking increases. Specifically, the relative popularity of these two topics rises by 4% and 2%, respectively, every decade. In contrast, topics like water resource, discharge, flood, and sediment demonstrate significant declines, with the relative popularity of water resource, discharge, and flood decreasing by 4%, 3%, and 3% per decade, respectively. These trends align with previous global-scale shifts in thematic evolution [6].

Fig. 7.

The topics with significant trends in topic annual popularity based on the DTM method. The dashed lines illustrate the annual popularity of each topic based on the DTM method, while the solid lines represent the estimated trend of popularity based on the Theil-Sen estimator.

3.3. Characteristics of basins’ publications

Fig. 8a illustrates the number of publications related to hydrological research across primary basins in China. The Yellow River and Yangtze River basins lead in terms of publication volume, with a total of 1,455 and 862 publications, respectively. The Haihe River basin (342), Northwest basins (347), and Songhua River basin (278) also show considerable publication activity, each exceeding 250 publications. The Southeast basins (227), Huaihe River basin (222), Pearl River basin (210), and Southwest basins (184) exhibit relatively lower publication counts, while the Liaohe River basin has the lowest publication count (50).

Fig. 8.

The overall publication status of each basin. (a) total number of publications in each basin; (b) temporal trends in publications across the basins. Map is edited on Chinese standard map GS (2019) 1682.

Fig. 8b depicts the temporal trends in publication volumes across the basins. Prior to 2003, publication volumes were generally low, but after that year, there was a steady increase, with the Yellow River and Yangtze River basins experiencing the most rapid growth, followed by notable increases in the Northwest and Haihe River basins.

We further analyzed the topics of publications in each basin, and Fig. 9 illustrates the popularity of different topics across the basins. Climate change and hydrological modeling emerge as prominent topics in all the basins, reflecting the significant attention given by Chinese hydrology to these issues. When focusing on individual river basins, the Yangtze River Basin places greater emphasis on discharge and flood, while the Yellow River Basin is more concerned with water resources, precipitation, and soil. Water resources attract considerable attention in the Haihe River basin, Northwestern basin, Huaihe River basin, and Songhua River basin, while flood is notably prioritized in the Southwest River basin and Huaihe River basin.

Fig. 9.

The top 10 popular topics across various basins. The topic popularity is defined as the sum of the popularity of that topic in all years of that basins.

Although the boundaries between these topics are not always clearly defined, the recall rate of basin information analysis, and the complex interactions among various hydrological disciplines, the findings indicate that the focus of hydrological research in China has shifted from the resource hydrology phase (2000–2010) to the ecological hydrology phase (2011–present) [46].

4. Discussion

As hydrology continues to evolve, its knowledge system has grown increasingly complex, making it both necessary and urgent to review its development. Traditionally, senior researchers in the field have provided profound summaries of its progress and offered forward-looking insights into its future. However, it must be acknowledged that conventional reviews inherently reflect qualitative assessments shaped by researchers’ personal expertise and perspectives, introducing a degree of subjectivity [6]. Additionally, the increasing diversification of research topics and the exponential growth of literature have made it increasingly challenging for individual researchers to grasp the field’s development comprehensively [18]. A quantitative review of existing research can help uncover the historical trajectory, progress, and future directions of the discipline [47]. Bibliometric methods facilitate a quantitative understanding of disciplinary advancements by analyzing temporal trends in publications, institutional contributions, and keyword networks. However, these methods fall short of providing deeper insights into disciplinary characteristics and more targeted information. For instance, in hydrology, basins serve as the fundamental unit, and hydrological models are essential research tools. Hydrologists seek to understand shifts in basin attention, emerging research hotspots, and trends in hydrological model usage—questions that traditional bibliometric methods struggle to address effectively [10]. Meta-analysis, which synthesizes findings from existing studies, has been widely applied across hydrology and geosciences to derive systematic conclusions [48,49]. However, it relies on manually extracting conclusions from individual publications, significantly limiting the number of studies that can be analyzed. As a result, meta-analyses typically incorporate only a few hundred to a thousand studies. Furthermore, sample selection and high heterogeneity in research findings can significantly impact the meta-analysis results [50]. Leveraging advanced artificial intelligence techniques, such as LLM and topic modeling, we have achieved automated processing and analysis of hundreds of thousands of publications, dramatically reducing manual workload. This represents a modest yet significant step forward from traditional bibliometric and meta-analysis approaches. The findings of this study—concerning basin attention, the popularity of hydrological models, and topic trends in China—hold considerable value for understanding the development of Chinese hydrology and guiding its future direction.

Although we initially retrieved 289,513 publications through WoS, only 4177 publications were ultimately used to analyze the progress of hydrological research in China. This reveals an obvious discrepancy in numbers, which can be attributed to two primary factors. Firstly, the limitations inherent in the LLM used for parsing research area names and the Google Maps geocoding interface employed to convert research area names to geographic coordinates inevitably constrained our utilization. In addition, to ensure comprehensive coverage of relevant studies, we employed a broad search scope in WoS, including any publications with the subject terms “hydrolo*” or “water cycle”. We prioritized obtaining as many relevant studies as possible, subsequently filtering them by research topic (hydrology) and research area (China), which further reduced the number of available publications. Overall, this difference between the initially retrieved and ultimately utilized number of publications reflects the bottom-up strategy necessitated by current technological limitations. In future research, we will explore more advanced search techniques to retrieve more relevant publications and employ more powerful open-source LLMs, such as DeepSeek R1 and Qwen 2.5, to enhance the accuracy of hydrological information intelligent parsing [19,20]. Moreover, academic papers published in Chinese journals are also highly representative of the progress in hydrological research in China [51,52]. However, due to significant differences between Chinese and English, challenges in data access, and issues of consistency, our study focused solely on English-language journals indexed by WoS, excluding Chinese-language articles published in Chinese journals. This limitation may restrict our deeper understanding of Chinese hydrology. In the future, it can be considered to integrate hydrological research included in China National Knowledge Infrastructure (CNKI) to consider Chinese publications.

The results show that SWAT, VIC, and XAJ models are the three most widely applied hydrological models in China, with relative popularities of 46.7%, 15.7%, and 11.9%, respectively, which is different from previous global studies. Addor & Melsen [53] analyzed the global popularity of seven hydrological models based on 1529 hydrological modeling studies published between 1991 and 2018, and found that VIC, Hydrologiska Byråns Vattenbalansavdelning model (HBV), and TOPography-based hydrologic model (TOPMODEL) were the most widely used, with the variation in model popularity being influenced by regional suitability and researcher preference. However, Addor & Melsen’s analysis focuses on large-scale and lacks analysis of the most popular SWAT. To contextualize Chinese hydrological research within the global framework, we employed regular expression matching to analyze hydrological model usage across 25,827 processed global hydrological publications. Fig. S2 illustrates the relative popularity of hydrological models globally from January 2000 to October 2023. SWAT remains the most extensively applied hydrological model (52.0%), likely owing to its broad applicability and user-friendly interface [54,55]. The VIC model was developed by the University of Washington [56]. It is a distributed hydrological model renowned for its high accuracy in runoff simulation, and it enjoys widespread use in the United States and China [57,58], securing the second spot in popularity in China at 15.6%. Globally, VIC remains the second most commonly used model (9.3%), although its relative popularity in China (15.6%) significantly exceeds that of the world. The XAJ Model, developed by Professor Renjun Zhao at Hohai University based on the Xinanjiang River Basin [59], is particularly suitable for China’s humid regions and China’s semi-humid regions in semi-wet seasons [60], ranking third in popularity in China at 11.8%. Despite its extensive use in China, the XAJ model’s global relative popularity is just 2.9%, ranking ninth. Overall, both in China and globally, the combined relative popularity of the top three hydrological models exceeds 68%, suggesting a potential issue of model homogenization in current hydrological research. Horton et al. [61] highlighted a critical gap in published studies, noting a lack of thorough discussion on whether the selected hydrological models are appropriate for specific research contexts or regions. Meanwhile, Babel et al. [62] emphasized that model diversity is essential for capturing uncertainties associated with hydrological model structures and implementation. The hydrological research in China remains predominantly reliant on models developed in North America and Europe, underscoring the urgent need to develop and promote models tailored to the unique characteristics of different regions and river basins across China [63]. It is important to note that the regular expression matching method can only extract potential model usage from titles and abstracts of studies, failing to capture the full diversity of hydrological models in practice. Future advancements may leverage artificial intelligence to intelligently parse model usage from abstracts and full texts, thereby reducing uncertainties in evaluating model popularity [64]. Fine-tuning existing generic LLMs using hydrology-related knowledge is also expected to enhance the capabilities of LLMs in related fields [65].

The analysis of basin attention reveals that the Yangtze and Yellow River Basins have garnered significantly greater focus than other basins, with the number of studies being 4.3 times and 2.5 times higher than that of the Hai River basin, which ranks third. This disparity is not only attributed to factors such as basin area, runoff volume, population, economic development, and regional characteristics, but is also deeply driven by policy directives and funding priorities. As China’s two largest rivers, the Yangtze and Yellow Rivers span 6300 km and 5500 km, respectively, possessing clear advantages in basin area and runoff volume compared to other river systems. The Yangtze River basin is also one of China’s most critical industrial hubs, contributing 45% of the national gross domestic product (GDP) and housing 43% of the total population in 2017 [66]. Consequently, issues such as water resource assessment and management, flood and drought mitigation, and the ecological and hydrological functions of lakes and wetlands have received substantial attention [67,68]. Meanwhile, the Yellow River, revered as China’s “Mother River” flows through ecologically fragile regions, making soil erosion and sediment regulation, ecological conservation, and sustainable development pressing concerns [69,70]. Given these factors, China has established the “Yangtze River Economic Belt Strategy” and “Ecological Conservation and High-Quality Development of the Yellow River Basin” [71,72]. At the same time, China has also allocated substantial funding for scientific research and launched a series of National Key Research and Development Programs, including “High-Efficient Development and Utilization of Water Resource”, “Comprehensive Management of Water Resources and Environment in Key Basins Such as the Yangtze and Yellow Rivers”, “Global Change and Response”, “Earth System and Global Change” [73,74]. These initiatives have significantly enhanced research efforts and management strategies for these critical basins.

Furthermore, we analyzed the attention levels of global basins from January 2000 to October 2023 based on 25,827 hydrological publications worldwide (Fig. S3). The results show that the Yangtze and Yellow River Basins secured the second and third positions in global basin attention, following only the Mississippi River. This reflects the growing strength of China’s hydrological research and the continued focus on the Yangtze and Yellow River Basins, which is consistent with previous global studies. Miao et al. [10] found that before 2000, global hydrological research primarily concentrated on basins in developed countries, including the Mississippi River, the St. Lawrence River, the Colorado River, and the Rhine River Basins. After 2000, the number of hydrological studies focusing on basins in developing countries, such as the Yangtze, Yellow, Congo, and Ganges River Basins, increased significantly. During the period from 2010 to 2023, the Yangtze and Yellow River Basins ranked first and third globally in basin attention, respectively, highlighting the progress of Chinese hydrological research and the growing emphasis on addressing hydrological challenges [12].

Finally, the results of topic modeling indicate that Chinese hydrology has transitioned from the stage of resource hydrology (2000–2010) to the stage of eco-hydrology (2011–present). This transformation can be attributed to the dual driving forces of social demand and environmental adjustment [46]. Between 2000 and 2010, China was undergoing rapid industrialization and urbanization, leading to increasingly prominent contradictions between water scarcity and supply-demand balance [75]. Consequently, Chinese hydrological research predominantly focused on resource hydrology, emphasizing the development, utilization, and optimization of water resources, such as watershed water balance analysis and the construction of water resource management models [76,77]. Since 2011, with the economic growth rate slowing and the emergence of sustainable development concepts, the nation’s attention to ecological and environmental protection has significantly increased. Policies such as the “Beautiful China” initiative, the implementation of the “River Chief System”, and the promotion of “Water Ecological Civilization” have marked ecological protection as the core task of water resource management [51]. Moreover, with the growing severity of global climate challenges, Chinese hydrology has shifted towards eco-hydrology, prioritizing the coupling of hydrological and ecological processes and reflecting a multidisciplinary integration [12,52]. This transformation has not only enhanced the scientific rigor and sustainability of water resource management but also provided a theoretical foundation and practical direction for addressing future complex hydrological and ecological challenges. In summary, hydrology has made significant strides over the past two decades in water resource assessment and management, understanding hydrological processes, and the response of water cycle components to climate change [51,52]. However, as pointed out by Blöschl et al. [78], numerous unresolved challenges persist in areas such as time variability and change, spatial variability and scaling, extreme event variability, measurements and data, and hydrological interfaces. Addressing these issues requires further exploration by the hydrological community, leveraging the latest research tools and methodologies [79].

5. Conclusion

The advancement of hydrological science in China represents a critical endeavor in understanding water cycle processes and addressing complex water resource management challenges. This study employed an innovative approach integrating artificial intelligence and natural language processing techniques to comprehensively analyze the development of Chinese hydrological research from 2000 to 2023. By systematically examining publication characteristics, institutional contributions, research topics, hydrological models’ popularity, and basin-specific research trends, we provide a quantitative assessment of the field’s evolution. Our findings reveal significant transformations in research focus, methodological approaches, and scientific collaboration, highlighting the transition from traditional resource-oriented hydrology to a more holistic eco-hydrological perspective driven by environmental sustainability and climate change imperatives.

The research identifies several key insights that have profound implications for future hydrological studies in China and globally. The predominant use of models like the soil and water assessment tool (SWAT), variable infiltration capacity model (VIC), and Xinanjiang model (XAJ Model), coupled with the increasing emphasis on climate change, ecological processes, and interdisciplinary research, underscores the dynamic nature of contemporary hydrology. The concentration of research in economically developed regions such as the Yangtze and Yellow River Basins reflects both scientific priorities and socio-economic challenges. Future research should focus on expanding methodological innovations, developing more localized hydrological models, integrating Chinese-language publications, and leveraging advanced artificial intelligence techniques to enhance a comprehensive understanding. By continuing to bridge technological advancements with environmental challenges, Chinese hydrology can play a pivotal role in addressing global water resource management, climate adaptation, and ecological preservation strategies.

CRediT authorship contribution statement

Jinlong Hu: Data curation, Formal analysis, Methodology, Visualization, Writing – original draft. Chiyuan Miao: Methodology, Supervision, Writing – review & editing. Yi Wu: Writing – review & editing. Jiajia Su: Writing – review & editing.

Biographies

Jinlong Hu is a PhD candidate at the Faculty of Geographical Science, Beijing Normal University, China. His research interests include hydrology, remote sensing, and deep learning.

Chiyuan Miao(BRID: 09726.00.52288) is a full professor in the Faculty of Geographical Science, Beijing Normal University, China. His research primarily focuses on climate change and its impacts on hydrological processes. He has published over 100 articles and has been recognized as a Highly Cited Chinese Researcher (2021–2024). His work has been cited 15,154 times, and he holds an h-index of 59 according to Web of Science.