2.1.1. Study area.

We examined the distribution of two species of killifish, native Aphaniops stoliczkanus (Day 1872) and the endemic Aphaniops kruppi (Freyhof, Weissenbacher & Geiger 2017) [10,11] throughout Oman’s wadi systems spanning from southern Dhofar to northern and eastern drainages [10]. Their occurrence spans two distinct eco-regions: The Oman Mountain eco-region with the designation ID 443 stretches over the southeastern Arabian Peninsula including Oman and United Arab Emirates (UAE) and is surrounded by various water bodies and the Rub’ al Khali Desert while the Southwestern Arabian Coast eco-region designated as ID 439 reaches into Oman at the southern edge of the peninsula [58,59]. The distribution pattern demonstrates Oman’s ephemeral and perennial waterways ecological connectivity which preserves aquatic biodiversity in dry regions [37].

Oman experiences regional differences in temperature patterns. The temperature range in Northern Oman varies between 32–48°C during May to September and falls to 26–36°C from October through April. During summer months coastal areas experience high temperatures up to 46°C along with humidity levels above 90%. The interior plains experience severe climate conditions as summer temperatures reach above 50°C followed by winter temperatures which remain below 23°C and above 15°C. The Dhofar region maintains a consistent temperature range of 30–35°C throughout the year while both Dhofar and the highland region experience moderate temperatures year-round [23].

Mountain ranges cover about 15% of Oman’s entire land area and significantly impact both its geographical features and hydrological systems where they stand out as the primary source of fresh renewable water [23]. The Al Hajar Mountains stretch over a 700 km arc from Musandam in the north to Ras Al Hadd in the east in northern Oman [21]. The Jabal Shams stands as its highest summit at 3,075 m above sea level. Whereas, the Dhofar Mountains located in southwestern Oman contain peaks that rise between 1,000 and 2,000 meters above mean sea level [10,23]. The Hajar mountain range contains numerous spring-fed streams called wadis even though it receives very little and irregular rainfall [60]. The rainwater gathers in the mountain’s subterranean aquifers and emerges occasionally as small lakes or creeks [61]. The region sustains multiple fish species within its freshwater environments which consist of wadis, pools, streams, falaj systems, and springs [12].

S1 Fig presents three spatial maps of Oman including (A) a digital elevation model that displays topographical variations throughout the landscape along with (B) the distribution of average annual temperature and (C) patterns of annual precipitation throughout the country.

2.1.2. Species occurrence data.

We collected occurrence records for Aphaniops kruppi (40 points) and Aphaniops stoliczkanus (90 points) from multiple sources for this study. Among these were field surveys conducted earlier in the Oman Mountain ecoregion (ID 443) and the Southwestern Arabian Coast ecoregion (ID 439) [59], publicly accessible datasets from the Global Biodiversity Information Facility (GBIF: https://www.gbif.org/; [62–66]), and recent publications [10–12,27,30,67]. Fig 2 and S1 Table, illustrates the distribution of these records.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Distribution of Aphaniops kruppi and Aphaniops stoliczkanus in Oman: species ranges and wadi flow capacity.

https://doi.org/10.1371/journal.pone.0346581.g002

Prior to analysis, the data was cleaned to ensure accuracy by removing duplicate entries and filtering out geographic outliers outside the study area in ArcGIS [68,69]. To reduce spatial autocorrelation (clustering bias) and balance data quality with sample size, we applied 1 km spatial filtering, retaining only one record per 1 km according to earlier ecological research [37,42,70–72]. This filtering removed sampling biases potentially caused by over-sampling of more accessible areas (i.e., wadis close to roads or settlements) while maintaining sufficient sample sizes for model training (A. kruppi: n = 40; A. stoliczkanus: n = 90). The 1 km filtering distance was chosen to match the spatial resolution of our environmental predictors (~1 km), thereby ensuring that occurrence records and environmental data are spatially aligned [72].

Due to the low number of occurrences of A. kruppi, we decided against using spatial blocking for model evaluation to ensure sufficient data for cross-validation [73,74]. Therefore, we utilized MaxEnt, as it has been demonstrated to be capable of modeling species with limited occurrence data [75,76].

We determined both species’ accessible area (M) by the FEOW eco-region [59] for the ID 443 and 439, which represent the geographical area that is accessible to the dispersal limited native/endemic species within fragmented wadi systems. This method was proposed as a way to define biologically relevant calibration areas with regard to dispersal limitations and biological/physical barriers [77].

2.1.3. Environmental data.

The distribution of freshwater fish is affected by a variety of environmental factors operating at different spatial scales [38,41]. We selected predictors related to climate, topography, and hydrology at a (30 arcsec~1 km) resolution [40] to model suitable habitats for A. kruppi and A. stoliczkanus (Table 1).The bioclimatic data (1979–2013) were sourced from CHELSA V2.1 (https://chelsa-climate.org), a high-resolution (~1 km) global climate dataset. To project future habitat suitability, we used the GFDL-ESM4 (Geophysical Fluid Dynamics Laboratory Earth System Model version 4) Global Climate Model from CMIP6 [78–80] under three shared socioeconomic pathways SSP1–2.6, SSP3–7.0, and SSP5–8.5 [81] across four time periods: 1981–2010, 2011–2040, 2041–2070, and 2071–2100. While multi-model ensemble methods are generally preferred as they provide a better means of accounting for the inter-model variability of different models; employing a single, well-developed general circulation model (GCM), such as GFDL-ESM4, is methodologically sound and provides internally consistent projections for assessing climate change impacts on species distributions [82,83]. Additionally, GFDL-ESM4 demonstrated skill in simulating climate dynamics for the Arabian Peninsula [32]. In Section 2.1.6, we describe a comprehensive three-part uncertainty framework for quantifying and managing the uncertainty associated with using a single GCM.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. List of predictor variables selected for the Aphaniops species distribution model.

https://doi.org/10.1371/journal.pone.0346581.t001

To minimize collinearity, we applied a Pearson Correlation Matrix (threshold: |r| > 0.7) to 19 bioclimatic variables, and the climate moisture index (cmi) – which is derived by subtracting precipitation amounts from potential evapotranspiration–, and the near-surface relative humidity (hurs) [79], thereby removing highly correlated predictors [31,37,50]. We selected pairwise correlation as opposed to Variance Inflation Factor (VIF) for a number of methodological reasons; (1) VIF is typically only applicable to presence-absence data structures and we are working with a presence background framework using MaxEnt, so VIF would be an inappropriate way to calculate variable redundancy in this case [84]; (2) Pairwise correlation has been shown to represent the common practice for selecting variables for inclusion in MaxEnt models prior to model building [85]; and (3), MaxEnt uses a variety of regularization parameters that can help reduce multicollinearity inherent in many MaxEnt models by penalizing very complex combinations of features [86]. This process yielded 11 bioclimatic variables for A. kruppi and 14 bioclimatic variables for A. stoliczkanus (Table 1).

We characterized habitats using nine geophysical variables (Table 1), including a digital elevation model (DEM) obtained from WorldClim 2.1 at 30 second spatial resolution (~1 km) and converted to ascii format so that it would be compatible with MaxEnt, along with derived metrics: slope (SL), topographic wetness index (TWI), flow accumulation (FA), sediment transport index (STI), stream power index (SPI), and terrain ruggedness index (TRI) [31,41], all of which were generated from DEM and processed in ArcGIS 10.8.2 Spatial Analyst hydrology tools [87]. Eco-region data from the WWF-FEOW database [59] helped identify ecological hotspots and assess climate change impacts. Soil properties were obtained from the Harmonized World Soil Database version 2.0 (HWSD) [88], a globally recognized resource for soil and land management studies.

2.1.4. Species distribution modeling with MaxEnt.

We used Maximum Entropy (MaxEnt v3.4.4) [91], a machine-learning algorithm based on the maximum entropy principle, to model current and future suitable habitats for Aphaniops kruppi and A. stoliczkanus. The model predicts species distributions by combining presence records with background environmental variables, constrained to the most probable geographic extent [76]. Moreover, both continuous and categorical variables and their exchanges can be used within this model. In order to prevent over-fitting, Maxent algorithm limits model complexity through regularization [55,92]. Additionally, MaxEnt is capable of creating consistent models even with a small sample size [37,41,76,84,93]. We used MaxEnt default settings, to generate 10,000 random background data points that are spatially limited to the entire defined accessible area (M) where a species can potentially be present and utilize environmental resources available at each location. This is done to prevent biased habitat suitability predictions due to sampling historical un-accessible geographic locations to the species [77,94]. A 10,000-point background sample was sufficient for describing all potential environmental characteristics on M, while remaining computationally efficient [95].

We employed MaxEnt using the default regularization (β = 1), 15 bootstrapped replications of the data, and an auto-feature class to provide optimal performance and avoid over-fitting; we also utilized a logistic function to produce the suitability values (0–1) for each cell. The use of bootstrapping was especially useful for our low sample size (i.e., A. kruppi, n = 40) since it provided a measure of the robustness of the uncertainty associated with the estimation of the distribution of the species [55,76,95–97]; whereas the use of regularization and the application of automatic feature selection were utilized to control the complexity of the model [84,92,98]. We ran the model with 500 iterations per replicate, splitting the occurrence data into training (70%) and testing (30%) subsets using a random test percentage of 30% and a training presence threshold of 10th percentile. Based on established standards for species distribution modeling, we evaluated model performance on independent test data not used during model training [73,99].

Then, we projected habitat suitability under three future periods (2011–2040, 2041–2070, 2071–2100) and compared these to the baseline (1981–2010). Suitability was classified into four categories: absent (0.0–0.25), low suitability (0.25–0.50), medium suitability (0.50–0.75), and high suitability (0.75–1.00). The use of this equal interval approach was made because of its simplicity and interpretability when comparing continuous suitability data for the purposes of species distribution modeling, which has been used in a variety of plant [100], mammal [101], and marine fish [102] studies. Other classifications are possible and have been proposed, such as classification intervals defined by statistical measures of model performance [103]; however, equal interval classification provides consistency and comparability among studies. It also is important to recognize that sites near classification boundaries (for example 0.24–0.26 or 0.49–0.51) should be viewed cautiously, as these areas of classification ambiguity represent zones where relatively minor differences in either environmental characteristics or model parameters can cause a shift in the classification category assigned. Sites located in these areas may require targeted field validation to provide assurance for suitability predictions.

Spatial changes in suitable areas were quantified using ArcGIS 10.8.2. Hence, we created 10 suitability maps featuring the current and future scenarios, and periods for each species.

2.1.5. Model validation.

Model validation was assessed using the Area Under the Curve (AUC) of the Receiver Operating Characteristic (ROC) quantifies overall model discrimination ability, ranging from 0 to 1, where 0.5 indicates random prediction and values approaching 1 denote high accuracy [43,104–106]. The True Skill Statistic (TSS) evaluates the balance between sensitivity and specificity, ranges from −1 to +1, the TSS value of +1 indicates perfect agreement, 0 indicates no better than random predictions, and −1 indicates perfect disagreement with scores ≥0.5 indicating robust performance [107,108].

(1)

(2)

(3)

For threshold-dependent validation, we employed three methods the default threshold (0.5), Maximum Training Sensitivity plus Specificity (MaxSS), and Minimum ROC Distance (MinROCdist), which identifies the threshold closest to the perfect classification point on the ROC curve. For each method, we calculated:

(4)

(5)

(6)

(7)

where TP is true positives, FP is false positives, TN is true negatives, FN is false negatives, and N is the sample size of the test set [108,109].

In addition to the threshold dependent metrics mentioned above, we examined the model’s calibration using an assessment approach consistent with the Continuous Boyce Index (CBI) to evaluate models’ calibration. CBI is based on a correlation between habitat suitability values assigned by a model for each location along a suitability gradient and frequency of occurrence [110–112]. However, we developed a robust version of this assessment, comparing the mean predicted habitat suitability at occurrence locations versus background locations. This method calculates a normalized performance ratio, defined as (mean presence/mean background – 1)/ (mean presence/mean background + 1); this ratio values ranging from −1 to +1. Similar to standard CBI interpretation, ratios greater than zero reflect that modeled habitat suitability values are well correlated with observed species occurrences; higher positive ratios indicate better calibration of the model [110–112]. Our approach retains the ecologically meaningful aspects of standard CBI analysis while being more robust to small sample sizes and specific data distributions in our study. In order to ensure statistical robustness, we reported means and standard deviations across 15 replicates. For this analysis and prevalence consistency checks (flagging deviations >10% from observed prevalence) we used the R packages PresenceAbsence and ecospat.

2.1.6. Uncertainty and extrapolation analysis.

To address uncertainty of projections, we implemented three methods to provide a comprehensive assessment of uncertainty in our model outputs. (1) To understand both the variability of our predictions and their reliability through time, we ran 15 bootstrapped model replicates to establish how well our predictions were stabilized. (2) We conducted a Multivariate Environmental Similarity Surface (MESS) analysis to determine areas where future climate projections represent novel conditions (extrapolation) compared to the current environmental space used for model training [113]. A negative MESS value indicates extrapolation risk, where model projections are less reliable [113,114]. The MESS analysis was completed for all 9 future scenarios using the dismo package within R [115]. We visualized MESS values using a continuous color gradient (red: −100 to blue: + 100) and classified MESS values into three ecological categories based on study-specific thresholds to facilitate ecological interpretation: strong extrapolation (MESS < −10), moderate extrapolation (−10 ≤ MESS < 0), and analog conditions (MESS ≥ 0). (3) Comprehensive Model Evaluation: As detailed in Section 2.1.5, we used independent test data and multiple validation metrics (AUC, TSS, CBI) to ensure robust model performance.

2.2.1. Study area.

We selected twelve freshwater wadis in Oman’s Hajar Mountain ecoregion, representing four areas: Darsait (D), Al Amirat (A), Al Khoud (K) in Muscat Governorate, and Ain Wadhah (AW) in Ad Dakhiliyah Governorate (Fig 3). We categorized these wadis by stream position (upstream, midstream, downstream) and assigned them station codes: Darsait (D1-D3), Al Amirat (A1-A3), Surur/Fanja/Al Khoud (K1-K3), and Ain Wadhah (AW1-AW3). All of these sites originate from the Hajar Mountain eco-region with permanent streams throughout the year. Additionally, all sites are accessible with presence of native freshwater fishes. Eight sites feature sandy substrates mixed with gravel (10–90%), while roads, dams, and recreation areas constitute the most frequent human disturbances. Riparian vegetation density varies across sites, with midstream areas typically supporting the greatest coverage (S2 Table).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Sampling locations and visual documentation of wadis in Northern Oman’s Hajar Mountain ecoregion.

https://doi.org/10.1371/journal.pone.0346581.g003

Recent research reported that A. stoliczkanus and A. kruppi mitochondrial haplotypes exist together in Wadi Fanja and Wadi Surur. These findings indicate substantial genetic linkages and possible introgressive hybridization between Aphaniops populations/species across Northern Oman [26,27]. Hence, Aphaniops spp. is considered the appropriate taxonomic designation for this habitat suitability assessment.

2.2.2. Habitat suitability analysis for Aphaniops spp.

We developed a Habitat Suitability Index (HSI) model for Aphaniops spp. following Brooks’ (1997) four step framework [52,116,117], collecting environmental and hydrological data on a quarterly basis from March 2022 to March 2023 across twelve wadis (A1-A3, AW1-AW3, D1-D3, K1-K3). Our initial step involved determining crucial habitat elements through the measurement of twelve environmental parameters across all locations. These included physical characteristics (depth ranging from 25−84 cm, width from 2.5–11.5 m, and velocity from 0.01–0.4 m/s), water chemistry (temperature between 28–31.6 °C, pH values of 7.9–8.6, electrical conductivity measuring 696.2–2609.5 μS/cm, total dissolved solids of 351.7–1331.2 mg/L, salinity between 0.33–1.33 ppt, dissolved oxygen levels from 5.8–10.4 mg/L, and biochemical oxygen demand ranging from 2.84–4.36 mg/L), water quality (turbidity between 1.1−22 NTU), and substrate quality (soil texture). We analyzed soil texture for particle size distribution using the Cosby pedotransfer function [118], then normalized the results to a 0–1 scale via logarithmic normalization [119] to align with our Gaussian modeling framework. The Aphaniops population abundance assessment involved temporary capture methods using foldable fishing traps (mesh size 3 × 3 mm, consisting of 4 nets). Fish welfare was considered through the process of sampling fish. Once fish were collected, they were placed in holding tanks supplied with oxygen and the fish were counted and identified. After that, fish were released back to the location from which they were captured in a healthy condition. No fish were kept, sacrificed, subjected to experimental manipulation, or even killed through a brief observation process for identification. The number of Aphaniops fish demonstrated considerable variation (979−17,708 individuals) across different wadis, with Wadi Aday (A3) sustaining the largest populations and Wadi Darsait (D3) sustaining the smallest (S3 Table).

Second, we developed habitat suitability curves for each parameter using Gaussian functions, calculating weighted means and standard deviations based on Aphaniops spp. abundance. Habitat suitability was modeled using the equation:

(8)

where S(x) represents the suitability index (0–1) for parameter value x. All parameters were constrained to ecologically plausible ranges to avoid extrapolation beyond observed conditions. This produced suitability scores between 0–1 for each parameter [120]. We then calculated the relative importance of each environmental parameter by comparing mean suitability values and their variability (standard deviation) across all sites.

Third, we utilized the Geometric Mean habitat suitability index instead of an arithmetic mean for parameter combination to obtain more conservative Habitat Suitability Index values since it reduces the overall score to zero whenever one parameter proves unsuitable thereby penalizing habitats with any critical limitations [53,121,122]. We also conducted a comparative assessment of HSI values with and without substrate texture to evaluate the specific contribution of this parameter to overall habitat suitability [123].

Fourth, we validated our model through statistical testing (Shapiro-Wilk), biological relevance assessment (comparing HSI with biodiversity metrics), and independent validation (AUC and TSS). For comprehensive model evaluation, we calculated the Boyce Index and generated a confusion matrix to assess classification performance, including sensitivity, specificity, precision, recall and F1 score metrics [104]. The Boyce Index is a useful presence-only, threshold-independent assessment metric that avoids pseudo-absence data assumptions [111].

Additionally, we used Non-metric Multidimensional Scaling to visualize community composition differences between sites. Through correlation analyses we examined the relationship between HSI values and various biodiversity measures including Shannon diversity, Simpson’s diversity, and evenness to determine our model’s biological significance. Finally, we applied the quantile classification method following Edosa and Erena [124] to classify HSI values into five zones based on data distribution: The quantile classification method divided HSI values into five zones: unsuitable for the bottom 20%, less suitable for 20–40th percentile values, moderately suitable for 40–60th percentile values, suitable for 60–80th percentile values, and highly suitable for the top 20% [124]. This data-driven approach ensures objective habitat quality assessment based on the actual HSI distribution within our study area.

To address potential over-fitting, and quantify uncertainty, we utilized bootstrap resampling (1000 iterations), to create 95% confidence intervals for correlation coefficients and Leave-One-Out Cross-Validation (LOOCV), as a means of evaluating model predictive performance for use with smaller ecological data sets [73,111,125–127].

We processed our data with R (version 4.4.3) packages dplyr, tidyr and magrittr; applied EnvStats and nortest for statistical validation and boot for bootstrap analysis; and used ggplot2, corrplot scales, extrafont, and ggpubr for visualization. ROCR, PresenceAbsence, ecospat, caret, and Metrics were used to develop ecological models and validation metrics. All packages were managed using pacman, and writexl facilitated data export.

2.3.1. Individual protected area analysis.

We analyzed habitat representation in Oman’s 31 protected areas through overlaying of species distribution models for A. kruppi and A. stoliczkanus on their respective protected areas. High-habitat suitability was determined using the 90th percentiles threshold. For each protected area, we calculated the area and percentage of suitable habitat for each species, the wadi density classification (high: > 0.0008; medium: 0.0004–0.0008; low: < 0.0004km/km²), and assigned a conservation priority score. All analyses were completed in R version 4.4.3 utilizing spatial packages (raster, sf, terra, dplyr, and exactextractr) for accurate area-weighted calculations in WGS84 coordinate reference system.

2.3.2. Quantifying conservation gaps and establishing targets.

The gap in protection was evaluated using Oman’s National Environment Strategy (2021–2040), developed by the Environment Authority in alignment with Oman Vision 2040 [128] and the Kunming-Montreal Global Biodiversity Framework targets (CBD 30x30) [129,130]. The national target for developing protected areas in wetlands for the year 2040 is 7.5% of the total habitat (25% of the 30% goal of the IUCN 30%), measured by the indicator “Percentage of beneficiary protected areas (Ramsar sites) of total wetland area”, with interim milestones of 10% by 2025, 15% by 2030, and 20% by 2035. Oman currently has 3 designated Ramsar sites (wetlands of international importance). We applied these national wetland protection targets to freshwater fish habitats by determining the area of protected high-suitability habitat for each species, calculating the rate of protection, and quantifying gaps from both national (7.5%) and international (30%) targets. Protection of wadis was evaluated based on specific density-based targets: 50% for high-density (critical connectivity corridors), 30% for medium-density (landscape connectivity), and 10% for low-density (minimal conservation priority). The output of the analyses was developed for conservation planning purposes using both ArcGIS (10.8.2) and R version 4.4.3 utilizing spatial packages (raster, sf, dplyr).

3.5.1. Model performance evaluation.

Assessment of habitat suitability for Aphaniops spp. exhibits strong performance metrics. The Boyce Index measurement of 0.894 demonstrates significant correlation between habitat suitability predictions and observed species occurrences (S7 Fig). With an AUC value of 0.778 the model shows acceptable discriminatory power while the TSS value of 0.667 validates excellent performance (S7 Fig). The Shapiro-Wilk test demonstrates that habitat suitability data follows normal distribution because W = 0.904 and p = 0.179 signify no significant statistical deviation (S7 Fig). The descriptive statistical analysis reveals an acceptable level of moderate positive skewness at 1.146 and platykurtic distribution with a kurtosis value of 1.103 for parametric analysis. The confusion matrix results show that classification accuracy stands at 83.3% with 10 correctly classified sites out of 12 total sites and includes 4 true positives (TP) and 6 true negatives (TN) while recording 2 false negatives (FN) and no false positives (FP). The confusion matrix analysis showed sensitivity (recall) at 66.7%, specificity at 100%, precision at 100% and F1 value at 80.0%.

3.5.2. Habitat suitability and biodiversity trade-offs in stream ecosystems.

Correlation analyses between Habitat Suitability Index (HSI) and diversity metrics consistently indicated negative correlations across all alpha diversity metrics within the stream ecosystems studied (Table 4 and S7 Table). Bootstrap uncertainty quantification confirmed the robustness of these relationships: Shannon diversity index exhibited a moderate, negative correlation with HSI (r = −0.577, 95% CI: −0.840 to 0.009, p = 0.049), as values varied from 0.678, at HSI of 0.496 in Wadi Darsait 3 (D3), to 0.034 at an HSI of 0.945 in Wadi Aday (A3), representing a 95% decrease across the HSI gradient. Evenness exhibited the strongest correlation (r = −0.589, 95% CI: −0.869 to −0.055, p = 0.044), while Simpson’s index demonstrated a negative trend with the confidence interval slightly crossing zero (r = −0.554, 95% CI: −0.849 to 0.011, p = 0.062) (Fig 6).

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Biodiversity and habitat suitability metrics for Hajar Mountain stream sites. HSI values, threshold classifications, and biodiversity metrics for 12 sites in Oman.

https://doi.org/10.1371/journal.pone.0346581.t004

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Relationships between habitat suitability classes and aquatic biodiversity.

(A) Shannon Diversity Index (r = −0.577, p = 0.049), (B) Simpson’s Diversity Index (r = −0.554, p = 0.062), (C) Evenness Index (r = −0.589, p = 0.044), and (D) NMDS ordination of community composition, colored by suitability class.

https://doi.org/10.1371/journal.pone.0346581.g006

Mean and standard deviation of the central tendency and variability of biodiversity metrics, based on the results of the bootstrap resampling for Shannon (mean ± SD: 0.607 ± 0.410, 95% CI: 0.377–0.829); Simpson (0.320 ± 0.235, 95% CI: 0.188–0.456); and Evenness (0.451 ± 0.296, 95% CI: 0.286–0.607) are presented in S8 Fig and S8 Table. The Leave-One-Out-Cross-Validation (LOOCV) showed moderate ability to transfer the HSI model to new sites (R² = 0.852, RMSE = 0.160, MAE = 0.133), an expected limitation due to the small number of study sites (n = 12) (S9 Fig) [73]. An evaluation of residuals of the model showed a systematic under-prediction bias (mean residual = 0.133) with the largest prediction error for site D3, A2, and K1 (residual = 0.323, 0.224, 0.223, respectively), indicating that these two wadis have unique ecological features not captured when the model is trained on other sites (S10 Fig and S9 Table).

The conclusions from the beta diversity analyses revealed clear spatial separation of stream communities based on their compositional differences (S10 Table). Moreover, the Bray-Curtis dissimilarity matrix [131] calculated the differences explicitly, finding relatively low dissimilarity values (0–0.143) between samples within the same stream grouping and higher dissimilarity values (and often exceeding 0.7) between stream groupings. For example, sites within group D displayed a high degree of similarity with each other (dissimilarity 0–0.143) but significantly dissimilarity to AW sites (dissimilarity 0.714–0.75). Across the AW group, dissimilarity was perfect (dissimilarity = 0), as it was for the K group.

The non-metric multidimensional scaling (NMDS) ordination analysis (Fig 6) revealed clear spatial segregation of stream communities with an excellent stress value of 0.044, indicating minimal distortion and high reliability of the 2D representation with little risk of misinterpretation [132]. AW sites clustered tightly in the upper right quadrant (NMDS 1 values = 0.543 to 1.025), indicating they shared similar community structure despite being different but physically separate sites, possibly indicating strong environmental filtering influences. K sites also clustered tightly in the lower right quadrant, indicating they exhibited similar community structure and respective environmental conditions or management practices (NMDS 1 values 0.370 to 0.661). A sites exhibited the most dispersion (NMDS 1 values −0.727 to 0.061), indicating that their community composition was more variable and could indicate that they are either exposed to increasingly diverse habitat conditions or potentially represent different stages of ecological succession. D sites clustered strongly in the negative NMDS 1 region (−1.096 to −0.866), indicating a level of uniqueness within their community composition and likely within another or different environment. Spatial interpolation (Fig 7) demonstrated an evident inverse geographic relationships between HSI (panel A) and diversity indices (panels B-D) during this study, where the areas of highest HSI (sites A3, D1, and K2) coincided with areas of lowest diversity, while sites A1, AW1, AW3, A2, and AW2 maintained higher diversity despite lower HSI values (Table 4).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 7. Spatial distribution of Habitat Suitability Index and biodiversity metrics using Inverse Distance Weighting (IDW) interpolation.

(A) Habitat Suitability Index, (B) Shannon Diversity Index, (C) Simpson’s Diversity Index, and (D) Evenness Index across the study region, demonstrating inverse geographic relationships between habitat suitability and biodiversity measures.

https://doi.org/10.1371/journal.pone.0346581.g007

3.5.3. Environmental parameter importance.

An analysis of mean suitability values for the environment parameters showed a hierarchy of importance. In total, there were seven tiers of important variables. Dissolved oxygen (DO) had the highest importance with a mean suitability of 0.771 ± 0.308, forming the first tier alone. The second tier included water velocity with a value of 0.729 ± 0.287. Stream width and substrate texture shared equal importance in the third tier, with values of 0.719 ± 0.297 and 0.719 ± 0.247, respectively. Water temperature formed the fourth tier with a value of 0.707 ± 0.354. pH comprised the fifth tier with a value of 0.694 ± 0.345. BOD formed the sixth tier with a value of 0.686 ± 0.309. The final tier included the water chemistry parameters: total dissolved solids (TDS), electrical conductivity (EC), salinity, depth, and turbidity, with values of 0.675 ± 0.343, 0.675 ± 0.343, 0.674 ± 0.342, 0.673 ± 0.286, and 0.672 ± 0.292, respectively (Fig 8 and S11 Table).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 8. Drivers of habitat suitability for Aphaniops spp.: Environmental variable contributions.

https://doi.org/10.1371/journal.pone.0346581.g008

3.5.4. Optimal habitat conditions.

The Optimal habitat parameter ranges for Aphaniops spp. were determined through examination of environmental measurements with Gaussian-derived suitability curves. Aphaniops spp. showed preference for certain ranges for all parameters. The physical habitat preferences were for intermediate depths (28.71–66.34 cm), intermediate stream width (3.63–8.62 m), and low to intermediate water velocity (0.04–0.24 m/s). Water chemistry parameters included optimal ranges found for temperature (29.29–30.78°C), slightly alkaline pH (8.07–8.50), moderate electrical conductivity (1164.46–2342.04 μS/cm), total dissolved solids (586.90–1187.57 mg/L), and salinity (0.57–1.18 ppt). The species preferred well-oxygenated water (DO: 6.42–8.66 mg/L) with moderate biochemical oxygen demand (2.83–3.93 mg/L) and low to intermediate turbidity (1.66–14.9 NTU) (Fig 9 and S12 Table).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 9. Gaussian-fitted habitat suitability curves showing optimal environmental ranges for Aphaniops spp.

https://doi.org/10.1371/journal.pone.0346581.g009

3.5.5. Site-specific habitat suitability.

The Habitat Suitability Index (HSI) analysis conducted across the twelve study sites, using the quantile classification method, illustrated significant spatial variation in habitat quality (Table 4). Three sites (25.0%) fell into the “Highly Suitable” classification: Wadi Aday (A3) with the highest HSI, Wadi Darsait 1 (D1), and Wadi Fanja (K2), with values of 0.945, 0.781, and 0.724, respectively. In addition, these same sites were notably associated with different patterns of species diversity. Two sites (16.7%) were classified as “Suitable”: Wadi Surur (K1) and Ain Wadhah 1 (AW1), with values of 0.635 and 0.587, respectively. Three sites (25.0%) had HSI values between 0.526 and 0.584 and were classified as “Moderately Suitable”: Ain Wadhah 2 (AW2), Wadi Darsait 2 (D2), and Wadi Al Amirat 1 (A1). Furthermore, two sites (16.7%) were classified as “Less Suitable”: Wadi Darsait 3 (D3) and Wadi Al Amirat 2 (A2), with values of 0.490 and 0.496, respectively. Two sites (16.7%) were classified as “Unsuitable”: Ain Wadhah 3 (AW3) and Wadi Al Khoud (K3), with values of 0.443 and 0.433, which had the lowest habitat quality conditions in the study area.

3.5.6. Substrate texture impact assessment.

Analysis showed that HSI values with and without substrate texture varied across sites (Fig 10 and S13 Table). In total, substrate texture improved habitat suitability predictions for nine sites, with the largest improvements to HSI values at Ain Wadhah 3 (AW3; 6.7%), Wadi Al Khoud (K3; 6.1%), and Ain Wadhah 2 (AW2; 5.6%), respectively. The inclusion of texture also resulted in decreased HSI values at three sites, including Wadi Surur (K1), Wadi Fanja (K2), and Wadi Aday (A3), and their values were −17.0%, −1.4%, and −0.8%, respectively. This suggests a complex interaction between substrate characteristics and other environmental variables (Fig 10 and S13 Table).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 10. Site-specific effects of substrate texture parameter inclusion on Habitat Suitability Index values.

https://doi.org/10.1371/journal.pone.0346581.g010

Characterizations of the substrate composition patterns provide insight into the relationships that may exist between substrate texture and habitat suitability. The highest-suitability site, Wadi Aday (A3), with an HSI value of 0.945, had a more balanced substrate composition of 47.45% sand and 48.1% gravel, with optimal amounts of each for Aphaniops spp. The sites with the largest improvements with texture inclusion had varied substrates: Ain Wadhah 3 (AW3) was characterized as gravel-dominant (60.44% gravel, 35.91% sand), Wadi Al Khoud (K3) as sand-dominant (80.62% sand, 12.47% gravel), and Ain Wadhah 2 (AW2) as largely gravel-dominant (90.49% gravel, 8.4% sand). In contrast, Wadi Surur (K1), which had the largest decrease (−17.0%) when substrate texture was included, exhibited a more varied substrate composition of 45.42% gravel, 35.23% sand, 14.26% silt, and 5.08% clay, resulting in a sandy loam texture. This indicates that more mixed substrate types might not be optimal for this specialized arid fish species (S2 Table).

4.1.1. Patterns of biogeography and key environmental drivers.

Habitat suitability maps (Figs 4 and 5) reveal separate distribution patterns of the two Aphaniops species in Oman. A. stoliczkanus occupies a broad range along Oman’s northern coastal areas within the Oman Mountain eco-region. While, the range of A. kruppi remains restricted to dispersed appropriate habitats along southern coastal areas associated with the Southwestern Arabian Coast eco-region.

The principal factors shaping A. kruppi’s restricted distribution pattern are the mean monthly climate moisture index (Cmi_m; 39.9%) and the mean diurnal air temperature range (Bio2; 18.3%). The Dhofar region’s monsoon-dominated environment provides stable annual temperatures between 30–35°C [23] throughout summer rainfall periods which maintain essential moisture conditions vital for survival of this species. The species’ stable presence within its limited range suggests it has specialized adaptations that enable it to thrive in the Dhofar region’s stable climate conditions contrasting with Oman’s interior plains where summer temperatures can reach over 50°C [23].In contrast, the broad distribution of A. stoliczkanus throughout northern Oman reflects its primary reliance on mean monthly precipitation during the coldest quarter (Bio19; 31%) which matches the winter precipitation trends from November to April found in the Al Hajar Mountains region of northern Oman [24,25]. The region receives up to 300 mm of annual rainfall which exceeds the precipitation levels found in interior regions (31 mm) and southern Oman (105 mm) [24]. A. stoliczkanus relies heavily on geomorphological factors such as the sediment transport index (STI; 20.2%) and stream power index (SPI; 13.3%) because these factors likely connect with the many spring-fed wadis and streams originating from the Al Hajar Mountains where rainwater gathers in underground aquifers then surfaces as small lakes or creeks [60].

This climate-specific specialization reflects similar distribution patterns found in Cyprinion muscatense where seasonal precipitation emerged as a vital factor for species distribution under climate change [42], and Garra shamal research which showed essential interactions between precipitation patterns and stream sediment dynamics for defining suitable habitats [31].

4.1.2. Projected impacts of climate change.

Future projections reveal species-specific vulnerabilities. A. stoliczkanus has broader yet unstable habitat patterns, with particular changes seen under SSP3–7.0 where low-suitability habitat grows from 0.02% to 1.04% in 2041–2070 before experiencing unexpected contractions. The medium-suitability habitat experiences erratic shifts with a 0.22% increase in 2041–2070 followed by a dramatic drop to −0.21% in 2071–2100. Later projections for high-suitability habitats show negative trends with reductions to −0.11% under SSP3–7.0 and −0.10% under SSP5–8.5.

For A. kruppi, this species experiences a rising trend in low-suitability habitat from 0.36% to 0.65% under SSP1–2.6 while medium-suitability zones moderately expand between 0.10% and 0.16%. The most dramatic shifts occur under SSP3–7.0, where low-suitability habitat increases reach 1.07% by 2071–2100.

The species’ vulnerability might be connected to its reliance on winter precipitation trends which will become unpredictable due to climate change while the alterations in stream geomorphology might be caused by extreme weather events such as cyclones Gonu (2007), Phet (2010), and recent storms that brought extraordinary rainfall to the region [36]. Furthermore, Aphaniops stoliczkanus faces amplified natural vulnerabilities to climate change due to anthropogenic changes in Oman’s freshwater ecosystems. In 2018, Oman established 155 groundwater and surface water dams [144,145] with many being located in northern areas where A. stoliczkanus populations live. New research findings demonstrate how dams disrupt river flow patterns and habitat connections [146] and predicts that dams will intensify warming effects because they block fish from moving to new areas as temperatures rise [147]. Beyond dams, freshwater systems face multiple threats including habitat destruction combined with high water extraction rates and pollution from domestic waste together with agricultural runoff and invasive species [10–12].

4.1.3. Climate novelty and uncertainty assessment.

Our MESS analysis highlighted major differences in climate novelty exposure, a pattern similar to other freshwater species being impacted by climate change [148]. A. kruppi has an extrapolation risk that is much higher than A. stoliczkanus due to A. kruppi being distributed less widely and having a narrower range of environments it can tolerate. Since there are few analog conditions available, we have to be cautious when interpreting the climate projections, because all of the projected future climates represent new conditions relative to those used to train models [149]. However, even though the analog zones are small in number, they do represent possible climate refugia, which could potentially provide better predictability for the results of future habitat suitability projections [150], making it important to prioritize the conservation of these areas.

Additionally, the fact that the extrapolation risk for both species increases as the emission levels increase and the time period increases, indicates an urgent need for conservation actions, especially for A. kruppi, which exhibits near-complete strong extrapolation under the worst-case scenario, and highlights the vulnerable nature of freshwater specialists in the arid Arabian peninsula [30].

This research demonstrates that Aphaniops stoliczkanus and A. kruppi display separate spatial distribution patterns because of their unique adaptations to local climate conditions. Conservation strategies must be tailored to each species’ specific needs: the conservation of A. kruppi relies on the preservation of moisture regimes in southern wadi systems due to monsoon pattern changes while A. stoliczkanus requires maintenance of natural flow regimes and sediment dynamics because of its sensitivity to stream power and sediment transport indices [31]. Conservation plans must focus on the specific hydrogeomorphological requirements of endemic fishes to protect them against climate change and human-induced habitat changes. These unique conservation needs of these species demonstrate why targeted strategies are crucial instead of generalized methods in the Arabian Peninsula due to growing aridity affecting aquatic biodiversity in temporary and permanent water channels [30,42].

4.2.1. Model performance and HSI distribution.

The habitat suitability model developed for Aphaniops spp. demonstrated Excellent predictive performance, as detailed in Section 3.2.1. Furthermore, promoted model reliability, with these values similar to or greater than other studies modeling fish habitat in freshwater [108,151], consistent with recommended approaches given the data-limited nature of the study [76,152]. Our validation process uncovered some nuance to the results; the HSI model had good descriptive validity for our sampling sites, but LOOCV indicated that the HSI model is fairly robust (R² = 0.852) but with moderately large errors (RMSE = 0.160, MAE = 0.133). We characterized this as “moderate transferability” because it is difficult to generalize an ecological model as complex as ours when trained using such a small sample size (n = 12) [73]. The results of the residual analysis suggested a systematic under-prediction bias (mean residual = 0.133) with the greatest prediction errors at sites D2 (0.343), A2 (0.224), and K1 (0.223), indicating that these wadis may possess unique environmental or habitat characteristics which were not adequately captured by the broader dataset (S9 Table).

The characteristics assigned to Aphaniops spp. as a taxonomic category, and that fact that hybridization has been documented between A. stoliczkanus and A. kruppi in northern Oman wadis with mixed mitochondrial haplotypes suggesting gene flow, further complicates conservation and taxonomy at distinct species levels [153,154]. The hybridization could facilitate adaptive introgression as well as climate resilience in species independently experiencing evolutionary pressures through the movement of adaptive alleles [155,156]. Although our model’s classifications were conservative as evidenced by our classification metric results, the bootstrap validation confirms that our fundamental ecological conclusion that there is an inverse relationship between HSI and biodiversity was still statistically valid; validating the specialization-dominance trade-off regardless of the previously mentioned predictive limitations [157,158]. The observed habitat suitability distribution classes across our study sites reflects an ecological reality that optimal conditions are generally rarer than suboptimal conditions [159,160].

4.2.2. Habitat suitability and biodiversity trade-offs in stream ecosystems.

This study demonstrates a counter-intuitive, inverse relationship between habitat suitability for desert killifish Aphaniops spp. and overall aquatic biodiversity within stream ecosystems (Table 4 and Fig 7). This finding is robustly supported through bootstrap validation of negative correlation of HSI to each of the alpha diversity metrics. This challenges the prevailing belief that more favorable habitat conditions result in increased biodiversity. Our findings reveal that a desert killifish Aphaniops spp. can thrive in highly-specialized habitat conditions that do not support overall high aquatic biodiversity, as expected from several other desert killifish species [161–163]. Instead, our observed model illustrates a typical specialization-dominance trade-off, in which habitat specialists can achieve numerical dominance in harsh environmental conditions, but only at the cost of community diversity [164,165]. In high HSI sites, Aphaniops spp. is likely acting as a specialist, where evolutionary adaption may enable them not only to exploit narrow environmental conditions but also to asymmetrically exclude generalist species that require broader niche breadths [166].

The negative relationships between HSI and biodiversity metrics (Table 4) indicate that as habitats become more suitable for Aphaniops spp., they become less suitable to other species and therefore the community becomes simpler with one dominant specialist [157,158]. This pattern aligns with habitat filtering theory, with harsh environmental conditions acting as strong ecological filters that differentially select for specialization with traits that filter out species that lack those traits [167,168]. Moreover, these trends may be explained by the Intermediate Disturbance Hypothesis [169] and competitive exclusion principles [170]. Furthermore, the pattern observed resonates with broader ecological theory depicting that specialist species are competitively dominant within their preferred habitats and develop conditions conducive to their own dominance due to positive-feedback mechanisms [171–173]. In extreme environments, like desert streams, environmental filtering dominates the prevailing assembly mechanisms and trait convergence for specialization leads to lower functional diversity [174–176]. This is intensified in aquatic desert ecosystems, with physiological convergence forcing even stronger selection where environmental filtering takes the form of temperature tolerances, oxygen availability, and osmotic regulation [176–179].

The spatial patterns illustrated in the NMDS analysis (as detailed in section 4.2.3) add further support for the specialization-dominance framework, it also demonstrates that environmental filtering operates differently across the different landscape contexts of our study area (Fig 6).

The spatial pattern of HSI values and diversity metrics across our study area provide additional evidence supporting our theoretical prediction. This spatial pattern indicates that the specialization-dominance trade-off operates at landscape scales, and specialist dominated communities form discrete areas in the metacommunity [180–182]. Of note, even though AW3 indicated a small improvement in habitat suitability (+6.7%) when substrate texture was considered in the model, the site still remained an unsuitable habitat (HSI 0.443). The flash flood event that changed AW3 flowing stream into isolated, inundated pools (recorded during the study period), created conditions that were some of the least suitable for Aphaniops spp., suggesting that a flood may create suboptimal habitat conditions even with some localized substrate improvements [183,184]. The breakdown in flow is a significant disturbance that irrevocably changes the physical template that shapes fish habitat quality and community structure [185, 186]. Specifically, pool isolation usually decreases habitat connectivity and alters the availability of resources for specialized stream fishes [187,188]. Such disturbances may disrupt the dominance of specialists temporarily, and potentially allow generalist species to colonize, as our results indicate that Aphaniops spp. is able to maintain its competitive superiority even under moderate disturbance, indicating its remarkable adaptation to extreme desert stream conditions [189,190].

4.2.3. Environmental parameter hierarchy.

The hierarchical analyses of environmental parameter presented in section 3.2.3 identified dissolved oxygen as the most influential parameter in determining habitat suitability, a finding consistent with the physiological requirements of cyprinodontiform fishes. The importance of dissolved oxygen for Aphaniops spp. is likely a common occurrence among cyprinodontiformes fishes, and the optimal dissolved oxygen levels of 6.42–8.66 mg/L would be similar to their related species [191]. Most freshwater fish require 5–9 mg/L DO to thrive, with levels below 5 mg/L inducing stress and concentrations under 2 mg/L proving fatal [192].

The preference for moderate to slow current velocities identified in our analysis has important implications for understanding killifish habitat selection. This suggests that the habitat preferred by Aphaniops species probably exists in springs, small streams and the margins of shallow lakes, where current velocities are generally low [27,193,194]. The norm also suggests that killifish prefer to occupy habitats with low or no water movement, which is supported by many species being adapted to lentic or slow habitats of sufficient depth and volume [195]. While stream width and substrate emerged as equally important parameters in our model, quantitative relationships between stream width and killifish population density remain poorly understood, representing a critical research gap requiring field studies combining hydromorphological surveys with population assessments in arid environments. This context-dependent importance of habitat variables aligns with hierarchical habitat selection theory [196,197].

The thermal tolerance range reported in this study demonstrates the evolutionary specialization of killifish in arid environments. Thermal specialization of the killifish is represented in our findings, and additional studies on related Iranian Aphanius species also support findings with more narrow tolerances [27,193,194]. Moreover, this optimal temperature range aligns with thermal preferences documented for related cyprinodontoid species in arid environments [161,198]. Aphaniops spp. have the ability to thrive in alkaline conditions, which requires specialized physiological adaptations. This indicates specialized physiological adaptations and adaptations that facilitate alkaline survival under conditions that would normally disturb embryonic development [199], and necessitate alterations to the gill morphology to manage extreme conditions [200]. Alkaline tolerance indicates evolutionary adaptations to arid systems [201].

Although water quality characteristics (EC, TDS, salinity) were moderately important to our model, they have been shown to be highly flexible by the extreme euryhaline abilities of Arabian Peninsula Killifish. Arabian Peninsula killifish show outstanding euryhaline abilities, with individuals tolerating extreme salinity ranges from freshwater oases (0.7–1.5 ppt) to hypersaline coastal lagoons (41–44 ppt), likely through specialized molecular adaptations and mechanisms [202]. Likewise, the level of tolerance we observed for turbidity will influence the ability of the killifish to forage as they move through different environments. Yet, visual detection and predation efficiency in closely related species decreases significantly above 10 FTU (Formazin Turbidity Units), with 32% reduction in feeding efficiency under algal turbidity compared to clear water [203]. These adaptions allowed them to colonize variable ephemeral pools and streams throughout the Arabian Peninsula [195]. While other water chemistry adaptations vary substantially for endemic species in the eco-sensitive habitats in the Arabian Peninsula; especially considering compounding effects of climate change, groundwater exploitation, agriculture runoff, invasive species, and pollution [12,30,204].

The complex role of substrate texture in our model, with both positive and negative effects at different sites, demonstrates flexible substrate requirements rather than absolute substrate preference. Overall, results suggested that Aphaniops spp. displayed substrate rather than requiring specific compositions. While, the highest-suitability site (A3, HSI: 0.945) noted optimal habitat condition with balanced substrate composition; (47.45% sand, 48.1% gravel). Sites that exhibited improvements with the inclusion of substrate texture exhibited different profiles; AW3 was highly gravel-dominated (60.44% gravel), K3 was highly-sand dominated (80.62% sand), while AW2 was highly gravel-dominated (90.49% gravel). Such substrate flexibility among substrate composition range has been reported among other killifish species [29,205,206].

4.2.4. Community assembly patterns (NMDS Analysis).

The patterns of specialization-dominance articulated in section 4.2.2 and their reference to NMDS suggest how these ecological mechanisms shift spatially across landscape contexts, with environmental filtering emphasizing different degrees of intensity dependent on local constraints. The NMDS analysis effectively evaluated community assembly mechanisms and provided an excellent stress value of 0.044, thus yielding a high level of confidence that the observed clustering was a true ecological difference rather than an analytical artifact [207,208]. This excellent fit allows for meaningful interpretation of spatial segregation patterns as corresponding to underlying community assembly mechanisms [209]. The spatial segregation evidently resulted in the identification of ecological clustering that reflected deterministic community assembly processes that varied by spatial scale [210], with environmental filtering being the most prevalent mechanism structuring communities within different landscape contexts [167]. The Site-specific community assembly patterns were as follows and summarized in (S14 Table):

• Ain Wadhah (AW) sites representing ‘Agricultural-Woodland Interface’: These sites clustered tightly on the upper right quadrant (NMDS1 values 0.543–1.025), demonstrating strong geological and topographic environmental filtering effects associated with mountain-stream interface landscapes [211]. The tight clustering indicates mountain-influenced hydrology and geochemistry where distinctive environmental conditions promote specialized community assemblages. Importantly, site AW3 underwent significant temporal changes over the study, changing from a connected flowing stream to isolated pools due to the effects of flash flooding. This resulted in its ecological characteristics altering while coexisting with its montane environmental condition. Overall, the homogeneous clustering suggests montane environmental filtering creates consistent selective pressures favoring species adapted to oligotrophic habitats, warm but thermally stable waterways, and specific mineral compositions based on the geology of mountains [167]. It is estimated that the mountain environment filtering selects for oligotrophic specialists and species that are adapted to low-conductivity, low-salinity mountain spring waters [212,213]. The unique hydrogeochemical signatures of mountain groundwater systems is due to how the surface water interacts with geological features [214], whereby the chemical evolution of groundwater is the result of the weathering of silicate and carbonate minerals [215]. These characteristics are expected to strongly influence downstream aquatic communities through mountain influences hydrogeochemical processes.
• Al Khoud (K) sites representing a ‘Traditional Wadi Systems’: These sites exhibited a tight clustering in the lower right quadrant (NMDS1 values 0.370–0.661), revealing mixed urban-agricultural-wadi landscape patterns similar across the entirety of the Ad Dakhiliyah region. This pattern indicates similar approximately moderate anthropogenic influence with similar arid lowland geology and wadi geomorphology [216–218]. These sites are representative of a moderate disturbance regime, and show settled-agricultural interface impacts including moderate urban engagement and palm cultivation, delimiting the predictable environmental filtering conditions but buffered by the natural wadi geomorphology [219,220]. This environmental filtering selects communities adapted to approximately moderate anthropogenic stress being shaped by semi-arid wadi hydrology, as well as geological constraints, and is in accordance with urban stream ecology studies showing that moderate levels of anthropogenic influence can lead to stable community compositions that are buffered by natural geomorphological features [221,222].
• Al Amirat (A) sites representing an ‘Urban Gradient Sites’: These sites exhibited the most spread across NMDS1 (−0.727 to 0.061), reflecting areas of abundant urban disturbance variability across Wadi Al Amirat and Wadi Aday systems. The on-site characteristics indicate a reasonable urbanization gradient: A1 = urbanization in a moderate state; A2 = mountain wadi ecosystems, fairly preserved; and A3 = developed heavily. The dispersion indicates a complex interaction between deterministic environmental filtering and stochastic assembly processes [223], whereby urbanization-induced habitat heterogeneity allows the assemblages to be constructed, in part, by the creation of microenvironmental mosaics [224,225]. The pattern of variability represents how urban gradients can result in complex community assembly dynamics in geographic regions [226], and once again, the same urban disturbance variability in levels overrides habitat similarities within the region to create new selective environments across different intensities of disturbance from moderate development to preserved systems to urban stream syndrome conditions [221,227].
• Darsait (D) sites representing a ‘Complete Urban Wadi System’: These sites formed a clearly identifiable grouping in the negative NMDS1 region (−1.096 to −0.866), reflecting their unique position as a complete wadi-to-sea continuum within Wadi Darsait system. These sites represent different watershed positions: D1 (urban headwaters), D2 (middle reaches with sewage treatment), and D3 (coastal discharge with marine influence). The tight clustering indicated that regional-scale environmental filtering and shared watershed identity had created a unique regional pool of species that was substantially different from the remaining wadi systems. This pattern reflects the combined influence of urban stream syndrome effects – characterized by flashier hydrographs, elevated nutrients, altered channel morphology, and reduced biotic richness with tolerant species dominance [220,221,228] alongside longitudinal river continuum dynamics [229,230] and unique anthropogenic influences including disruption of longitudinal connectivity [231,232] and wastewater treatment impacts [228,233] and marine connectivity [234] that create environmental conditions not found in other watershed groups.

4.2.5. Beta diversity and community turnover.

Patterns of beta diversity, measured using Bray-Curtis dissimilarity analysis [131], help to show how habitat suitability-diversity trade-offs occur across the landscape. Generally speaking, the patterns reflect the fundamental area-heterogeneity trade-off; increasing environmental heterogeneity allows suitable conditions for more species, while consequently decreasing the effective area available for individual species [235]. The within-group homogeneity (dissimilarity values of 0–0.143) was starkly different from the between-group heterogeneity (often over 0.7), which indicate there was strong community turnover along environmental gradients [236]. The perfect within-group similarity observed in the AW and K groups (dissimilarity = 0), indicates environmental filtering may function as an important assembly mechanism locally [237]. Alternatively, we found significant dissimilarity between D and AW sites (0.714–0.75), which suggests there was strong turnover between the different watershed systems, and shows that species sorting mechanisms were operating at regional scales [210]. These beta diversity patterns support the NMDS results, indicating that the specialization-dominance tradeoff is an aspect of metacommunities, in which community assemblies are formed based upon the environmental filtering at the landscape scale.

4.3.1. Quantifying the protection gap.

The scale of this mismatch is quantified in Table 5. The amount of high-suitable habitat is extremely small (0.31–1.34%) for both species and represents an extremely small percent of Oman’s goal of protecting 7.5% of all critical wetlands by 2040. Therefore, the total amount of critical habitat to be protected to meet the national goal is 24-folds for A. kruppi and 6-folds for A. stoliczkanus higher than the amounts currently being protected. Furthermore, there is an even larger deficit regarding the wadi corridor needed to maintain the populations of our fish species. As shown in S12 Fig, only 2.6% of the high-density wadis are currently located within the boundaries of PAs when it is estimated that 50% of the wadis need to be included in PAs in order to preserve the necessary connectivity for metapopulations. Therefore, approximately 97.4% of the wadis are unprotected. This represents a significant gap in planning for conservation as the linear nature of the wadis are vital for metapopulation persistence and genetic connectivity in fragmented landscapes [238,239].

4.3.2. Key priority areas for conservation action.

Our results identify specific PAs that represent the current core of protected habitat for each species. These should be targeted for increased management and potential strategic expansion (see Fig 12). For A. kruppi, the Jabal Samhan Reserve (PA 14), Al Wusta Wetland Reserve (PA 9), and Khor Kharfout Archaeological Reserve (PA 25) reserves are most important. For A. stoliczkanus, the Al Buraimi Oasis Nature Reserve (PA 2), Al Serien Nature Reserve (PA 8), and Western Mountain Nature Reserve (PA 31) reserves are most significant. Enhancing the management of these reserves and strategically expanding them will provide the greatest efficiency in advancing national conservation goals.

4.3.3. Policy integration and future challenges.

The findings of this study align with global trends where freshwater ecosystems are consistently underrepresented in protected area networks [5]. The identification of priority areas and the immediate need to protect wadi corridors directly support Oman Vision 2040’s strategic direction of achieving “Effective, Balanced and Resilient Ecosystems” [128]. However, achieving this goal will require confronting compound threats such as climate change altering precipitation regimes upon which these species depend [22,24] and the increasing threats from humans, such as dams, extraction of freshwater, invasive species and agricultural runoff, and pollution threatening unprotected wadis, especially in central Oman [12,19,240].

Therefore, future conservation must include comprehensive, landscape-level strategies. Moreover, the described hybridization between A. stoliczkanus and A. kruppi necessitates a shift from species to landscape-level conservation planning that protects the whole Aphaniops species complex while retaining connectivity to enable adaptive gene flow and hybrid zones as potential sources of evolutionary change [241,242]. Ultimately, conservation for Oman’s endemic fish species will require a complicated and multi-facetted approach that maximizes the potential to increase protections for the critical wadi corridors [30], carefully manage freshwater within current protected areas, and develop buffer zones around key sites [51] to ensure the long-term existence of Oman’s unique freshwater biodiversity.