Moving on up: Vertical distribution shifts in rocky reef fish species during climate‐driven decline in dissolved oxygen from 1995 to 2009

Abstract Anthropogenic climate change has resulted in warming temperatures and reduced oxygen concentrations in the global oceans. Much remains unknown on the impacts of reduced oxygen concentrations on the biology and distribution of marine fishes. In the Southern California Channel Islands, visual fish surveys were conducted frequently in a manned submersible at three rocky reefs between 1995 and 2009. This area is characterized by a steep bathymetric gradient, with the surveyed sites Anacapa Passage, Footprint and Piggy Bank corresponding to depths near 50, 150 and 300 m. Poisson models were developed for each fish species observed consistently in this network of rocky reefs to determine the impact of depth and year on fish peak distribution. The interaction of depth and year was significant in 23 fish types, with 19 of the modelled peak distributions shifting to a shallower depth over the surveyed time period. Across the 23 fish types, the peak distribution shoaled at an average rate of 8.7 m of vertical depth per decade. Many of the species included in the study, including California sheephead, copper rockfish and blue rockfish, are targeted by commercial and recreational fisheries. CalCOFI hydrographic samples are used to demonstrate significant declines in dissolved oxygen at stations near the survey sites which are forced by a combination of natural multidecadal oscillations and anthropogenic climate change. This study demonstrates in situ fish depth distribution shifts over a 15‐year period concurrent with oxygen decline. Climate‐driven distribution shifts in response to deoxygenation have important implications for fisheries management, including habitat reduction, habitat compression, novel trophic dynamics and reduced body condition. Continued efforts to predict the formation and severity of hypoxic zones and their impact on fisheries dynamics will be essential to guiding effective placement of protected areas and fisheries regulations.


| INTRODUC TI ON
Greenhouse gas emissions have driven global increases in atmospheric and ocean temperatures, which enhance ocean stratification. As the ocean surface layer becomes more buoyant, transport of highly oxygenated surface waters into the ocean interior is reduced (Keeling & Garcia, 2002). Increases in seawater temperature also reduce oxygen solubility (Schmidtko et al., 2017). Marked declines in dissolved oxygen (DO) concentrations and shoaling of the oxygen minimum zone (OMZ) have been observed globally since 1960 (Stramma et al., 2008). These anthropogenic processes are superimposed on natural multidecadal oscillations (Deutsch et al., 2011;Stramma et al., 2019), seasonal patterns (Boyer et al., 1999;Connolly et al., 2010) and storm impacts (Van Dolah & Anderson, 1991;Xu et al., 2019) on ocean oxygen content.
The ecological impacts of reduced ocean oxygen concentrations include altered microbial processes and metabolic rates, changes in predator-prey dynamics and lateral and vertical distribution shifts in marine organisms (Deutsch et al., 2015;Gilly et al., 2013). Hypoxia disproportionately impacts large taxa, including crustaceans, echinoderms and fish, and is associated with decreased fecundity, habitat reductions and a loss of diversity McClatchie et al., 2010;Sato et al., 2018;Stramma et al., 2012). Exposure to hypoxic conditions over a short period of time can often be tolerated by temporary metabolic reductions (i.e. Chew et al., 1990); however exposure over long time periods can lead to growth restrictions, increased risk of predation and mortality (van den Thillart et al., 1994). The impact of low DO concentrations is highly variable between fish species (Davis, 1975;Gray et al., 2002). In a broad review of fish response to hypoxia, Gray et al., (2002) found that many actively swimming fish exhibit growth restrictions at concentrations of 4.2 ml/L and metabolic rates decreased at 2.8 ml/L for benthic fish; mortality can occur for many species at 1.4 ml/L. The observed and forecasted expansion of hypoxic waters have the potential to impact commercial fishing productivity and create regulatory challenges across political boundaries (e.g. Cheung et al., 2012).
The upper 3000 m of the Northeast Pacific has lost over 15% of its oxygen over the last 60 years, with the OMZ expanding at a rate of 3.0 m/year (Ross et al., 2020). Hydrographic data from the California Cooperative Oceanic Fisheries Investigations (CalCOFI) program demonstrate DO declines and OMZ shoaling beginning in the 1980s in the southern California Current System (CCS) . In this time period, the hypoxic boundary has shoaled to depths as shallow as 90 m in parts of Santa Barbara Channel and areas off Point Conception . This expansion and shoaling of the OMZ has the potential to impact fish populations and communities through community reorganization and habitat compression. Previous studies provide insight on the effects of low DO on fish survival, fitness and distribution in the productive California Coastal Current (e.g. Chan et al., 2008;Davis et al., 2018;Flannery, 2018;Gallo, Hardy, et al., 2020;Keller et al., 2017;McClatchie et al., 2010).
Laboratory experiments comparing fish behaviour and metabolic rates between normal and low DO treatments provide a basis to predict fish response to changes in their native habitat.
For example, juvenile rockfish species (including gopher rockfish, Sebastes caratus; copper rockfish, S. caurinus; and black-and-yellow rockfish, S. chrysomelas) from central California that were exposed to hypoxic conditions (DO concentration of 3.15 ml/L) exhibited increased metabolic costs, exploration behaviour and predation mortality compared to normoxic controls (Davis et al., 2018). The swimming performance of juvenile copper rockfish (Sebastes caurinus) and black rockfish (Sebastes melanops) from northern California decreased in hypoxic conditions (DO concentration of 2.8 ml/L and 1.4 ml/L, Flannery, 2018). In laboratory experiments with juvenile rockfish collected from Central California exposed to hypoxic conditions, copper rockfish exhibited behavioural changes such as reduced escape time, and blue rockfish (Sebastes mystinus) experienced elevated mortality rates (Mattiasen et al., 2020). These studies indicate that declining DO may lead to distributional shifts in California rockfish populations, or cause a decrease in survival and fecundity in persistent populations.
Oxygen concentrations have been repeatedly identified as a significant predictor in pelagic and demersal fish distribution (Gallo & Levin, 2016;Netburn & Koslow, 2015). In a study by Gallo, Beckwith, et al., (2020), a remotely operated vehicle (ROV) was used to survey benthic fish communities in the Gulf of California at depths ranging from 200 m to 1400 m. Oxygen level was the best predictor of fish community composition and diversity, and declines in oxygen predicted by a global climate model are expected to drive a reduction in diversity by 2081-2100 (Gallo, Beckwith, et al., 2020). Observations from an autonomous lander at depths from 100-400 m off the coast of San Diego indicate that benthic communities transitioned from fish dominated to invertebrate dominated along a declining oxygen gradient (Gallo, Hardy, et al., 2020). West Coast Groundfish Bottom Trawl surveys conducted within a known hypoxic zone off the coast of Oregon show significantly lower weight to length ratios in five of six groundfish species in low DO regions (<1 ml/L) relative to moderate regions (>1 ml/L, Keller et al., 2010). A temporary anoxic event in the California Current large marine ecosystem was accompanied by the near-complete mortality or abandonment of the anoxic zone by rocky reef macroscopic benthic invertebrates and fish (Chan et al., 2008).
Changes have also been detected in fishery productivity between normal and low DO environments. US West Coast Groundfish Bottom Trawl catch per unit effort was positively associated with DO for 19 of 34 groundfish species in hypoxic (DO <1.43 ml/L) or severely hypoxic (DO <0.5 ml/L) environments (Keller et al., 2017). Total catch per unit effort and species richness were also positively associated with DO concentrations within hypoxic waters (Keller et al., 2010(Keller et al., , 2015(Keller et al., , 2017. Periodic declines in ichthyoplankton abundance in the southern California Current corresponding to low oxygen observed during CalCOFI surveys from 1951 to 2008 indicate that hypoxia may also reduce mesopelagic fish recruitment (Koslow et al., 2011), although not for all species (Koslow et al., 2019)

| Oceanographic data
Oxygen concentration data were pulled from the California To determine the impacts of depth and time on fish abundance, fish were first separated by taxon and life-history status. Life history was divided into two categories: young of the year (YOY) and non-YOY. YOY were defined for each taxon as individuals with a total length less than the average length at 1 year, as predicted by the von Bertalanffy growth function using taxon-specific growth parameters following Claisse et al. (2014) (Table S1).
Separate models were developed for each unique combination of taxon and life history status that was observed during at least 6 years (half of the study period), and had observation years spanning at least a decade. Fish abundance was modelled using Poisson regression as follows: where N is the number of fish within a specific taxon, t, and life-history stage, s, counted on a transect; D is the mean depth of the transect; Y is the year of the survey; D*Y is the interaction between depth and year; and L is the transect length. The depth squared term was added so that fish count estimates were not forced to monotonically increase or decrease across the large depth range observed in this study. This allows the models more flexibility to define the relationship between fish abundance and depth, creating a more realistic fit for species that may have a peak in distribution in this range. All Poisson models were tested for overdispersion using the AER library in R with alpha set at α = 0.05 (Kleiber & Zeileis, 2008; R Core Team, 2019). When  (Table 1); shading around the regression lines represents the 95% confidence intervals for each depth category. Mild hypoxia is defined by the black dashed line at 107 μmol O 2 /kg (~2.45 ml/L), and hypoxia is defined by the black solid line at 61 μmol O 2 /kg (~1.4 ml/L) overdispersion was found, a quasi-Poisson model was run in place of a Poisson regression.
Model predictions were generated for each survey year across the range of depths that a species was observed. The interaction between depth and year provides information on whether the modelled fish type is moving deeper or shallower over time. We define the annual peak distribution as the depth of maximum predicted fish abundance for a given survey year calculated from the model coefficients. Cases were excluded if a modelled peak in distribution across the three reefs did not occur within the taxon's observed depth range during any of the years within our time series.
The model-estimated change in the depth of peak distribution over time was calculated for each fish type. This depth change is defined as the difference between the peak depth distribution of the fish type in its first observed year and its final observed year, divided by the span of observation years. This method may result in an underestimate of the change in peak distribution depth because our surveys were limited to depths between 44 m and 365 m. Due to this limitation, we were not able to observe or model peaks in fish densities that may have moved shallower than 44 m over the 15-year observation span.

| RE SULTS
The analysis of oxygen concentration data at three CalCOFI stations revealed a significant decline in oxygen over the time period 1990-2019 (p < 0.001; Figure 3, Table 1). Deeper depths were charac-  (Table S2). Analogous models assessing changes in temperature and salinity at the same CalCOFI stations from 1990 to 2019 found no significant temporal trend (Table S3).
There is a wide range of definitions for environmental hypoxia in the literature (Hofmann et al., 2011). For the purposes of this study,  (Table S4).
Species-and life stage-specific changes in peak depth distribution of fish over time were examined using the interaction term between depth and year in the Poisson and quasi-Poisson models.
Out of the 60 modelled combinations of fish taxonomic group and life stage, 27 of the models had a statistically significant interaction between depth and year. In four of these fish types, there was no model-estimated peak in depth distribution within the observed depth range (Figure 4; Table 2). In the remaining 23 fish types with a significant interaction between depth and year, the depth of peak fish abundance became shallower over time in 19 of the fish types, consisting of 15 non-YOY taxa and four YOY taxa (Figure 4; Table 2).   the rapid redistribution of marine species to more suitable habitats and the extirpation of populations that are unable to relocate will have unprecedented consequences on marine ecosystem structure and the provisioning of ecosystem resources (Breitburg et al., 2018;Keeling et al., 2009;Levin et al., 2009).
The distributions of fish and invertebrate species, including species with high commercial or recreational fishery value, have been associated with DO in previous studies (Chan et al., 2008;Gallo, Beckwith, et al., 2020;Gallo, Hardy, et al., 2020;Keller et al., 2010Keller et al., , 2015Keller et al., , 2017. However, it is difficult to disentangle ecosystem response to persistent or seasonal low DO due to local bathymetry or local-scale oceanographic processes from multidecadal trends in DO Reef fish at the deepest site, Piggy Bank, and over time at the mid-depth site, Footprint, were exposed to DO concentrations below  (Davis et al., 2018;Flannery, 2018). The observed shift in peak distributions of fishes to shallower depths may be explained by emigration to areas with higher DO concentrations, as supported by behavioural responses to hypoxia in laboratory settings (Davis et al., 2018;Flannery, 2018). whereas anthropogenic warming is projected to correspond to a decrease in nutrient availability (Behrenfeld et al., 2006;Polovina et al., 2008). Further study is required to tease apart the relative impacts of both natural and anthropogenic climate processes and predict future changes in fish distribution.
Fishing regulations and non-linear population dynamics also have the potential to cause changes in fish distribution over time.  (Karpov et al., 2012). Furthermore, fish recruitment, especially in rockfish species, is not consistent over time (e.g. Zabel et al., 2011). For example, a known pulse in juvenile rockfish recruitment occurred in 1999, and this cohort was followed at natural reefs and offshore oil and gas platforms until at least 2004 (Love et al., 2006;Meyer-Gutbrod et al., 2019). Although this study does not critically examine changes in abundance over time, Water quality measurements were not collected concurrently with the visual surveys over the rocky reef system in this study.
The comparison of fish distribution change with oxygen was enabled here by the consistent and long-term sampling of oxygen concentration and temperature at three CalCOFI stations nearby; however, similar surveys could be improved with the collection of in situ oceanographic measurements. Precise DO data collected at each transect would be ideal for building mechanistic models of the effects of DO on fish population and community dynamics.
Surveys that include in situ oxygen concentration sampling would also be useful for identifying fish response to seasonal patterns in DO. While all fish surveys and CalCOFI station sampling included in this study occurred within a 64-day seasonal window over the 15-year time series ( Figure S1), fish survey effort was not high enough to parse out potential seasonal variability within that survey window. The fall period examined in this study is concurrent with a seasonal shift from summertime upwelling-driven hypoxia to higher bottom-water DO levels in the winter, although these oscillations are less pronounced in the southern portion of the California Current such as the Santa Barbara Channel (Peterson et al., 2013).
Most rockfish species, however, have small home ranges and high site fidelity with limited seasonal movements (Green et al., 2014, Jorgensen et al., 2006, Tolimieri et al., 2009. Although observations are limited, known seasonal movements of species such as copper and blue rockfishes are small in scale and occur over the summer, and therefore are less likely to impact our study (Matthews, 1990 The identification of hypoxic environments, tracking their spatial and temporal dynamics, and predicting the response of fish species to this environmental degradation is critical to supporting meaningful fishery management. Oxygen stress will emerge in nearly half of the global no-take marine protected areas by 2050 under the business-as-usual climate projection RCP8.5 (Bruno et al., 2018).
Hypoxic conditions may result in the degradation of 55% and complete loss of 18% of the available demersal habitat within the estab- invertebrates in search of oxygen refugia have the potential to further complicate fishery management by increasing the risk of bycatch (Craig & Bosman, 2013).
Declining oxygen will exceed the range of natural variability in most of the global ocean by 2052 (Henson et al., 2017). Climate models predict a continued decline of up to 7% in global oceanic DO concentrations in the next century (Bopp et al., 2013;Keeling et al., 2010). Monitoring the formation and severity of these low oxygen habitats and the ecosystem response will be a critical component to the effective placement of marine protected areas and the regulation of recreational and commercial fisheries. This study demonstrates significant changes in the depth distribution of rocky reef fish species over a 15-year time period and underscores the need for fisheries management that is responsive to variable, and potentially unprecedented, environmental conditions.

ACK N OWLED G EM ENTS
The

DATA AVA I L A B I L I T Y S TAT E M E N T
The fish survey data that support the findings of this study are