Project Contacts
Kevin See, QCI

Current Status
Operational Spring/Summer Chinook summer version. R-scripts available on GitHub, currently developing an R package.

Funding
ISEMP

Estimating carrying capacity for rearing parr and spawning adults, and identifying the habitat characteristics that influence capacity is a challenging aspect of Chinook and steelhead recovery planning efforts. ISEMP is using quantile regression forest (QRF) models to estimate the (potentially) non-linear relationships between fish and habitat, while incorporating interactions between habitat variables.  Moreover, QRF models can potentially be used to predict carrying capacity post-restoration to effectively direct restoration actions.

Chinook Summer Parr Capacity

A QRF approach was used to derive empirical estimates of carrying capacity from fish and habitat data collected by ISEMP, CHaMP, and several collaborating entities across a range of interior Columbia River subbasins and habitats. The predicted 90th quantile of fish density was used as a proxy for capacity. Globally available GIS data were used in an extrapolation model to upscale reach-level estimates and produce spatially continuous maps of capacity within a watershed. Such extrapolations can produce estimates of total capacity, as well as provide indications of where on the landscape restoration actions should be targeted. QRF predictions of juvenile Chinook capacity correlated well with independent estimates from a spawner-recruit model, providing robust evidence that we are identifying habitat factors that limit Chinook densities. For juvenile steelhead, we are evaluating whether fish density or biomass is the appropriate response metric and anticipate completing this analysis by September 2017.

Over-winter Parr Capacity

Based on the validation of the summer parr QRF model, we would like to extend that same framework to modeling over-winter capacity. This would be an empirically based estimate of over-winter carrying capacity, which relies on pairing habitat sampling and fish surveys. CHaMP habitat data should suffice to describe winter habitat; however, because fish may exhibit very different seasonal behavior, and display entirely independent fish-habitat relationships between summer and winter, winter fish surveys at CHaMP sites are required to fit a winter QRF model.

To date, we have 43 winter fish surveys (29 in the Entiat, 2 in the John Day, and 12 in the Lemhi) with fish data. This is too small a sample size, with not enough variety in habitat (i.e., too few watersheds), to fit a reliable QRF model. We hope to increase the number and spatial scope of winter fish surveys in future years.

Findings and Uses

QRF models can be used to:

  • Identify important habitat variables associated with juvenile parr or adult redd capacity,
  • Estimate carrying capacity from measurements of those habitat variables,
  • Predict potential improvements to carrying capacity due to restoration actions, and help identify what and where restoration can be most cost-effective, potentially providing a new cost metric: capacity (or fish) per dollar.

In combination with models of capacity for other life stages such as redds and over-wintering, QRF can also be used to help identify which life-stage is most capacity-limited in a watershed and by what habitat limiting factors, so that restoration projects are targeted  to achieve the largest population response.

Estimates of spring/summer Chinook parr capacity as log(parr/m2) for the Lemhi River subbasin.

Partial dependence plots, depicting how capacity shifts as the habitat metric changes, assuming all other habitat metrics remain at their mean values. Tick marks along x-axis depict observed values, and the subbasin they came from.

Partial dependence plots, depicting how capacity shifts as the habitat metric changes, assuming all other habitat metrics remain at their mean values. Tick marks along x-axis depict observed values, and the subbasin they came from.

 Spawner recruit data from several watersheds. Solid lines are the spawner-recruit curve, dashed lines are the estimated capacity and shaded polygons depict the 95% confidence intervals of capacity. Blue corresponds to Beverton-Holt models, green to Ricker models, and red to QRF estimates. The QRF solid curve is a Beverton-Holt model with the capacity parameter fixed to the QRF estimate of capacity.


Spawner recruit data from several watersheds. Solid lines are the spawner-recruit curve, dashed lines are the estimated capacity and shaded polygons depict the 95% confidence intervals of capacity. Blue corresponds to Beverton-Holt models, green to Ricker models, and red to QRF estimates. The QRF solid curve is a Beverton-Holt model with the capacity parameter fixed to the QRF estimate of capacity.

Products

Publications

Coming soon!

Data

Coming soon!