Advances in Freshwater Fish Habitat Suitability Determination Methods: A Global Perspective

Abstract

Freshwater fish habitat simulation is a vital technology for assessing the state and dynamics of aquatic ecosystems under changing environments. Based on a comprehensive dataset spanning 1991–2024, this study constructs a domain knowledge map by integrating co-citation analysis, keyword burst detection, and social network metrics. The bibliometric results quantitatively identify leading contributors and trace the field’s exponential growth. Complementing this, a critical technical review reveals a significant paradigm shift in modeling methodologies: moving from traditional univariate suitability curves to advanced multivariate and artificial intelligence (AI)-based frameworks. Despite these advancements, our analysis highlights critical gaps in addressing habitat connectivity and broad environmental stressors. To overcome these limitations, we propose a novel framework that integrates landscape pattern indices with circuit theory to quantify habitat patch arrangement and ecological flows. Furthermore, we advocate for future research to explicitly incorporate climate change scenarios (e.g., thermal regime shifts) and geomorphological processes. This study offers both a macroscopic overview of the discipline’s evolution and a roadmap for developing robust, ecosystem-based management tools.

1. Introduction

The concept of “habitat” was first proposed by Grinnellian scholars in the United States in 1917, defined as the spatial environment inhabited by specific organisms—essentially the ecological milieu in which they reside. Fish habitat, in particular, refers to the aquatic realms occupied by fish, encompassing all essential environmental factors required to sustain vital activities, including hydrology, hydrodynamics, topography, substrate, water quality, and biological components. These elements fluctuate under external pressures and biological feedbacks, thereby directly impacting fish growth, reproduction, fattening, overwintering, and other critical life-history functions [1]. Within aquatic ecosystems such as rivers and lakes, fish species exhibit high diversity and extended lifespans; they typically occupy apex positions in food webs and span diverse trophic levels from primary to higher consumers. Consequently, fish are highly responsive to habitat alterations, serving as pivotal bio-indicators for the health status of river and lake ecosystems [2].

Since the middle of the last century, intensive human activities have severely compromised aquatic ecosystems, leading to a progressive contraction of suitable fish habitats. Consequently, habitat research has gained increased prominence. Since the 1970s and 1980s, comprehensive research and empirical applications regarding river habitat suitability have been conducted in countries including the United States, South Africa, and Australia [3,4]. Specifically, since 1974, the U.S. Fish and Wildlife Service (USFWS) has pioneered several habitat-centric assessment frameworks, such as Habitat-Based Environmental Assessments, Habitat Evaluation Procedures (HEP), and Habitat Suitability Indices (HSI) [5]. Simultaneously, the U.S. Environmental Protection Agency (EPA) developed the Rapid Bioassessment Protocol (RBP) to evaluate stream integrity using indigenous algae, fish, and macroinvertebrates [6]. Similarly, since 1991, Austria’s fisheries policy has consistently emphasized ecological conservation to maintain indigenous fish diversity. Notably, Muhar [7] conducted a habitat quality assessment across 52 major Austrian rivers (excluding the Danube), focusing on river morphology, watershed connectivity, and hydrological conditions. In 1994, South Africa’s Department of Forestry, Fisheries, and the Environment implemented the River Health Program, utilizing fish, invertebrates, riparian vegetation, and habitat conditions as ecological indicators, while employing the Habitat Integrity Index to quantify the impacts of major anthropogenic disturbances [8]. Furthermore, the European Union adopted the Water Framework Directive (WFD) in 2000, establishing a unified management framework and legal basis for water resource protection [9]. Concurrent developments in Australia (2000) introduced the River Condition Index and the River Status Survey Method. In 2003, the UK Environment Agency developed a standardized River Habitat Survey (RHS) to assess habitat conditions through in situ assessments of physical characteristics, including channel morphology, flow regime, vegetation structure, land use, pools, riffles, and artificial structures (Table S1) [10,11].

Research on river habitat conservation and evaluation in China has evolved along two primary trajectories. First, studies have scrutinized the critical environmental elements of habitats for rare and endangered species or individual taxa. For instance, Yi [12,13] developed habitat suitability models for the Chinese Sturgeon to simulate and predict habitat availability under various hydrological scenarios. Second, research on fish habitat suitability and protection has emphasized the harmonization of development and conservation within major river systems and their tributaries. In 1995, the State Council Gorges Project Construction Committee and the Institute of Hydrobiology, Chinese Academy of Sciences, launched a comprehensive survey of the Chishui River Basin. Subsequently, in 1999 and 2000, Academician Cao Wenxuan proposed the establishment of the “Chishui River Upper Yangtze River Endemic Fish Nature Reserve” at the CPPCC National Committee. In April 2005, the State Council approved the incorporation of the Chishui River—a pivotal tributary of the upper Yangtze—into the “National Nature Reserve for Rare and Endemic Fishes in the Upper Yangtze,” effectively prohibiting the construction of cascade hydropower stations along its mainstream. Further efforts include the 2007 establishment of a fish sanctuary in the Luosuo River, a major tributary of the lower Lancang River, and a 2013 study on alternative fish habitats for hydropower development on the lower Jinsha River. Significantly, on 1 July 2021, the “Technical Code for Fluvial Aquatic Habitat Protection” (NB/T 10485-2021) was implemented to standardize the design, construction, and operation of river aquatic habitat mitigation and protection measures [14].

Globally, numerous scholars have extensively investigated fish habitats, focusing primarily on three thematic domains: modeling frameworks [15,16,17,18,19,20], the selection of indicator systems [21,22,23], and management strategies [21,24]. To systematically quantify and summarize research progress, this paper retrieved fish habitat-related literature from the Web of Science Core Collection spanning 1991 to 2024. Using bibliometric tools CiteSpace and VOSviewer, a visualized analysis was conducted to objectively elucidate research methodologies, core research clusters, and emerging trends. The principal contribution of this paper lies in integrating bibliometric mapping with the methodological evolution of habitat suitability modeling. This dual-perspective framework provides a novel lens for understanding the field’s developmental trajectory, specifically aiding in the identification of modeling methods that remain under-explored.

2. Methods

2.1. Data Sources

The Web of Science Core Collection was selected as the primary data source, with the search restricted to English-language literature. The applied search string was defined as TS = ((“fish habitat” AND “simulation”) OR (“fish habitat” AND “model”) OR (“fish habitat” AND “evaluate”) OR (“fish suitability”)), spanning the period from 1 January 1991, to 12 May 2024. This query yielded a total of 915 records, comprising original research articles, conference proceedings, conference abstracts, and reviews. To enhance data integrity and minimize systematic errors, the retrieved documents underwent a rigorous screening process (detailed in the “Data Cleaning” section and illustrated in Figure 1a), yielding a final dataset of 882 valid papers.

The chronological publication trend is depicted in Figure 1b,c. To visualize the long-term evolution amidst annual fluctuations, a 5-year moving average (MA) trend line was superimposed. The MA line reveals a sustained upward trajectory, indicating identifying intensifying interest from the scientific community. To quantify this growth objectively and identify its critical evolutionary stages, we performed a segmented regression analysis. The regression model showed a high goodness-of-fit (R² = 0.935), explaining over 94% of the variability in the dataset and corroborating the trend observed in the MA line.

Specifically, the statistical analysis identified 2007 as a significant breakpoint (p < 0.001), dividing the field into two distinct phases:

• Incipient Stage (1991–2007): Prior to the breakpoint, the field experienced moderate growth (linear slope, m = 1.31). Annual publication volumes generally fluctuated between 1 and 24, a period of foundational accumulation.
• Rapid Expansion Stage (2007–2023): Post-2007, the scholarly output witnessed a substantial acceleration. The growth rate shifted upward with a steeper slope (m = 2.29). This accelerated trend culminated in a peak of 66 articles in 2023, reflecting the increasing integration of global eco-hydrology research and habitat suitability modeling.

2.2. Data Cleaning

Raw data retrieved from databases may contain redundant duplicates or thematic inconsistencies if not rigorously screened. Consequently, pre-analytical data cleaning is imperative to preclude quality issues from compromising the results and to ensure the efficacy of the visualized analysis. This study adopts the “DEAN” data cleaning framework, which effectively mitigates the adverse effects of data quality anomalies. The “DEAN” process systematically addresses duplicate entries, erroneous records stemming from retrieval deviations, inconsistent keyword variants (Aliases) characterized by identical semantic meaning but divergent structures, and non-informative low-frequency or confounding keywords (Noise). Following the “DEAN” methodology, the workflow and results of this data cleaning are delineated in Figure 1a and Table S2. Detailed procedural information regarding the research methods is provided in the Supplementary Materials and Table S3, thereby facilitating the reproducibility of this article.

2.3. Data Analysis and Visualization

Data processing and visualization were integrated to systematically examine the research landscape across authorial, geographic, and thematic dimensions. First, to identify core research forces, Price’s Law was applied to determine the distribution of prolific authors, where the minimum publication threshold (m) for “core author” designation was calculated using Equation (1):

$$m=0.749\times \sqrt{{{\displaystyle n}}_{\max }}$$

(1)

where n_max represents the number of papers by the most prolific author in the dataset. Authors with a publication count ≥ m are considered core contributors to the research domain. Subsequently, collaboration network for authors with more than four publications was visualized using an overlay view incorporating a temporal dimension, where the color gradient reflects scholarly activity levels over time. In these bibliometric maps, node size corresponds to publication volume, and connecting lines denote collaborative relationships.

To evaluate global contributions, a spatial analysis of the top 30 countries was performed using VOSviewer 1.6.2 and Scimago Graphica 1.0.53, where node size and edge thickness signify scholarly output and inter-country collaboration strength, respectively. Finally, to identify research hotspots, a keyword co-occurrence network was constructed for 882 documents, filtering 53 keywords with a frequency threshold of ≥20. CiteSpace was further employed to generate keyword clustering timeline views, enabling the characterization of the temporal evolution and structural transitions of the identified research themes.

3. Results and Discussion

3.1. Bibliometric Analysis of the Author

By analyzing the authors of the literature, we can learn about the representative scholars and core research forces in this research field. Goldberg [25] postulated that concerning a specific topic, half of the total publications are contributed by a cohort of highly prolific authors.

Based on the criteria defined in Section 2.3, the threshold for core authors in this study was determined to be m ≈ 3.59 (n_max = 23). Consequently, authors with a minimum of four publications were designated as core authors. A total of 67 core authors collectively contributed 399 papers, accounting for 45.2% of the total corpus. This proportion falls short of the 50% threshold stipulated by Price’s Law, indicating that a highly centralized core author group in freshwater fish habitat research has not yet fully coalesced. This phenomenon is likely attributable to the inherent issues of geographical scale. Research in this field is fundamentally tied to diverse and localized aquatic ecosystems across different regions, which inherently favors a decentralized research structure rather than a globally concentrated core group of scholars.

The results of the author collaboration network (Figure 2a) reveal the absence of a cohesive network structure among prolific authors, suggesting that large-scale global research synergies have yet to form—an observation that aligns with the aforementioned Price’s Law analysis. As shown in Figure 2a, the interaction between core scholars remains fragmented, with most collaborations occurring within small, localized research groups rather than integrated global networks. Figure 2b tabulates highly productive authors with at least 10 publications. Among them, Heinz G. Stefan is the most prominent, with 23 publications spanning 1991 to 2024. His research primarily examines the effects of climate change, water quality, and temperature on habitat simulation [26,27,28,29]. Following him are Yao, Weiwei and Chen, Qiuwen, with 17 and 13 publications, respectively. An examination of their scholarly output indicates that these three scholars predominantly focus on environmental hydrodynamics, fish habitat simulation, and ecohydraulics [29,30,31,32,33], reflecting research expertise largely rooted in water resources engineering.

3.2. Bibliometric Analysis of the Journal

Bibliometric analysis of publications spanning the past thirty-five years reveals that research in this field has been predominantly centered on water resources and ecology, with minimal representation in multidisciplinary journals. Figure 2c ranks the top 10 journals by publication volume. Notably, those with a contribution exceeding 30 articles include River Research and Applications (39 articles), Ecological Modelling (34 articles), and Marine Ecology Progress Series (31 articles).

Among these, Water stands out as an open-access journal, underscoring the pivotal role that the expansion of open-access publishing has played in advancing this domain in recent years. Although scholarly debate persists regarding optimal open-access models, the paradigm that research results should be freely accessible is universally acknowledged. Citation metrics indicate that the Journal of Hydrology, a premier publication in hydrological science, commands the highest impact; it accounts for 18 articles with a mean citation rate of 36.28 per article. This suggests that publications in this journal exhibit superior scholarly quality and have garnered substantial recognition within the water conservation community (Figure 2d). The Journal of Hydrology specializes in original research and comprehensive reviews, focusing on critical hydrological issues through robust multi- and interdisciplinary lenses.

3.3. Bibliometric Analysis of the Country

The spatial distribution of global contributions to freshwater fish habitat research (Figure 3) exhibits significant heterogeneity, with research output concentrated in a few leading nations. To move beyond subjective periodization, this study utilized segmented regression analysis (as verified in Section 2.1, R² = 0.935) to identify 2007 as the objective statistical breakpoint. Consequently, the international research landscape is delineated into three primary phases: the Foundational Stage (1991–2007), the Rapid Expansion Stage (2008–2024) and Entire Study Stage (1991–2024).

As illustrated in Figure 3a, the Foundational Stage (1991–2007) was characterized by the dominance of North American and European nations. During this interval, the United States (104 papers) and Canada (33 papers) emerged as the primary pioneers, followed by the United Kingdom (6 papers), Germany (3 papers), and Norway (2 papers). This early momentum was significantly bolstered by institutional frameworks such as the EU Water Framework Directive (2000) and the UK’s River Habitat Survey (2003). In contrast, contributions from developing nations remained scarce; for instance, China’s output was limited to a single study focused on the Chishui River—a pivotal tributary of the upper Yangtze that was incorporated into a National Nature Reserve in 2005.

The Rapid Expansion Stage (2008–2024) marks a profound shift toward global diversification (Figure 3b). Following the 2007 breakpoint—which coincides with the establishment of the Luosuo River fish sanctuary in China—publication volumes surged across both developed and developing regions. While the United States and Canada sustained their leadership, increasing their totals to 379 and 155 papers, respectively, the network significantly expanded to include emerging contributors such as India (14 papers), Portugal (14 papers), Iran (9 papers), South Africa (8 papers), and Brazil (5 papers). Notably, China experienced an unprecedented trajectory, with its output soaring from a single paper in the foundational phase to 119 in the second (Figure 3b,c). This quantitative breakthrough was underpinned by proactive government mandates for ecological restoration, most notably the 2021 implementation of the “Technical Code for Fluvial Aquatic Habitat Protection” (NB/T 10485-2021).

To further explore the structural characteristics of these contributions, the co-occurrence network of the top 30 productive countries was analyzed (Figure 3d). The network reveals a highly collaborative global research environment, albeit centered around a few primary hubs. The United States serves as the undisputed central node, exhibiting the highest total intensity of cooperation (reaching 87) and maintaining the strongest bilateral ties with Canada, the United Kingdom, and China. A dense intra-regional collaboration cluster is also evident among European nations, including Germany, France, and Spain. Although China has achieved a quantitative breakthrough and acts as a pivotal bridge in the Asian network, its international cooperative links are still expanding to match the density of its North American and European counterparts. Despite this numerical expansion (the third position globally with 121 total publications), a pronounced disparity exists between publication volume and international impact. While the United States maintains the most substantial influence—accounting for 43% of the total corpus and leading in citation impact (32.32 citations per paper)—the mean citation count for Chinese scholars (10.88) remains modest compared to Canada (29.57) (Figure 3e). This indicates that while China has achieved significant numerical expansion, enhancing the international academic recognition and qualitative impact of its contributions remains a critical objective for future research.

3.4. Keyword Analysis

The keyword co-occurrence network (Figure 4a) provides a comprehensive mapping of the thematic essence and research hotspots in freshwater fish habitat science. Based on the established bibliometric criteria, the analysis reveals the core conceptual framework and the primary evolutionary trends of the field. In Figure 4a, the diameter of each node corresponds to the frequency of the keyword, with larger nodes highlighting prominent research hotspots. The edges connecting nodes signify the strength of their thematic association; increased line thickness denotes a higher frequency of co-occurrence within the same literature. Representative terms in the field comprise high-frequency keywords such as “fish habitat,” “habitat model,” “river,” “water temperature,” “climate change,” “suitability index,” “management,” “restoration,” and “impacts.”

Notably, beyond the primary search terms, high-frequency keywords including “management,” “restoration,” and “impact” underscore that a fundamental objective of fish habitat research is the reclamation of natural environments to optimize economic, social, and ecological benefits [13,21,34,35,36]. The prevalence of “water temperature” and “climate change” reflects a burgeoning focus on these drivers [34]. These evolutionary trends are further substantiated by density visualizations and burst keyword analysis (Figure 4b and

Table 1. Analysis of fish habitat keyword emergence intensity from 1991 to 2024.

Keywords	Year	Strength	Begin	End	1991–2024
Instream Flow	1996	4.72	1996	2001	▂▂▂▂▂▃▃▃▃▃▃▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂
Simulation	1995	5.28	2008	2014	▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▃▃▃▃▃▃▃▂▂▂▂▂▂▂▂▂▂
Streams	1995	5.12	2011	2013	▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▃▃▃▂▂▂▂▂▂▂▂▂▂▂
Optimization	2011	4.36	2011	2013	▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▃▃▃▂▂▂▂▂▂▂▂▂▂▂
Water Quality	1999	4.79	2012	2016	▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▃▃▃▃▃▂▂▂▂▂▂▂▂
United States	2012	4.19	2012	2017	▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▃▃▃▃▃▃▂▂▂▂▂▂▂
Management	2000	4.42	2013	2018	▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▃▃▃▃▃▃▂▂▂▂▂▂
Climate Change	1999	5.91	2017	2024	▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▃▃▃▃▃▃▃▃
Impact	2005	4.14	2019	2022	▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▃▃▃▃▂▂
Suitability	2014	7.00	2021	2024	▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▃▃▃▃
Habitat Suitability	2012	7.17	2022	2024	▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▃▃▃
Yangtze River	2018	5.01	2022	2024	▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▃▃▃

Note: The red color of the table indicates years with a significant increase in keywords, while the blue color indicates years with no significant change in keywords.

).

The keyword clustering timeline view (Figure 5) effectively illustrates the temporal evolution and structural transitions of the research themes. The Modularity Q score (0.5366 > 0.3) and Weighted Mean Silhouette S score (0.7829 > 0.7) validate the significance and high consistency of the cluster network. As illustrated in Figure 5, the research landscape has evolved from early foundational studies on basic physical parameters (e.g., #1 dissolved oxygen and water temperature) toward more integrated and predictive domains, such as #0 habitat suitability and #9 climate change. Notably, the persistent activity in clusters #3 (habitat modeling) and #8 (fish habitat) in recent years aligns with the surging interest in ecological restoration and the Yangtze River basin mentioned in the keyword burst analysis (Table 1).

Climate change exerts profound pressures on aquatic ecosystems, specifically jeopardizing the habitat suitability for stenothermic cold-water species. It can trigger shifts in habitat selection behaviors and phenological changes in breeding seasons. Altered environmental regimes, such as elevated water temperatures, may compel species to modify their habitat preferences and reproductive strategies. Under divergent climate scenarios—ranging from RCP2.6 (low emissions) to RCP8.5 (high emissions)—habitat distributions are projected to undergo significant contractions, with stricter warming trajectories leading to a substantial sequestration of suitable habitats for cold-water taxa. These emerging domains represent current scholarly frontiers. Furthermore, there is a discernible shift toward prioritizing ecological outcomes over purely economic development. Against this backdrop, the Yangtze River basin has emerged as a focal point for fish habitat research [35,37].

3.5. The Impact Mechanism of Habitat on Fish

Fish habitats are governed by multifaceted drivers, necessitating the selection of indicators that exert significant influence on habitat conditions to ensure accurate environmental characterization. A fish habitat encompasses the spatial extent required to support comprehensive life-history processes, including spawning grounds, feeding areas, overwintering zones, and migratory corridors that connect distinct developmental stages. Pivotal habitat constituents include factors that facilitate fish growth and reproduction, such as water temperature, topography, flow regimes, water quality, and prey availability. The identification of key indicators is paramount for the precision of habitat modeling. These factors must represent the full spectrum of ecological conditions while remaining empirically measurable. Established methodologies for factor selection include expert judgment and Principal Component Analysis (PCA) [38]. Simulated habitat variables predominantly feature water depth, flow velocity, substrate composition, sediment content, water temperature, dissolved oxygen (DO), and hydrological connectivity. Despite the absence of standardized selection criteria, a meta-analysis of peer-reviewed literature reveals that water depth (21%), flow velocity (19%), and substrate (16%) constitute the most prevalent descriptors. These are followed by water temperature (5%), habitat coverage (5%), and slope (4%), alongside secondary variables such as salinity, DO, riparian proximity, channel width, water-level fluctuations, and conductivity.

Bathymetry exerts a profound impact on the density, abundance, and taxonomic diversity of fish populations. Vertical gradients in water depth typically correlate with declining water temperature and DO concentrations, which subsequently modulate fish growth and metabolic rates, as well as drive significant ontogenetic shifts in habitat utilization across different life stages [39]. Furthermore, stratified light attenuation at varying depths dictates the photosynthetic productivity of aquatic macrophytes, thereby regulating the availability of autochthonous food resources [40]. Lake depth is intrinsically coupled with surface area and serves as a proxy for hydraulic storage capacity. Consequently, fluctuations in depth induce dynamic shifts in habitat structure and nutrient stoichiometry, ultimately shaping the reproductive success and growth trajectories of fish species.

Within freshwater ecosystems, flow velocity constitutes a pivotal environmental driver that dictates both spatial occupancy and behavioral patterns. Reduced flow regimes may induce more frequent movement and intra-specific competition, potentially reallocating energy from somatic growth to locomotory activity [41]. Conversely, excessively high velocities (>2 m/s) can compel fish into high-intensity exercise, triggering physiological stress and swimming fatigue, which consequently inhibits growth [42]. It is imperative to note that responses to hydraulic conditions are highly species-specific; thus, these effects cannot be generalized across an entire community. The relationship between flow rate and growth remains multifaceted, contingent upon the specific ecological requirements of each species. Beyond physiological and behavioral modulations, variations in flow velocity alter habitat availability. Reduced discharge typically diminishes the wetted area, thereby restricting spatial niches for foraging, refuge, and reproduction—a contraction that disproportionately impacts habitat specialists. Nonetheless, under specific geomorphological contexts, lower velocities may engender novel microhabitats that benefit early life-history stages, such as juvenile fish.

Substrate represents a foundational element of fish habitats, exerting a decisive influence on spawning success and community structure [43]. Organic substrates—comprising detritus such as leaves, wood chips, mosses, and macrophyte roots—function as critical food resources and refugia for benthic animals and aquatic biota [44]. The mineral matrix [45] is characterized by distinct grain size, sorting, and embeddedness, which collectively govern sediment heterogeneity and bed roughness. These physicochemical attributes mediate biological interactions including competition and predation, while facilitating the formation of interstitial spaces. These voids serve as essential microhabitats for spawning, egg incubation, juvenile refuge, and invertebrate colonization [46]. Embeddedness—the degree to which coarse clasts (e.g., cobbles, gravels) are infilled by fine sediments—directly modulates the accessibility of these spaces. Excessive embeddedness induces substrate compaction and impairs oxygen exchange, thereby compromising habitat suitability for early life stages. In contrast, well-sorted substrates with low embeddedness preserve interstitial voids, supporting higher biodiversity and recruitment success. Furthermore, substrate characteristics modulate near-bed hydrodynamics, influencing swimming efficiency [47], while particle size distributions impact reproductive indices, including fecundity, fry size, and survival rates [48].

Water temperature serves as a determinative environmental driver for fish, exerting profound impacts on their physiological, biochemical, and life-history trajectories [49]. Thermal fluctuations modulate critical functions—including feeding, respiration, and reproduction—across ontogenetic stages [50]. Temperature regimes, encompassing climate-driven shifts and coupled water-air temperature variations, directly regulate cellular proliferation rates in fish [51]. Furthermore, thermal shifts can alter the phenology of benthic and planktonic communities, thereby indirectly modulating piscine energy acquisition and somatic growth [52].

Water quality constitutes another fundamental dimension of fish habitats, as piscine vital activities are intrinsically linked to specific physicochemical regimes. Research characterizing habitat water quality primarily scrutinizes dissolved oxygen (DO), nutrients, heavy metals, and anthropogenic pollutants [53]. DO concentrations are paramount for fish survival and metabolic homeostasis [54], governed by a complex interplay of water depth, flow dynamics, irradiance, temperature, and biological demand [55]. Fluctuations in ambient DO directly modulate blood oxygen saturation levels, subsequently impacting growth, fecundity, and other fitness-related traits. Moreover, DO is synergistically coupled with other abiotic and biotic factors, thereby exerting cascading influences on fish health and behavioral ecology.

Hydrological connectivity is a quintessential factor in fish habitat integrity, underpinning essential processes such as migratory movements, ichthyoplankton dispersal, spatial expansion, and foraging [56]. While extensive research has quantified habitat extent, evaluations typically rely on habitat suitability indices (HSI) or qualitative categorization (e.g., unsuitable, suboptimal, and optimal) rather than crude volumetric metrics. However, spatial configuration and inter-patch connectivity remain under-explored. Relying exclusively on habitat area—while ignoring quality or functional connectivity—may precipitate erroneous assessments and misguided management interventions. Notwithstanding the significance of total habitat area, spatially fragmented and isolated patches can obstruct the flux of matter and energy, hindering movement and recruitment, which may ultimately trigger population senescence [57].

The conceptual framework and the interrelationships between these environmental factors are visually summarized in the accompanying Graphical Abstract (provided as a standalone visual summary). Fish and associated taxa constitute an integrated ecosystem where interspecific competition and mutualism are vital for somatic development. Connectivity across diverse habitat types is essential for the survival and persistence of aquatic biota. While emerging studies have begun to incorporate habitat connectivity, a robust quantitative evaluation framework remains elusive. In landscape ecology, habitat assessment typically synthesizes factors such as patch area, spatial configuration, and fragmentation to evaluate landscape architecture. By leveraging the stochastic properties of electron random walks in circuit theory and accounting for the inherent stochasticity of ecological flows, species can be modeled as electrons and landscape patches as resistive nodes, enabling the quantification of inter-patch connectivity. Lower resistance and higher current density signify enhanced connectivity. We propose an integrated approach using landscape pattern indices to analyze the spatial arrangement of habitat patches, coupled with circuit theory for connectivity assessment. Initially, the aforementioned methods are employed to identify fish habitats and derive both quantity (area) and quality (spatial distribution and connectivity) metrics. Subsequently, the Genetic Algorithm–Analytic Hierarchy Process (GA-AHP) is applied to assign relative weights to these dimensions. Finally, a weighted composite index (ranging from 0 to 1) is calculated, where elevated values denote superior habitat suitability (Figure 6).

4. Methodological Evolution and Synthesis of Fish Habitat Modeling

While the bibliometric analysis in Section 3 provides a macro-scale mapping of the key contributors and general themes in freshwater fish habitat research, it also identifies ‘habitat’ as the foundational concept across all thematic clusters. To complement these quantitative findings, this section provides a qualitative synthesis of the methodological evolution that serves as the technical cornerstone of the field. By transitioning from the ‘who and where’ of the research landscape to the ‘how’ of scientific execution, we evaluate the progression of modeling paradigms—from early univariate approaches to modern AI-driven frameworks—thereby offering a comprehensive discussion on the state of the art in habitat suitability assessment.

The primary objective of fish habitat research is to construct robust modeling frameworks and establish rigorous methodologies for habitat suitability assessment. A comprehensive synthesis of diverse modeling studies reveals that methodological approaches can be systematically categorized into five distinct paradigms: (1) univariate suitability curves; (2) conditional habitat selection criteria (e.g., the MesoHABSIM framework); (3) fuzzy logic-based inference models; (4) multivariate statistical models; and (5) machine learning-driven evaluation approaches. A detailed summary of the advancements in habitat suitability research is tabulated in Table S4.

4.1. Univariate Suitability Curves

The Habitat Suitability Index (HSI), established by the USFWS, remains the primary univariate framework for habitat quantification [58]. HSI models quantify species-specific requirements by mapping environmental variables to normalized indices ranging from 0 (unsuitable) to 1 (optimal). These relationships are typically visualized as curves plotting variable values (abscissa) against suitability indices (ordinate). This architecture underpins over 90% of extant fish habitat models, including the widely adopted PHABSIM [59,60,61]. Figure 6a depicts a representative curve used in these applications.

Functionally, HSI models evaluate individual environmental drivers independently before aggregating results via mathematical functions. While offering high computational efficiency and interpretability, the underlying assumption of variable independence often overlooks synergistic or antagonistic interactions. Consequently, such simplifications can constrain the ecological realism of model outputs.

Despite advanced developments, univariate curves remain predominant in practice. To address their limitations, several alternative paradigms have emerged: (1) fuzzy logic models for stochastic uncertainty; (2) multivariate statistical or probabilistic frameworks for variable interdependencies; and (3) conditional habitat selection paradigms, such as MesoHABSIM. Unlike univariate methods, conditional approaches evaluate variables relative to their environmental context (e.g., depth suitability as a function of specific velocity or substrate), thereby enhancing ecological fidelity within heterogeneous lotic systems.

4.2. Conditional Habitat Selection Criteria

MesoHABSIM, developed by Professor Piotr Parasiewicz, addresses the limitations of micro-scale models like PHABSIM in large-scale river ecosystems. By partitioning fluvial systems into discrete hydromorphological units (HMUs)—such as pools, riffles, and runs—it integrates topography, land cover, and hydrology to evaluate habitat suitability. This meso-scale approach enables management across expansive river systems by navigating complex environments defined by high discharge variability and multi-species requirements. The framework simulates biological responses under divergent flow scenarios, providing rigorous, evidence-based recommendations for flow optimization [62,63].

Renowned for its scalability, MesoHABSIM supports hierarchical, multi-species assessments. While early versions utilized logistic regression for superior accuracy over univariate models, contemporary implementations employ empirically validated conditional habitat selection criteria. Despite its computational complexity, the framework’s methodological versatility makes it a robust instrument for managing intricate hydrological regimes. A synthesis of its technical attributes is tabulated in Table S4 [64].

In practice, MesoHABSIM has seen extensive global deployment, notably in the Quinebaug River basin (USA) to support restoration under fluctuating flows. The systematic workflow involves assimilating multi-source datasets to delineate HMUs, followed by a rigorous quality assessment for each unit. Ultimately, geospatial suitability maps are generated to guide flow optimization, a pipeline successfully replicated across numerous watersheds [63].

4.3. Fuzzy Logic

Fuzzy logic [65] provides a robust mathematical framework for modeling complex systems defined by uncertainty and imprecision. Unlike deterministic models, it allows variables to hold partial membership in multiple categories simultaneously (0 to 1), making it ideal for fragmented or stochastic environmental datasets [66].

In these systems, environmental drivers like flow velocity are represented by linguistic variables (e.g., “low,” “high”). Heuristic “if-then” rules link these to habitat suitability [67]. The workflow involves five steps: (1) defining fuzzy sets and rules; (2) input fuzzification; (3) rule evaluation; (4) output aggregation; and (5) defuzzification via the centroid method to derive a final index (Figure 6b) [68].

By integrating empirical data with expert knowledge, fuzzy logic enhances ecological fidelity, as demonstrated by the CASiMiR model [20,69,70]. However, computational overhead escalates exponentially with input dimensionality. Furthermore, fuzzy logic struggles to represent abrupt ecological thresholds or non-linear state shifts. For instance, it may depict rapid reconfigurations—such as sudden temperature shocks—as gradual transitions, potentially compromising accuracy in dynamic environments. To address this, Niu [71] explored improving model responsiveness to such environmental shocks.

4.4. Multivariate Statistical Frameworks

Traditional HSI and fuzzy logic methodologies often struggle to account for synergistic interactions between habitat drivers, thereby oversimplifying ecosystem complexities. For example, rheophilic fish requiring large substrate as velocity refugia may shift preferences toward lentic environments if such shelters are absent. To address these non-linear dependencies, multivariate frameworks—including Multiple Linear Regression (MLR) [72,73], Principal Component Analysis (PCA) [74,75], Logistic Regression (LR) [76,77], Generalized Linear Models (GLMs) [78,79], Generalized Additive Models (GAMs) [80,81], and Maximum Entropy models (MEMs/MaxEnt) [82,83]—integrate inter-variable dependencies to accurately elucidate the compounded influence on aquatic biota.

While MLR standardly characterizes linear associations, its utility is constrained by multicollinearity [84]. Consequently, PCA is implemented for dimensionality reduction to bolster model stability [85], whereas LR provides superior statistical rigor for presence–absence modeling, as demonstrated in winter flounder distribution studies [86]. Within the MesoHABSIM evolutionary trajectory, coupling LR with the Akaike Information Criterion (AIC) has significantly enhanced ecological inference [87,88]. Furthermore, Random Forest (RF) methods excel in managing high-order interactions and non-linear, high-dimensional spectral data [89,90], marking a shifting preference toward LR and RF for complex classification tasks.

For responses deviating from Gaussian or monotonic patterns, GLMs and GAMs offer flexible architectures and smoothing splines to model intricate ecological nexuses [91]. Similarly, MaxEnt has emerged as an indispensable niche modeling tool [92,93], performing reliably even under data-poor conditions to identify drivers of habitat loss, such as fragmentation and altered hydrological regimes [94].

4.5. Machine Learning-Based Evaluation Methods

Fish–habitat relationships involve multifaceted, non-linear interactions. Within the AI landscape, machine learning (ML) paradigms—particularly Artificial Neural Networks (ANNs)—excel at quantifying suitability by elucidating multiscale dependencies without a priori data assumptions [95]. Various architectures have been implemented, including Fuzzy Neural Networks (FNNs) [96], Fuzzy Habitat Preference Models (FHPMs) [95], Preference Leveling models (PPLs), Multivariate Adaptive Regression Splines (MARS) [97], Discriminant Factor Analysis (DFA) [98], Species Distribution Models (SDMs) [99], and Regression Trees (RTs) [100]. Among these, multilayer perceptrons (MLPs) and feedforward backpropagation networks consistently outperform conventional regression in modeling stochastic ecological processes.

Recently, hierarchical ANN variants (e.g., BPNNs, DNNs) have proliferated. Empirical studies confirm their efficacy: Muñoz [101] implemented an ensemble MLP architecture to assess the ecological niche of Barbus haasi, revealing that integrating biotic interactions significantly augmented model precision. Akkan [102] employed an MLP to predict fish condition factors, achieving remarkable performance across training and validation phases (R² > 0.99). Li [103] synergized HTI acoustic telemetry with BPNNs to model grass carp migratory trajectories, yielding spatially stable results. Furthermore, Brosse [104] corroborated that ANNs consistently exceed multiple linear regression in predicting ichthyofaunal distribution and abundance. Concurrently, deep learning models—specifically Convolutional Neural Networks (CNNs)—have gained traction for vision-based habitat monitoring, excelling in automated classification, detection, and semantic segmentation [105].

Despite high fidelity, ML models often face limited interpretability (“black-box” issues) and constrained generalizability [106]. To address this, hybrid frameworks integrating ML with fuzzy logic or statistical theory are proposed to enhance transparency. As the field shifts toward Explainable AI (XAI) and high-resolution datasets, AI-driven Habitat Suitability Criteria (HSC) offer more objective, data-driven alternatives to traditional heuristic curves, supporting real-time auditing and improved transferability. Complementing these data-driven paradigms, mechanistic frameworks—such as bioenergetic and agent-based models (ABMs)—explicitly simulate ecological processes [107]. These approaches effectively bridge the gap between empirical and process-based methods, as summarized in Table S4.

4.6. Scale-Consistent Criteria and Model Verification

Habitat modeling frameworks operate across a hierarchical continuum of spatial scales, ranging from microhabitats (e.g., water depth, velocity, and substrate composition) to mesohabitats (e.g., fluvial units such as pools and riffles) and macrohabitats (e.g., watershed-scale fragmentation and river connectivity) [108]. Each tier captures discrete ecological processes; consequently, the applied habitat suitability criteria (HSC) must be meticulously aligned with the corresponding scale to ensure ecological relevance. For instance, expert-elicited flow velocity preference curves are optimal for microhabitat models, whereas macrohabitat assessments necessitate landscape-level metrics such as barrier density or hydrological longitudinal connectivity [109].

Notably, suitability parameters lack universal transferability across scales. Projecting fine-scale preferences onto macro-scale models can precipitate scale mismatches, thereby compromising the integrity of ecological inferences. Recent breakthroughs in artificial intelligence (AI) facilitate the derivation of scale-adjusted suitability functions from empirical datasets, augmenting model adaptability while preserving mechanistic interpretability.

Equally imperative is the rigorous validation of models using robust statistical protocols [110,111]. Model benchmarking typically entails systematic data partitioning—including training/validation splits, k-fold cross-validation, or nested resampling—to mitigate overfitting [112]. Standardized performance metrics, such as the coefficient of determination (R²), root mean square error (RMSE) [113], area under the receiver operating characteristic curve (AUC), and the true skill statistic (TSS), are employed to quantify predictive precision and generalization robustness [114]. For example, Muñoz [101] and Akkan [102] implemented these protocols to validate ANN-driven suitability predictions, underscoring the necessity of objective performance audits in identifying optimal modeling criteria.

At the macrohabitat echelon, Parasiewicz [109] elucidated how machine learning can be synergized with conceptual models to quantify systemic habitat alterations. In a pan-European meta-analysis, the authors scrutinized over 1.2 million anthropogenic barriers, estimating that exceeding 200,000 km of historically free-flowing habitat has been fragmented by impoundments. They correlated structural attributes (e.g., barrier typology and density) with functional impairments, including loss of connectivity and taxonomical community shifts. These results highlight the necessity of embedding macro-scale indicators into habitat suitability frameworks, particularly for regional-scale conservation prioritization.

5. Conclusions

This study provides a systematic synthesis of freshwater fish habitat research (1991–2024), integrating bibliometric mapping with a critical review of methodological evolution. Our analysis identifies a thematic shift from foundational physical parameters toward complex predictive modeling and climate-resilience strategies. The principal contribution of this work lies in elucidating how the field has transitioned through five major paradigms: from univariate suitability curves to advanced AI-driven and hybrid frameworks.

The methodological strengths of current habitat simulations reside in their capacity to integrate multi-domain datasets (hydraulics, water quality, and ecology) to forecast habitat shifts under divergent flow regimes. However, critical limitations remain regarding the explicit integration of fluvial geomorphological processes and the high data intensity required for precise HSI-based architectures. Addressing the scarcity of high-fidelity biological data and the “black-box” nature of machine learning is essential for future model transferability.

Moving forward, habitat modeling must evolve into interdisciplinary, scale-consistent frameworks that incorporate climate stressors and anthropogenic disturbances. The convergence of aquatic and terrestrial modeling paradigms, supported by Explainable AI (XAI) and mechanistic models, will provide a more rigorous scientific foundation for basin-scale conservation governance and informed water resource management.