What can be learned from fishers’ perceptions for fishery management planning? Case study insights from Sainte-Marie, Madagascar

Local support is critical to the success and longevity of fishery management initiatives. Previous research suggests that how resource users perceive ecological changes, explain them, and cope with them, influences local support. The objectives of this study were two-fold. First, we collated local fishers’ knowledge to characterize the long-term socio-ecological dynamics of the small-scale fishery of Sainte-Marie Island, in Madagascar. Second, we empirically assessed the individual- and site-level factors influencing support for fishery restrictions. Our results indicate that fishers observed a decline in fish abundance and catch sizes, especially in nearshore areas; many also perceived a reduction in fish sizes and the local disappearance of species. To maintain their catches, most fishers adapted by fishing harder and further offshore. Accordingly, fishers identified increased fishing effort (number of fishers and gear evolution) as the main cause of fishery changes. Collectively, our results highlight that the transition from a subsistence to commercial fishery, and resulting changes in the relationship between people and the fisheries, was an underlying driver of fishery changes. Additionally, we found that gender, membership to local associations, coping mechanisms, and perceptions of ecological health, were all interlinked and significantly associated with conservation-oriented attitudes. Conservation-oriented attitudes, however, were not associated with fishers’ willingness to decrease fishing. In the short-term, area-based restrictions could contribute to building support for conservation. In the long-term, addressing the underlying causes of the decline will necessitate collaborations among the various groups involved to progressively build livelihood flexibility. Collectively, our study provides additional insights on the individual- and site-level factors influencing support for fishery restrictions. It also highlights the importance of dialoguing with fishers to ensure that fishery management plans are adapted to the local context.


1: Catch sizes
We used generalized additive models (GAMs) to model temporal changes in best catch sizes for 13 of the most frequent target species family (local Malagasy name) identified from the interviews. GAMs were ideally suited to the structure of our data and the nature of our analysis because; 1) their non-parametric smoothing function (hereafter referred to as smoothers) allowed us to model nonlinear temporal trends (62,63); 2) they can incorporate both continuous and categorical variables; 3) they can accommodate random effects; and, 4) they estimate the shape of the relationship from the data itself (we did not have to specify any a-priori shape). For these reasons, GAM represented a flexible and powerful approach to model temporal trends in best catches, as well as their nature and timing.

Load data
Here, we will use the "Catches.txt" dataset. In the code below, I named it ten_main.

GAM fitting
We use negative binomial GAMs and a logit link. A first model (Mod1) included different intercept for each species, a smoother for time, and its interaction with species. As perceptions can vary among fishermen, we fit fisherman identity as a random intercept and slope. To investigate plausible alternative hypotheses, we also construct several additional candidate models: • Mod2 included the species intercept and a smoother for time.
• Mod3 includes the species intercept.
• Mod4 includes the time smoother.

GAM Model selection
We identify the best model as the one with; the lowest AIC criterion, highest restricted log-likelihood, and highest explanatory power. Furthermore, we can evaluate the significance of the fixed effect using Wald's test.
anova.gam(Mod1,Mod2,test="F")# the interaction is not significant. The plots show that the model fits well, but K is low. Let's try to increase K and see if we can get p > 0.05. However, when increasing k, edf and k don't really change. This shows that the model is fine; for this reason, default k (k=10) is fine.

Fishing site locations
Here, we used generalized linear models (GLM) to model temporal changes in best catch sizes for 13 of the most frequent target species family (local Malagasy name) identified from the interviews.

GLM fitting
To model temporal changes in fishing distance from the shore, we specified a poisson distribution and a log link. As the magnitude of the perceived changes could vary by gear, we fitted a first model (Mod1b) with an interaction between fishing gear and time as a fixed effect. Mod2b included an intercept for gear and time; Mod 3 and 4 included either one of the gear intercept or time.