Anatomy of a range contraction: Flow-phenology mismatches threaten salmonid fishes near their trailing edge

Climate change is redistributing life on Earth, with profound impacts for ecosystems and human well-being. While repeat surveys separated by multi-decadal intervals can determine whether observed shifts are in the expected direction (e.g., poleward or upslope due to climate change), they do not reveal their mechanisms or time scales: whether they were gradual responses to environmental trends or punctuated responses to disturbance events. Here we document population reductions and temporary range contractions at multiple sites resulting from drought for three Pacific salmonids at their ranges’ trailing edge. During California’s 2012-2016 historic multi-year drought, the 2013-14 winter stood apart because rainfall was both reduced and delayed. Extremely low river flows during the breeding season (‘flow-phenology mismatch’) reduced or precluded access to breeding habitat. While Chinook (Oncorhynchus tshawytscha) experienced a down-river range shift, entire cohorts failed in individual tributaries (steelhead trout, O. mykiss) and in entire watersheds (coho salmon, O. kisutch). Salmonids returned to impacted sites in subsequent years, rescued by reserves in the ocean, life history diversity, and, in one case, a conservation broodstock program. Large population losses can, however, leave trailing-edge populations vulnerable to extinction due to demographic stochasticity, making permanent range contraction more likely. When only a few large storms occur during high flow season, the timing of particular storms plays an outsized role in determining which migratory fish species are able to access their riverine breeding grounds and persist.

Methods
Hydrological data

Daily flow data were retrieved from USGS for 13 watersheds in northern California, including coastal watersheds from Marin County to Del Norte County (summarized in Carlson et al., SI Appendix, Table S2). The magnitude and timing of flows from the 2013-14 winter were compared to the long-term record (specifically, the most recent 40 complete water years 1983-2022, except in two cases: Walker Creek, where gage records commenced after 1983 leading to inclusion of 39 complete water years, and Austin Creek, where the gage was reactivated in 2004, leading to inclusion of 18 complete water years). A water year in California is defined as October 1 – September 30 (e.g., the 1983 water year corresponds to the period October 1, 1982 to September 30, 1983). We include gage data from Russian River, Austin Creek, Noyo River, Elder Creek, South Fork Eel River, and the Mattole River as representative of conditions in systems where we also have biological observations. Additionally, we present hydrologic data for 7 additional watersheds that spanned a north-south gradient of coastal northern California watersheds, providing a broader regional context for the conditions experienced during the 2013-14 winter. See SI Appendix, Table S1 for a summary of gage data included in our analysis.

For each of the 13 watersheds, we used the e-flows Functional Flows Calculator to identify the onset of winter flow conditions using the ffcAPIClient package in R (version 0.9.8.3, https://github.com/ceff-tech/ffc_api_client). This tool identifies the wet-season start timing as the date that sufficient baseflow has accrued based on a magnitude threshold and a rate of change threshold (46). We obtained the winter onset timing for each year from 1983-2022. The dates were then ranked from smallest to largest using data from the 1983-2022 water years, and the 2013-14 conditions were identified within these ranked conditions. Here, the higher the rank, the later the winter onset date.

We considered the “typical” adult salmonid migration window to extend from 1 November to 31 January, capturing the onset of migrations for the three salmonids (Chinook, coho, and steelhead trout) in the study region.  We computed the mean flow for this window for each winter across the 1983-2022 water years, in each watershed, and then rank-ordered the observations, again identifying the 2014-14 conditions within the ranked conditions. Here, the lower the rank, the more unusually low the flow conditions were.

Biological data

We then combined observations from multiple monitoring efforts to illustrate the consequences of extremely delayed rains for three species of salmon and trout near their trailing range edge. While sampling designs differed among the monitoring efforts (SI Appendix), they fell into two general categories: monitoring of adults on the breeding grounds and/or monitoring of newly hatched juvenile fish (SI Appendix, Table S1). Newly hatched salmonids typically have restricted movement for the first months of their lives (26), so their abundance can indicate the success of adult breeding. The adult monitoring data allowed us to explore the timing and location of breeding, while juvenile monitoring data revealed patterns of juvenile production and cohort strength.

We used a bootstrapped Kolmogorov-Smirnov test to determine if adult arrival timing in 2013-14 was delayed relative to the other years of study (where the cumulative distribution for “all other years” was calculated from the pooled data set across years). We used ks.boot() function in the R package “Matching” with 1000 bootstraps. The test returns the Kolmogorov D (“distance”) statistic and associated p-value (in our case, for a one-tailed test). We present the D statistics in Figure 2, which represents the degree of difference measured as the maximum vertical distance between the reference distribution (“all other years”) and the 2013-14 distribution.

To determine whether the juvenile or adult abundance was significantly reduced in 2013-14 relative to other years of study, we asked whether the abundance estimate fell outside the 95% CI of estimates based on “all other years” of data. The 95% CI was estimated by the mean ± standard error (SE). SE was calculated as the standard deviation divided by square root of the sample size.