Estimating carrying capacity for juvenile salmon using quantile random forest models

Establishing robust methods and metrics to evaluate habitat quality is critical for the recovery of endangered Pacific salmonids (Oncorhynchus spp.). A variety of modeling approaches are used for status and trend monitoring of anadromous species throughout the Pacific Northwest, USA, but current methods may fail to capture the complex relationship between fish and habitat and are often limited in predictive power beyond specific watersheds. Further, the focus on species distribution and abundance is not easily manipulated to predict carrying capacity and traditional stock-recruitment analyses are reliant on long-term data which are not always available. In this study, we developed a quantile random forest model to provide estimates of habitat carrying capacity for Chinook salmon (O. tshawytscha) parr during the summer months, at both the site and watershed scale. Quantile random forest models allow for the consideration of noisy data, correlated variables, and non-linear relationships: common features in fish-habitat datasets. We leveraged Columbia Habitat Monitoring Program data to select habitat co-variates and predict capacity at those sites. We also identified a set of globally available attributes to extrapolate capacity estimate predictions throughout wadeable streams within the Columbia River basin. Total capacity estimates for watersheds closely matched estimates from alternative fish productivity models. Carrying capacity estimates based on quantile random forest models, like those presented here, provide managers a framework to guide the identification, prioritization, and development of habitat rehabilitation actions to recover salmon populations.

Automated summary

Establishing robust methods and metrics to evaluate habitat quality is critical for the recovery of endangered Pacific salmonids. Modeling approaches like the quantile random forest model developed in this study allow for estimation of habitat carrying capacity at different scales, considering noisy data, correlated variables, and non-linear relationships. These models provide managers with a framework to guide habitat rehabilitation actions and recover salmon populations.