Spatiotemporal variation in habitat quality and quantity impacts individual fish energetics by influencing growth, energy storage, and survival, ultimately shaping population dynamics. However, few studies have explicitly linked individual energetic condition to variation in habitat across both space and time. We investigated the influence of freshwater habitat characteristics and season on both physiological (e.g., percent lipid, energy density) and morphological (e.g., relative condition factor) metrics of juvenile coho salmon energetic condition in interior British Columbia, Canada. Physiological metrics responded to spatial and temporal variation in habitat characteristics. Among sites, a higher percent lipid was associated with lower water temperature and higher energy density, and elevated stream nutrient concentrations. Across seasons, energetic condition declined over summer (July-September) and again over winter (September-April). Post-winter percent lipid and energy density values converged on proposed lower thresholds for survival. Collectively, this research shows the potential for using physiological energetic condition metrics as indicators of spatial and temporal variation in habitat quality. Quantifying how habitat changes affect juvenile salmon energetic condition could improve the evaluation of different habitat protection and restoration actions.

Spatial Model Data:

This dataset (Cumulative_Model.csv) contains the physiological, morphological, and habitat data used to evaluate spatial drivers of energetic condition in juvenile coho salmon (Oncorhynchus kisutch) across multiple sites in the North Thompson watershed (2019–2023). The data support analyses examining how local habitat characteristics influence percent lipid, protein, energy density, and relative body condition (Kn) in Age-0+ and Age-1+ juveniles. The dataset includes raw physiological measurements, derived condition metrics, site-level habitat covariates, and model outputs used in hierarchical spatial analyses.

Physiological measurements represent the whole-body composition of juvenile salmon and include lipid content, moisture, protein, carbon content, and derived energy density. Lipid content was quantified using a modified chloroform–methanol extraction suitable for small-bodied juveniles (0.76–14.88 g). Each individual was homogenized, and duplicate ~0.20 g subsamples were extracted. Whole-body lipid mass and percent lipid were calculated from extraction ratios scaled to total wet mass. Replicates with high variability (CV > 20%) were re-run, and averaged values were retained.

Moisture, protein, and carbon content were derived from duplicate ~0.30 g subsamples dehydrated at 80°C for 24 hours and combusted to ash at 500°C. Percent protein was estimated following Trudel et al. (2005) as the proportion of tissue not accounted for by water, lipid, and carbon. Total energy density (kJ g⁻¹ wet weight) was calculated using standard lipid and protein caloric conversion factors specific to coho salmon (Brett & Groves 1979; Trudel et al. 2005).

The dataset includes fork length and body mass measurements for each individual. These measurements were used to derive Le Cren’s relative condition factor (Kn), a mass-independent index of body condition that compares observed mass to expected mass for a given length. Expected mass was generated from population-specific allometric parameters (a, b) estimated from >5,000 individuals collected during long-term mark–recapture surveys (2019–2023). Kn > 1 indicates better-than-average condition for a given length, while Kn < 1 indicates poorer condition. Fork length measured in the field was used to avoid fin-damage issues affecting lab measurements.

Each fish is associated with site-level environmental variables used to explain spatial variation in condition. Variables include accumulated thermal units (ATUs), conspecific density (fish/m²), large wood counts, pool–riffle ratio, elevation, total phosphorus, and stream gradient. Juvenile coho salmon densities in tributary sites were estimated using a hierarchical Bayesian closed mark-recapture model (HBM) developed by Korman et al. (2016) that was based on Wyatt (2002). All covariates were standardized (mean-centered and scaled by one standard deviation). A Pearson correlation matrix was used to exclude highly collinear variables (|r| < 0.80 retained).

The dataset was fed into hierarchical Gaussian (energy density and K condition) and beta regression models (for percent lipid and protein) describing condition metrics as a function of individual traits (mass, age, and their interaction) and site-level habitat features. Models were fitted using R (2023.12.1) and RStudio.

The general model structure includes individual-level fixed effects (mass, age, mass × age), site-level random intercepts, allowing individuals from the same site to share similarities (e.g., genetic, environmental), and site-level predictors: environmental variables influencing the site-level intercept. 

Model selection was based on AICc across all subsets of habitat covariates (global model dredged). When no single best model was identified (AICc weight < 0.9), model-averaged parameter estimates and unconditional confidence intervals were generated using the natural-average method. Relative variable importance was calculated from summed AICc weights for each variable across the model set. Predicted condition values and model performance metrics, including RMSPE (%), were calculated for each condition metric.

Across condition metrics, fish mass was consistently the strongest predictor of energetic condition—4× stronger than any habitat variable for percent lipid, and 6× stronger for both Kn and energy density. Age and mass showed clear interactive effects, with age-1+ fish approximately 25% heavier and exhibiting different allometric scaling relationships than age-0+. Habitat features exhibited more modest effects, and conspecific density did not significantly influence condition.

Seasonal Data

This dataset (Temporal_analysis_data.csv) includes the information required to evaluate seasonal changes in juvenile coho salmon energetic condition for Age-0+ and Age-1+ fish. Seasonal analyses test hypotheses about how conditions change between midsummer (July–September) and the following spring (April), and how these seasonal patterns differ among sites.

For each site, the dataset contains measurements of mass, percent lipid, percent protein, relative condition factor (Kn), and energy density (ED) collected in July, August, September, and April across multiple years. Each record includes the month of capture, allowing month-specific condition estimates and comparisons.

To quantify seasonal trends in condition, a separate regression model was built for each condition metric at each site. Mass, month, and their interaction were included as predictors to account for allometric changes in condition and the possibility that the mass–condition relationship differs across months.

Percent lipid and percent protein, which are proportional measures, were modeled using beta regression with a logit link. Energy density and Kn, which are continuous metrics, were modeled using Gaussian linear regression. Mass was centered prior to analysis to improve the interpretability of month-specific effects.

Months were treated as a categorical variable (July, August, September, April), enabling comparison of both the average condition in each month and the month-specific relationship between condition and mass. For each site and condition metric, the dataset includes: 

• Month-specific mean condition for an average-mass individual (calculated from the model intercept and month coefficients).
•  Month-specific allometric slopes represent strongly condition scales with body mass in each month.
• Interaction effects identifying whether the mass–condition relationship changes seasonally.

To assess statistical differences among months, pairwise comparisons were performed with Bonferroni correction to control family-wise error when testing multiple month combinations.