Catch-and-release (C&R) is widely promoted to reduce harvest mortality and support sustainable recreational fisheries. However, understanding its ecological, social, and management implications remains limited by the lack of large-scale, long-term datasets revealing whether C&R practices differ among species and change over time. We analysed over one million caught fish (2011–2024) across 39 species and 1 286 Swedish inland fisheries, reported via the online licence platform iFiske AB. The prevalence of C&R increased from ~53% to 71% during 2011-2024, though trajectories and magnitudes differed among species. Size-specific analyses showed that longer individuals were released less frequently than shorter conspecifics for four of the six main target species (northern pike, European perch, zander, European grayling, brown trout, and Arctic char). Our findings indicate that recreational fisheries are shifting towards lower exploitation rates, but also highlights a size-dependent concern regarding the harvest of long individuals that are critical for recruitment and ecosystem functioning, are harvested more frequently. These insights are essential for assessing the sustainability of recreational fishing and for guiding management strategies that balance harvest opportunities with long-term population viability, while emphasising angler education to ensure responsible release practices.

Through a research collaboration with iFiske, we accessed a dataset of 593 842 anonymised catch report entries from 2011–2024, including both public and non-public entries. Each report contained information on the date of angling, species captured, fish sizes, and whether fish were released. To ensure consistency, we first excluded reports involving non-angling methods (e.g. passive gear), representing approximately 1.4% of the total. We then excluded zero-catch reports (37%). Finally, we removed reports lacking release information (<1%). This resulted in a final dataset of 354 562 report entries covering 1 080 197 individual catch records of 39 fish species across 1 286 fisheries.

We tested for temporal trends in catch-and-release (C&R) practices using generalised linear mixed models (GLMMs), fitted with the glmmTMB package. Further, we tested for an association between fish length and C&R practice with generalised additive mixed models (GAMMs), implemented in the mgcv package and fitted using the bam() function. The response variable in all models was the fate of individual fish (binary: kept or released).

(i) To test for overall temporal trend across all fish species and fisheries, we fitted a model including year as a fixed effect, with species identity as a random intercept to capture baseline differences among species. At the fishery level, we included random intercepts and random slopes for year, allowing fisheries to vary in their baseline level and in how C&R changed over time. Syntax model (i): glmmTMB(Fate ~ Year + (1 | Species) + (1 + Year | Fishery), family = binomial (link = 'logit')).

(ii) To test for species-specific temporal patterns, we focused on the six most frequently captured native species for which management objectives typically include both conservation of wild populations and sustainable harvest: northern pike (Esox lucius, Esocidae), European perch (Perca fluviatilis, Percidae), zander (Sander lucioperca, Percidae), European grayling (Thymallus thymallus, Salmonidae), brown trout (Salmo trutta, Salmonidae), and Arctic char (Salvelinus alpinus, Salmonidae). Rainbow trout (Oncorhynchus mykiss, Salmonidae), although among the most frequently reported species overall, was excluded because it is non-native in Sweden and does not reproduce naturally in Swedish waters, occurring primarily in put-and-take fisheries where stocked fish are generally harvested and management focuses on maintaining catch opportunities rather than conserving self-sustaining populations. In this model, we included the interaction between year and species identity as fixed effects to capture potential differences in temporal trends among species. At the fishery level, we added random intercepts and random slopes for year, allowing fisheries to vary in their baseline C&R probability and in how this probability changed over time. Species-specific temporal trends were estimated using marginal slope estimates derived from the fitted GLMM using the emmeans package (Searle et al., 1980). Estimated slopes (coefficients) of the year variable were extracted for each species, together with standard errors and 95% confidence intervals, and tested against zero to test for the presence of statistically significant temporal trends. Syntax model (ii): glmmTMB(Fate ~ Year * Species + (1 + Year | Fishery), family = binomial (link = 'logit')).

(iii) To evaluate whether C&R practices varied with fish body length, we restricted the dataset to the six focal species. Because C&R practices were expected to vary over time and analyses (i) and (ii) already addressed these temporal trends, we focused on contemporary size-selective patterns by using only the five most recent years of data (2020–2024). In Swedish recreational fisheries, anglers traditionally report the body mass of fish rather than their total length, reflecting a long-standing emphasis on weight-based records. To facilitate interpretation for an international audience and allow length-based analyses, total length (cm) was estimated from reported body mass using species-specific conversion factors based on power functions from historical length–mass monitoring data of fish in Swedish waters (Table S1). In cases where anglers reported the combined mass of multiple individuals caught together, typically for smaller or released fish, the total mass was divided evenly to approximate individual body weight before applying the conversion functions, as in Flink et al. (2024). This procedure may introduce some additional uncertainty at the individual level, but is unlikely to affect overall species-specific patterns given the large sample size. To improve model fit, extreme observations at the upper end of the weight range, where only a few individuals were recorded, were removed for each species, and separate GAMMs were then fitted. In each model, total length was included as a predictor using a smooth function (thin plate regression splines, k = 5) to capture potential non-linear length effects. To account for among-fishery variation, we included random intercepts and random slopes of the length–C&R relationship at the fishery level. Syntax model (iii): bam(Fate ~ s(Total length, k = 5) + s(Fishery, bs = "re") + s(Total length, Fishery, bs = "re"), family = binomial (link = 'logit')).

All models were checked for overdispersion, zero-inflation, and uniformity of residuals using simulated residuals generated by the DHARMa package (Hartig 2022). All GLMMs were estimated using maximum likelihood, while GAMMs were estimated using restricted maximum likelihood (fREML). Model predictions were obtained using the predict() function from glmmTMB and mgcv. Predictions were calculated on the link (logit) scale, then back-transformed to probabilities using the inverse logit function. Confidence intervals (95%) were derived from prediction standard errors. For graphical presentation, predictions were shown at the population level by excluding random fishery effects. All statistical tests were evaluated at a significance level of α = 0.05.