A Machine Learning Tool for Design of Behavioral Fish Barriers in the Sacramento-San Joaquin River Delta
Survival of out-migrating juvenile salmonids (Oncorhynchus spp.) through the Sacramento-San Joaquin River Delta averages less than 33 percent, depending on water flow through the delta, and is partially governed by the distribution of fish among three Sacramento River distributaries: Sutter, Steamboat, and Georgiana sloughs. Behavioral altering structures in the junctions of the distributaries can effectively increase entrainment into favorable routes, thereby increasing through-delta (Verona to Chips Island, California) survival. The effectiveness of these structures, hence forth called “behavioral barriers,” are dependent on shape, length, location, barrier type, and water velocity, which is governed by Sacramento River discharge (hereinafter referred to as “flow”).
We developed a machine learning tool to optimize behavioral barrier designs at up to three junctions within the Sacramento-San Joaquin Delta for improving through-delta survival of juvenile winter-run Chinook salmon (Oncorhynchus tshawytscha). This barrier optimization tool (BOT) works by evolving barrier solutions in one to three junctions by repeatedly simulating survival of populations of Sacramento River origin fish as they pass through the Delta. Over approximately 6,000 simulations per junction, the BOT converges on barrier designs that result in the greatest average survival given simulated environmental conditions. Survival at each iteration of the model is simulated using a modified version of the salmon travel time and routing simulation (STARS) model. In the BOT, STARS is modified by replacing probabilistic route determinations with an individual based model (IBM) that simulates fish behavior to predict the entrainment rates in each junction. The IBM allows the flexibility to explore how entrainment changes with evolving barrier designs. We used juvenile winter-run-sized Chinook salmon catch data collected at Knights Landing from 1997 to 2011 to create realistic arrival and spatial distributions of simulated fish within the BOT that varied among water years (hereafter years). We demonstrated the capabilities of the BOT by comparing optimized barrier solutions and their resulting simulated improvement in survival among three scenarios that differed in the number of junctions with barriers (Georgiana Slough, Steamboat Slough, or both) and the barrier operational period (early: November 1–March 15, or late: January 1–April 30). In this initial demonstration of the BOT we only considered a bioacoustic fish fence (BAFF) at Georgiana Slough and a floating fish guidance structure (FFGS) at Steamboat Slough.
The increase in simulated through-delta fish survival ranged from 1.0 to 6.3 percent among the optimized barrier designs. The most effective Georgiana Slough barrier design predicted improved survival by 6.3 percent and was chosen by the California Department of Water Resources (DWR) as the Georgiana Slough salmon migratory barrier planned for operation annually from 2023 to 2030 at Georgiana Slough in response to the 2020 California Department of Fish and Wildlife’s (CDFW) Incidental Take Permit Minimization Measure 8.9.1 (California Department of Fish and Wildlife [CDFW], 2020). When barriers were simulated in both junctions, the percentages of simulated winter-run Chinook salmon interacting with a barrier at Steamboat or Georgiana sloughs were 95 percent given the early operational period and 48 percent given the late operational period. When barriers were simulated at both sloughs, the optimal barrier at Steamboat Slough effectively routed fish into the Sacramento River. This is because the Georgiana Slough barrier reduced routing into Georgiana Slough where survival is low, which resulted in higher survival for fish routed down the Sacramento River at Steamboat Slough than fish routed down Steamboat Slough. Whereas when no barrier was simulated at Georgiana Slough, the optimized barrier at Steamboat Slough routed fish into Steamboat Slough. This is because survival was higher through Steamboat Slough than the Sacramento River and Georgiana Slough combined. The greatest improvement in survival (6.3 percent) was predicted over the earlier operational period with only a barrier at Georgiana Slough.
Swyers, N.M., Blake, A., Stumpner, P., Burau, J.R., Burdick, S.M., and Anwar, M.S., 2024, A machine learning tool for design of behavioral fish barriers in the Sacramento-San Joaquin River Delta: U.S. Geological Survey Open-File Report 2023–1095, 38 p., https://doi.org/10.3133/ofr20231095.
ISSN: 2331-1258 (online)
Table of Contents
- Executive Summary
- Barrier Optimization Tool Overview
- The Genetic Algorithm
- Testing and Scoring Candidate Barrier Solutions
- Integration of Models
- The Individual Based Model
- The Applied Computational Framework
- Running Optimizations
- Summary and Conclusion
- References Cited
- Appendix 1
|USGS Numbered Series
|A machine learning tool for design of behavioral fish barriers in the Sacramento-San Joaquin River Delta
|U.S. Geological Survey
|California Water Science Center, Western Fisheries Research Center
|ix, 38 p.
|Sacramento-San Joaquin River Delta
|Online Only (Y/N)
|Google Analytic Metrics