The Chena River hosts one of the largest populations of Chinook Salmon in the Alaska portion of the Yukon River drainage, which has recently experienced poor Chinook returns throughout. Past work suggests that these fish do poorly when they grow slowly due to cold, high-flow conditions during their first summer in the river as juveniles, and the population does well when juveniles experience warm, low-flow years and grow larger. Because high flow drives low water temperatures and vice versa, it was previously unclear to what extent each factor—flow and temperature—affects the fish population.

We collected data to run foraging and bioenergetics models that simulate how flow and temperature affect growth. We invented two sampling devices, a suction pump to sample drifting prey and a computer vision system to measure the inedible debris that occupy most of a fish’s feeding effort. We combined our data with USGS and NOAA records to predict temperature, turbidity, prey, and debris on a daily basis throughout our study years (2019-2020) and others.

This dataset contains the data we collected and compiled from other sources on fish growth, temperature, invertebrate drift, and drifting debris.

Temperature data were collected by placing Hobo data loggers at our study sites (see Study Site Map.png) for two years. We used a machine learning model in Mathematica (gradient-boosted trees) to predict water temperature at the USGS Two Rivers hydrograph (data available from USGS from late-2013 onward) based on streamflow, day-of-year, and air temperature at the Fairbanks airport. We then used that regression, streamflow, and air temperature as predictors for machine learning models (by the same method) to estimate daily temperature (degrees C) at all of our study sites from 1967 to present.

Invertebrate drift data were collected using a suction pump as described in the associated Hydrobiologia publication:

 Neuswanger. J.R., Schoen, E.S., Wipfli, M.S., Volk, C.J., and Schoen, E.R. 2022. A suction pump sampler for invertebrate drift detects exceptionally high concentrations of small invertebrates that drift nets miss. Hydrobiologia (in press). DOI: 10.1007/s10750-022-04849-1.

Juvenile Chinook salmon were sampled via dip net and minnow trap at approximately two-week intervals throughout the summers of 2019 and 2020. Fork length was measured to the nearest mm and mass to the nearest 0.1 g. This produced two fairly high-resolution time series of juvenile Chinook Salmon growth, which we combined with data from the literature, our own past studies, and colleagues' sampling reports to the Alaska Department of Fish and Game to produce several additional time series of fish growth.

Debris data were collected using a Sony mirrorless digital camera to photograph a flow-through chamber at two-second intervals, with both the camera and powerful LED lights controlled by a Raspberry Pi microcomputer. Hundreds of thousands of these debris images were then analyzed using a custom algorithm based on the OpenCV library to count and measure drifting debris particles fish might mistake for prey. Several paragraphs of details about this device can be found in our project's final report to the Alaska Sustainable Salmon Fund, and up-to-date Python code for the debris counter analysis algorithm is available on Github at https://github.com/JasonNeuswanger/DebrisCounterAnalysis.

The spreadsheets in "Completed Field Datasheets" include all environmental data associated with our collections (e.g. turbidity, windspeed) in a consistent format. The date and site name listed on those sheets corresponds to a code "YYYY-MM-DD SiteName" that was used to reference data collected there in subsequent files.

Temperatures throughout these files are given in degrees C. Drift concentrations and energy contents are given in #/m³ and J/m³, respectively. Masses are given in grams and fish lengths in mm.

The "Site Predictions" folder contains observed, predicted, and "best" temperature columns for each of our study site. The "observed" column usually only includes 1-2 years of data. The "best" column uses the observed data when available and the predicted data otherwise.