Unprecedented Warming and Increased Herbivory Drive Collapse of Japan's Kelp Forests

ABSTRACT

Kelp forests underpin entire ecosystems and provide key services for human society, yet are threatened by anthropogenic stressors, most notably ocean warming and overgrazing. While Japanese kelp forests have substantial historical and cultural value, responses to recent environmental change remain poorly resolved. Here, we assessed how recent Kuroshio‐associated warming and increased herbivory were linked to kelp forest collapse across approximately 200 km of the Izu Peninsula (central‐eastern Japan). We combined 40 years of in situ kelp biomass (1980–2024) and 12–14 years of canopy‐cover records (2009–2025) with satellite‐derived sea‐surface temperature and marine heatwave metrics. Herbivorous fish assemblages and bite rates were surveyed monthly (2021–2022) across reefs spanning healthy, mosaic, and barren states. Laboratory experiments quantified lethal and sub‐lethal thermal thresholds of juvenile Ecklonia cava using temperature‐ramping (20°C–33°C) and growth trials (22°C vs. 28°C). Spatial analyses compared thermal‐threshold exceedance and seasonal grazing‐window expansion with observed canopy loss, evaluating whether warming and herbivory were consistent with the spatial extent of decline. Warming along the Izu coast intensified during the recent Kuroshio Large Meander, increasing from 0.15°C decade−1 (1982–2016) to 2.98°C decade⁻¹ (2017–2024), with cumulative marine heatwave intensity rising from 2.34°C to 63.22°C·day decade⁻¹. Over the same period, canopy‐forming E. cava cover fell from ~75% to zero, and peak biomass declined from 2.37 kg DW m⁻² to zero, indicating local extirpation and likely regional functional loss. Laboratory assays showed rapid physiological decline above 29.5°C and mortality at 32°C–33°C; temperatures exceeded during summer 2023/2024. The grazing season by warm‐adapted herbivorous fishes expanded by up to 86 days, amplifying kelp loss across heat‐stressed reefs. Together, extreme thermal stress and expanded herbivory were strongly associated with a shift from kelp forests to barren reefs, highlighting how physiological stress and trophic processes can contribute to rapid kelp loss in warming boundary‐current systems.

Rapid warming during the Kuroshio Large Meander intensified marine heatwaves beyond kelp physiological limits while simultaneously extending grazing windows of thermally expanding herbivorous fishes. This synergistic interaction between climate‐driven thermal stress and prolonged herbivory triggered kelp forest collapse and a persistent shift to barren reefs, positioning this western boundary current system as a sentinel warning of threshold‐driven ecosystem collapse under accelerating ocean warming.

1. Introduction

Kelp forests are among the most productive and structurally complex coastal marine ecosystems (Pessarrodona et al. 2022). By creating extensive three‐dimensional habitats, kelp species sustain high biodiversity and support critical ecological functions, including nutrient cycling, shoreline protection, and carbon cycling (Eger et al. 2023; Graham et al. 2007; Steneck et al. 2002; Wernberg et al. 2016). Kelp species are distributed across approximately one quarter of the global coastline (Jayathilake and Costello 2020), where they function as foundation organisms and have elevated ecological and socioeconomic importance (Eger et al. 2023; Steneck et al. 2002). However, kelp species and the ecosystems they underpin have declined in many regions in response to recent environmental change (Smale et al. 2013; Wernberg et al. 2019).

In recent decades, kelp forest declines have been driven by the cumulative effects of both large‐scale climatic stressors, particularly ocean warming and increasing marine heatwaves (MHWs), and localised anthropogenic pressures such as eutrophication, habitat modification, reduced water clarity, and overgrazing by herbivorous fish and sea urchins (Bennett et al. 2015; Connell et al. 2008; Krumhansl et al. 2016; Ling et al. 2015; Smale 2020; Tait et al. 2021). Regionally, trajectories vary substantially; while kelp forests in some areas exhibit stability, others have undergone regime shifts to sea urchin or turf‐dominated systems with reduced ecological functionality (Filbee‐Dexter and Scheibling 2014; Filbee‐Dexter and Wernberg 2018; Mora‐Soto et al. 2024; Smale et al. 2025; Tait and Schiel 2011). The loss of canopy‐forming kelps often initiates cascading effects, such as declines in fish and invertebrate diversity and the proliferation of opportunistic turf algae that inhibit kelp recruitment (Layton et al. 2019; O'Brien and Scheibling 2018). Understanding the mechanisms driving shifts in habitat and ecosystem structure across local, regional and global scales is therefore essential for predicting future trajectories of kelp forests and developing effective conservation and restoration strategies.

Climate change, particularly ocean warming, has emerged as a dominant force structuring contemporary kelp forests (Smale 2020; Wernberg et al. 2024). In addition to gradual ocean warming trends, many regions are increasingly affected by MHWs, which are characterised by acute thermal anomalies that can exceed physiological thresholds and trigger widespread canopy collapse (Smale et al. 2019; Smale and Wernberg 2013; Wernberg et al. 2016). MHWs are becoming more frequent, intense, and prolonged as a result of anthropogenic climate change (Frölicher et al. 2018; Oliver et al. 2018), thereby exerting greater thermal stress and compressing post‐disturbance recovery windows by extending the duration and recurrence of extreme temperatures (Smale et al. 2019; Wernberg et al. 2016). Concurrently, longer‐term ocean warming trends are driving the poleward migration of many species as they track favourable thermal habitat (Lenoir et al. 2020; Poloczanska et al. 2013). In mid‐latitude regions often dominated by kelp forests, this ‘tropicalisation’ process has introduced novel herbivores and competitors into formerly temperate systems (Kumagai et al. 2018; Vergés et al. 2014, 2016). These interacting physical and biological stressors can act synergistically, whereby heat‐induced physiological stress reduces kelp resilience and growth, while intensified herbivory suppresses recruitment and prevents recovery, potentially pushing coastal ecosystems beyond ecological thresholds into persistent alternative states.

On Japan's temperate reefs, intensified herbivory by fishes is a major driver of ‘isoyake’, the Japanese term for barren grounds devoid of kelp and other canopy‐forming seaweeds. Several species are considered to contribute to this persistent grazing pressure, most notably the rabbitfish ( Siganus fuscescens ) and the Japanese parrotfish ( Calotomus japonicus ), which remove substantial kelp biomass through blade browsing and stipe scraping. In addition, the Japanese sawtail ( Prionurus scalprum ) and largescale blackfish ( Girella punctata ) have been identified as significant grazers in southwestern and central Japan, where foraging contributes to macroalgal deforestation (Barrientos et al. 2022; Vergés et al. 2022). Feeding activity typically intensifies during late summer/early autumn and at water temperatures > 20°C (Tanaka et al. 2012), which often follow periods of kelp recruitment when juvenile kelp plants are particularly susceptible to grazing, thereby suppressing recovery. In addition, sea temperatures during late summer/early autumn are conducive to larval settlement and post‐settlement survival of several warm‐adapted herbivorous fishes, particularly during warmer periods or years, which further heightens grazing pressure (Kumagai et al. 2018; Tanaka et al. 2012; Vergés et al. 2014, 2016). As such, ocean warming and intensified herbivory interact to truncate recovery windows, reduce resilience, and accelerate kelp forest collapse.

The Izu Peninsula, at the climatic transition between temperate and subtropical waters off southeastern Honshu, has historically supported dense kelp forests typically dominated by Ecklonia cava and Eisenia bicyclis, both of which are found towards their warm trailing range edge in the region. The coastal climate of Izu is shaped by the Kuroshio Current, a western boundary current warming faster than many ocean regions (Sugimoto 2025). During large‐meander phases, the Kuroshio's path shifts shoreward, pushing warm offshore water toward the Izu–Tokai coast, elevating coastal sea temperatures and promoting MHW activity (Hirata et al. 2025). An extended large meander persisting from August 2017 to April 2025 (the longest since records began in 1965) produced sustained thermal anomalies and heightened variability (Hirata et al. 2025; Japan Meteorological Agency 2025), creating a natural experiment (and contemporary analogue) for assessing how prolonged warming and compounding drivers influence kelp population structure and wider ecosystem resilience.

We focused on the Izu Peninsula as a model sentinel region for assessing the resilience of kelp forests faced with concurrent rapid ocean warming and increased herbivory. Long‐term kelp biomass and canopy‐cover records were used to quantify the timing and magnitude of kelp decline, while satellite‐derived sea surface temperature (SST) and MHW metrics were used to characterise chronic and acute thermal forcing. Controlled laboratory experiments were used to estimate the thermal sensitivity and growth response of juvenile E. cava; contemporary fish surveys and bite‐rate assays to quantify herbivorous fish abundance, grazing intensity, and seasonal activity windows; and spatial mapping was used to evaluate how recent warming, experimentally derived thermal thresholds, and herbivore grazing windows corresponded with kelp presence and functional loss across the Izu Peninsula. Our integrated approach aimed to assess the resilience of Izu's kelp forests to intensified abiotic and biotic stress and to elucidate the mechanisms driving change, which in turn will inform predictions of resilience and recovery under future climate scenarios.

2. Materials and Methods

2.1. Study Site and Species Description

This study was predominantly conducted in the coastal waters surrounding Shimoda, Shizuoka Prefecture, Japan (34.67° N, 138.95° E), on the southeastern Izu Peninsula. The coastline of Shimoda is characterised by small bays with rocky reefs extending to depths of ~20–25 m that historically supported extensive kelp forests. The study area is a semi‐rural rocky‐reef coastline with no major industrial development adjacent to any of the monitoring sites. The dominant canopy‐forming kelp, Ecklonia cava Kjellman 1885 (Laminariales), is a perennial brown alga widely distributed across temperate rocky reefs of southern and central Japan. It exhibits a marked seasonal cycle in which new blades form in autumn/winter, elongation peaks in late winter/spring, and blade thickening and sorus development occur through summer (Yokohama et al. 1987). Old blades then senesce through autumn, so that minimal standing stock biomass occurs in winter and maximum biomass is observed in summer. Eisenia bicyclis (Kjellman) Setchell 1905 forms dense understories or secondary canopies in more exposed habitats. Together, these kelps contribute substantially to reef structure and productivity, and underpin kelp forests on the Izu Peninsula, as well as other regions across central/southern Japan and South Korea (Choi et al. 2024; Terada et al. 2021), supporting diverse assemblages of algae, invertebrates and fishes. The Izu Peninsula is located towards the warm trailing range edge of both species. In this study, E. cava served as the focal species for physiological and experimental analyses, whereas E. bicyclis was monitored only through field surveys of canopy cover.

2.2. Environmental Conditions and Marine MHW Characterisation

Daily sea surface temperature (SST) data were obtained from the UK Met Office GHRSST Level‐4 OSTIA Global Historical Reprocessed Foundation SST analysis (0.05° resolution) for the period 1982–2023. Because this product terminates on 2023‐12‐31, SST for 2024 was taken from the GHRSST Level‐4 MUR Global Foundation SST analysis (v4.1) and bilinearly resampled up to 0.05° to maintain consistent spatial resolution. These datasets were used to characterise SST and marine heatwave (MHW) conditions across the wider Japanese shelf region (26°–46° N, 127°–147° E), with focused analyses in two areas: the Izu Peninsula (34.5°–35.2° N, 138.6°–139.3° E) and Shimoda (centred at 34.67° N, 138.95° E). Satellite‐derived SST was used because it provides a spatially consistent, multi‐decadal temperature record across the study region, allowing comparison of long‐term warming, marine heatwave exposure, and spatial patterns of thermal‐threshold exceedance. However, it should be acknowledged that SST may not capture fine‐scale nearshore or benthic thermal variability experienced by kelps at depth. Comparison with previous in situ temperature measurements at the study site (Agostini et al. 2021) showed that satellite SST tracked local temperature patterns well, although satellite‐derived values were approximately 0.9°C higher than in situ data (Figure S1).

For each pixel, as well as for the Izu Peninsula area‐mean time series, MHWs were detected following the Hobday et al. (2016) framework using the heatwaveR package. A daily climatology and seasonally varying MHW threshold were computed from the 1982–2011 baseline period. Spatial trends in SST and cumulative MHW intensity (°C·day·decade⁻¹) were then estimated separately for 1982–2016 and 2017–2024 to contrast pre‐ and during‐Kuroshio Large Meander conditions (Spatial trends for the 1982–2024 are shown in Figure S2). Representative event‐level time series for Shimoda were extracted for 2016 and 2024, illustrating MHW characteristics during a pre‐meander year and during the period of peak warming and MHW activity.

2.3. Percent Cover (Abundance) Surveys

Long‐term changes in kelp cover were reconstructed from two monitoring programs. The Japan Ministry of the Environment's Monitoring Site 1000 program established permanent 2 × 2 m quadrats at 5–9 m depth in Oura Bay (e.g., Terada et al. 2021). Annual surveys have been conducted from 2009 to 2023 (typically September–October), with the percent cover of kelps recorded to the nearest 5%. A complementary program initiated by the authors in 2013 monitored a shallower site (2–4 m depth) in the same bay. Ten permanent 50 × 50 cm quadrats were photographed monthly, and percent cover was estimated from images using an 8 × 8 point grid (see Heitzman et al. 2022 for details). Although this second program focused on sessile benthic communities such as corals and understory algae, kelp recruits frequently established in these quadrats, providing valuable information on understory kelp and recruitment dynamics. However, distinguishing between juvenile E. cava and E. bicyclis from photoquadrats was not possible, and cover estimates represent both kelp species combined. Thus, the annual surveys were interpreted primarily as records of canopy persistence and adult cover, whereas the monthly photoquadrats were interpreted as records of recruitment dynamics rather than species‐specific recruitment.

2.4. Standing Stock Biomass Surveys

To capture long‐term changes in E. cava biomass, we collated four survey periods spanning 1980–2024, all conducted at ~5–10 m depth within the same area within Oura Bay, Shimoda. The earliest dataset (1980–1981; Yokohama et al. 1987) consisted of monthly harvests of 1 m² quadrats (n = 2 per month), with all thalli collected by SCUBA divers, dried, and weighed as dry biomass (kg DW m⁻²). A subsequent dataset (1996–1997; Serisawa, Akino, et al. 2001) used seasonal harvests of 1 m² quadrats (n = 3–4 per survey), in which thalli were separated into blades and stipes and weighed as fresh (wet) biomass (kg FW m⁻²). During our own surveys in 2015–2016, we applied the same quadrat harvest method (1 m² quadrats, n = 2–4 per month) and measured both fresh and dry biomass, allowing the calculation of a conversion factor (mean ± SE: 6.93 ± 0.12 g FW per g DW, n = 165). This closely matched the ratio reported by Serisawa, Akino, et al. (2001; 6–7 times), which we used to standardise the 1996–1997 values to dry biomass. In the most recent surveys (2023–2024), divers conducted rapid timed searches (5–10 min per transect, n = 3) across the historical forest area at 5–10 m depth. No thalli or recruits were detected; biomass was therefore recorded as zero. Standardising all datasets to dry biomass allowed direct comparison of standing stock across decades, although differences in sampling frequency and seasonal coverage mean that these data are best interpreted as broad changes in E. cava standing stock among survey periods rather than fine‐scale seasonal comparisons with equal precision across all decades.

For each survey period, we fitted seasonal generalised additive models (GAM) to describe the annual standing stock trajectory of E. cava. Because 2015–2016 has a mid‐year gap in observations, we retained a biologically realistic seasonal shape by scaling the 1980–1981 seasonal GAM to the 2015–2016 magnitude using a log‐space least‐squares factor. The 95% CI ribbon for 2015–2016 reflects the raw 2015–2016 variability (month‐wise envelope). From each fitted seasonal curve, two biologically relevant summary statistics were extracted: (i) peak seasonal biomass (the maximum predicted value across the annual cycle), and (ii) mean seasonal biomass (the average predicted value across all 12 months). To estimate uncertainty in these summaries, we implemented a non‐parametric case bootstrap within each survey period. Specifically, raw quadrat observations were resampled with replacement 1000 times, the seasonal model was re‐fitted to each resample, and the two summary metrics were recomputed. The 2.5th and 97.5th percentiles of the bootstrap distributions were taken as 95% confidence intervals. For 2015–2016, the bootstrap distributions were scaled by the same factor applied to the seasonal curve.

2.5. Thermal Tolerance and Growth Rates

We conducted two separate laboratory experiments to assess the sensitivity of E. cava juveniles to warming, the first to establish critical upper limits and the second to assess growth rates at sub‐lethal temperatures. For the first experiment, ten individuals (juveniles, ~15–20 cm length) were collected in May 2024 from a remnant E. cava population in a small inlet on the opposite side of Oura Bay (34.6639° N, 138.9395° E). Individuals were placed into individual tanks supplied with running unfiltered natural seawater, aeration, and artificial light (12 h day⁻¹, ~3 mol m⁻² day⁻¹). Temperature was controlled with a 300 W heater and logged every 5 min (HOBO pendant, ONSET). After a three‐day acclimation at ambient (collection) temperature (mean 20.26 ± SD 0.45°C), tank temperature was increased by 1°C every two days until reaching a target of 33°C (mean 33.45 ± SD 0.56°C). Light (measured at the bottom of the tank) was mean 74.78 ± SD 9.46 μmol m⁻² s⁻¹. Maximum photosynthetic efficiency (F _v /F _m ) was measured every two days (shortly before temperature increase) with a PAM fluorometer (Junior PAM, Heinz Walz GmbH) following 30 min dark adaptation, with two technical replicate spots per individual averaged before analysis. Upper critical temperature was inferred from sharp declines in F _v /F _m (< 0.6, approaching 0.1). For the growth experiment, ten additional juveniles were collected in June 2024 from the same donor population used in the thermal experiment. Following a three‐day acclimation, half the tanks (n = 5) were held at 22°C (mean 22.07 ± SD 0.34°C), while the other half were gradually warmed (1°C every two days) to 28°C (mean 27.8 ± SD 0.17°C). Net growth was assessed over 35 days from scaled weekly photographs, calculating planar surface area (cm²) relative to week‐0 (ImageJ; Schneider et al. 2012).

2.6. Fish Surveys on Abundance and Grazing Pressure

Fish assemblages, grazing pressure, and habitat state were quantified from May 2021 to March 2022 at three sites (each circa 2000 m²) spanning a gradient in kelp condition: a kelp‐forest site (34°39′58.4″ N, 138°56′27.3″ E), a mosaic site where kelp forest and barrens co‐occurred (34°39′54.7″ N, 138°56′33.9″ E), and a barren site devoid of adult kelp (34°39′07.3″ N, 138°57′44.3″ E). Surveys were conducted approximately once per month between 09:00 and 11:00 h; adverse weather prevented sampling at the kelp‐forest site in May 2021 and at the barren site in November 2021 and February 2022. The sites differed primarily in kelp condition, although some differences in exposure and local reef configuration were unavoidable. In particular, the barren site was more exposed than the two Oura Bay sites, and this site‐level context is considered when interpreting among‐site comparisons. The barren site had previously supported a healthy kelp forest, as documented by earlier surveys (E. cava was first observed to be extirpated here in 2020). A map of the three survey sites is provided in Figure S3.

At each site, belt transects (5 × 20 m) were established, with the number of transects determined by the extent and configuration of suitable contiguous rocky reef at each site: six at the kelp‐forest site, eight at the mosaic site, and seven at the barren site. Because each reef state was represented by a single site, comparisons among forest, mosaic, and barren conditions should be interpreted as local contrasts among habitat states rather than fully spatially replicated tests of reef‐state effects. Transects were surveyed by SCUBA; depth was recorded on each occasion (maximum ~7 m, mean ~4 m). We only focused on herbivorous fishes. All herbivorous fishes encountered within each transect (up to 5 m ahead of the observer) were identified to species in situ and counted, while divers simultaneously recorded continuous forward‐facing video footage to validate field identifications (following Cattano et al. 2020, apart from belt‐transect length). For Siganus fuscescens (Houttuyn, 1782), a key herbivorous fish species, juveniles and adults were recorded separately to capture known ontogenetic differences in abundance and feeding behaviour. Herbivorous fishes were classified into two feeding groups: browsers ( Calotomus japonicus Valenciennes, 1840; Girella punctata Gray, 1835; and adult S. fuscescens ), and grazers ( Acanthurus nigrofuscus (Forsskål, 1775); Prionurus scalprum Valenciennes, 1835; and juvenile S. fuscescens ) (grouping based on: Choat et al. 2002; Rasher et al. 2013).

To quantify grazing pressure based on bite rates, six bottom‐mounted video cameras (60 fps) were deployed for one‐hour periods during each survey. Cameras were used to quantify bite rates on focal kelp thalli or turf substrata rather than to estimate grazing per unit reef area. Cameras were placed ~3 m from natural kelp stands so that entire focal thalli were visible within the field of view. To minimise differences in visibility between kelp forests and barren habitats, cameras were mounted on elevated stands approximately 30 cm above the seabed and, where possible, positioned on rocks to further elevate them above the canopy. This arrangement was used to maintain a broadly comparable field of view among kelp‐forest, mosaic, and barren habitats. When natural kelp still obscured the lens, the camera position was adjusted to ensure the focal thallus remained visible. Where kelp was absent, E. cava or E. bicyclis thalli were transplanted from a nearby donor forest, fixed to sandbags with cable ties, and positioned as focal plants. Reference footage from barrens without transplanted or natural kelp plants was also obtained. Transplanted kelp was used throughout at the barren site and, following natural kelp loss, from October 2021 at the mosaic site and from November 2021 at the kelp‐forest site. Video was analysed to enumerate discrete bite events (including consecutive bites by the same individual); uncertain identifications were excluded. Browsers were scored for bites on kelp thalli or stipe, whereas grazers were scored for bites on turf algae on rock; during periods of kelp introduction only browser bites on transplanted kelp were counted. Bite counts were converted to hourly rates (bites h⁻¹) per species, with juveniles and adults tallied separately for S. fuscescens .

Because kelp availability differed among sites and over time, bite rates were standardised within sites to enable temporal comparisons independent of absolute canopy abundance. Raw bite rates are provided in Figure S4 to allow comparison of absolute grazing magnitude. Although cameras were elevated to maintain a broadly comparable field of view among habitats, bite rates should still be interpreted as focal‐thallus or focal‐substratum grazing rates rather than fully area‐standardised estimates of reef‐wide grazing pressure. For each site, we identified the maximum raw bite rate observed on kelp (for browsers) and on turf (for grazers) across the study and divided monthly means, their standard errors, and raw observations by this site‐specific maximum to obtain relative grazing intensity (0–1 scale). This scaling preserves intra‐site seasonal dynamics while minimising confounding from between‐site differences in kelp availability.

To place grazing in a thermal context, daily bite records were aligned with daily sea surface temperature (described above). For each species, we constructed a bite‐weighted temperature distribution from the study‐period records (using only the barren site) and took the bite‐weighted 5th percentile as the lower operating threshold for active grazing. We did not use an upper bound as these tropical and subtropical fishes routinely experience higher temperatures than those experienced during our surveys, so an upper bound would not have provided a consistent operational limit in this system. We only used the barren site as the lack of natural kelp in that site gave us a constant kelp density throughout (whereas kelp condition changed over time at the other two sites). Using the long‐term daily SST, we then derived annual window metrics from calendar years while allowing windows to bridge past 31 December. A day was considered suitable if SST ≥ the lower bound. To reduce spurious crossings, suitability required ≥ 5 consecutive days; short interruptions of ≤ 2 days were bridged when identifying continuous runs. For each recorded species and year, the first suitable day of the year for grazing (first day of the longest suitable run), the last suitable day (last day of that run), and the span (days in the longest run) were ascertained. To test for regime‐scale change associated with the Kuroshio Large Meander, we compared these metrics between 2009–2016 (pre‐meander) and 2017–2024 (Large Meander).

2.7. Spatial Extent of Exceedance of Kelp Thermal Thresholds

To illustrate how kelp distribution across the Izu Peninsula may respond to the conditions described above, we evaluated (i) whether local mean SST was altered, (ii) whether local SST exceeded lethal thresholds for kelp derived from our thermal tolerance experiment, and (iii) whether the annual thermal grazing window derived from our fish‐survey grazing data was extended. Whereas the long‐term temperature and MHW trends were derived from the longer‐term UK Met Office GHRSST Level‐4 OSTIA Global Historical Reprocessed Foundation SST analysis (0.05° resolution), the GHRSST Level‐4 MUR Global Foundation SST analysis (v4.1) product (0.01° resolution) was used here to provide fine‐scale spatial mapping of recent thermal conditions across the Izu coast. The baseline distribution of seaweed forests was compiled from Hasegawa (2010). We conducted qualitative presence/absence surveys at selected sites during 2013–2019 and resurveyed sites where kelp was present during 2019–2024. These surveys recorded whether canopy‐forming kelps were present, functionally absent, or associated with barren habitat, but were not used to estimate quantitative percent cover or biomass. The qualitative observations were used to contextualise spatial patterns of recent warming, exceedance of experimentally derived thermal thresholds, and expansion of herbivore grazing windows across the Izu coast.

2.8. Statistical Analyses

All analyses were conducted in R v.4.3.1 using the tidyverse (dplyr, tidyr, ggplot2).

Long‐term SST trends and annual mean cumulative intensity were quantified using the Theil–Sen slope estimator (trend::sens.slope). Annual changes in kelp canopy cover were analysed with a generalised linear mixed model (GLMM) with a beta‐binomial error distribution and logit link (glmmTMB::glmmTMB). Percent cover values were expressed as binomial successes, with year (centred) included as a fixed effect and quadrat as a random intercept. Model residuals were evaluated using simulation‐based diagnostics in DHARMa. To evaluate whether temporal autocorrelation influenced inference, we conducted generalised least‐squares sensitivity analyses with first‐order autoregressive residual structure [AR(1); nlme::gls]. For SST and MHW metrics, GLS models were fitted to annual time series. For annual kelp‐cover responses, GLS models were fitted to annual mean cover after logit transformation. For the monthly survey, GLS models retained month as a fixed seasonal effect. These sensitivity analyses were used to assess whether the direction of temporal change was robust to temporal‐dependence assumptions. Across all SST/MHW and kelp‐cover analyses, AR(1) sensitivity models retained the same direction of temporal change as the primary analyses, indicating that temporal autocorrelation did not alter the interpretation of the main trends.

Seasonal patterns of E. cava standing stock biomass across decades were analysed with generalised additive models (GAMs; mgcv::gam) using cyclic cubic splines for month, fitted separately for each survey period. Models were fitted by restricted maximum likelihood (REML) with shrinkage, and 95% confidence intervals were obtained using gratia.

To estimate the upper critical temperature at which PSII performance declined, technical PAM measurements were first averaged for each tank at each target temperature. A linear model with mean F _v /F _m as the response variable, mean temperature during the preceding 12 h as the continuous predictor, and tank included as a fixed blocking factor was first fitted. Including tank as a fixed effect allowed each tank to have its own intercept, thereby accounting for baseline differences among tanks while estimating a common temperature response. Segmented regression (segmented::segmented) was then applied to this model to estimate a common breakpoint (ψ, °C) and slopes below and above ψ. Because each juvenile was held in an individual tank, tank corresponded to the experimental unit. The threshold estimate is therefore interpreted as an experimental estimate of the temperature at which juvenile E. cava from this population showed a sharp decline in PSII performance, rather than as a universal species‐level thermal limit.

To test for differences in growth rate between treatments, we fitted a linear mixed‐effects model (lme4::lmer) with week (0–5), temperature (22°C vs. 28°C) as a fixed effect and tank as a random intercept. The temperature × week interaction was used to test whether weekly growth rate differed between treatments. Weekly growth slopes were estimated with emmeans::emtrends, and contrasts between treatments were reported with 95% confidence intervals. To evaluate whether temporal autocorrelation influenced inference, we also fitted a generalised least‐squares sensitivity model with first‐order autoregressive residual structure [AR(1); nlme::gls] to tank‐by‐week mean growth values, using week, temperature treatment, and their interaction as fixed effects. This sensitivity analysis supported the same interpretation as the primary mixed‐effects model.

For all models, residuals and assumptions were checked using simulation‐based or residual diagnostics in DHARMa, car, and performance packages.

For each fish species, the lower thermal operating bounds for grazing activity were identified from observed bite‐rate–temperature relationships (described above). Using these thresholds, we quantified the first and last suitable days of each year and compared their timing between 2009–2016 (pre‐meander) and 2017–2024 (Large Meander) with Wilcoxon rank‐sum tests, applying Benjamini–Hochberg correction for false discovery rate.

3. Results

3.1. Environmental Conditions and MHWs

Sea surface temperatures across the wider Japanese shelf have warmed markedly over the last four decades, with the strongest acceleration along the Kuroshio path and around the Izu Peninsula, and a second pronounced hotspot emerging in the zone of confluence where the warm Kuroshio Current meets the cold Oyashio Current (off the east coast of central/northern Japan) (Figure 1). From 1982 to 2016, SST trends were positive but relatively modest along most of coastal Japan (typically < 0.25°C decade⁻¹), whereas during the Kuroshio Large Meander (2017–2024) warming intensified to several °C decade⁻¹ along the Tokai and Izu coasts and in the Kuroshio–Oyashio confluence region, placing the Izu Peninsula and the offshore mixing zone among the most rapidly warming nearshore and shelf hotspots (Figure 1A,B). Within the Izu Peninsula, the mean warming rate increased from 0.15°C decade⁻¹ (95% CI: 0.14–0.16) during 1982–2016 to 2.98°C decade⁻¹ (95% CI: 2.93–3.03) during 2017–2024.

FIGURE 1.

Spatial and temporal evolution of ocean warming and marine heatwaves (MHWs) in the Izu region. (A, B) Trends in sea surface temperature (SST; °C·decade⁻¹) for 1982–2016 (A) and 2017–2024 (B). The blue rectangle indicates the Izu Peninsula, enlarged in the inset; Shimoda site is marked with a blue point. (C, D) Trends in annual mean cumulative MHW intensity (°C·day·decade⁻¹) over the same two periods. (E, F) Example daily time series for Shimoda in a pre‐meander year (2016; E) and during the Kuroshio Large Meander (2024; F), showing observed SST (black line), climatological mean (grey line), MHW threshold (green line), and MHW events (red shading; darker tones indicate higher MHW categories). (G, H) Long‐term changes in (G) the number of MHW days per year and (H) annual cumulative MHW intensity (°C·day) for the Izu‐area (averaged across the region; blue) and Shimoda (red). The grey band denotes the Kuroshio Large Meander period (approximately August 2017–April 2025).

Trends in annual cumulative marine heatwave (MHW) intensity showed a similar shift. Prior to 2017, increases in heat accumulation were weak and spatially patchy, but during the Large Meander they became strongly positive along the Kuroshio Current, with particularly rapid intensification around the Izu Peninsula and in the Kuroshio–Oyashio confluence zone (Figure 1C,D). Likewise, cumulative MHW intensity within the Izu region accelerated from 2.34°C·day decade⁻¹ (95% CI: 2.15–2.53) before 2017 to 63.22°C·day decade⁻¹ (95% CI: 57.52–68.93) during the Large Meander period. At Shimoda, daily temperature records illustrate how event characteristics have changed: compared with the pre‐meander year 2016, 2024 was characterised by more frequent, longer, and higher‐category MHWs (Figure 1E,F). Consistent with these patterns, both the number of MHW days per year and the annual cumulative intensity (°C·day) increased steadily from 1982 to 2024 for the Izu‐area (averaged) and for the Shimoda site, with record‐high values during the Large Meander period (Figure 1G,H).

3.2. Percent Cover (Abundance) Surveys

The percent cover of canopy‐forming kelps declined sharply over the survey period (Figure 2). In the ‘Monitoring Site 1000’ quadrats (5–9 m depth), canopy cover of E. cava averaged 75%–80% in 2009–2010, then fell below 5% by 2020 and was absent by 2022–2023 (Figure 2B). Generalised linear mixed‐effects models (GLMMs) confirmed a significant negative temporal trend for E. cava (−5.23% ± 1.04% cover year⁻¹, p < 0.001). E. bicyclis comprised a smaller fraction of canopy cover (typically ≤ 10%) and exhibited a similar decline, but response‐scale (back‐transformed) contrasts were not statistically significant. Interestingly, E. bicyclis canopy was undetected after 2014, then briefly reappeared in 2019 before being undetectable from 2020 onward (Figure 2B).

FIGURE 2.

(A) Representative images depicting temporal changes in kelp cover at the principal survey site in 2016, 2020, and 2024. (B, C) Temporal change in kelp percent cover. (B) kelp canopy coverage from the annual monitoring program (from 2009 to 2023). (C) Understorey kelp coverage from the same survey, as well as understorey coverage of kelp from the monthly photoquadrat survey (from 2013 onwards). Blue: Ecklonia cava, Green: Eisenia bicyclis, Purple: Mixture of E. cava and E. bicyclis. Points for the annual monitoring in B and C show individual quadrat results, and points in C for the monthly monitoring show the mean monthly coverage. Lines are beta‐binomial mixed model ±95% CI. Where coverage was zero, it is indicated with an ‘x’ shape. Photo credits: Ben P Harvey and Shigeki Wada.

Understorey E. cava also trended downward (non‐significant on the response scale), reaching zero cover by 2013, briefly reappearing in 2019, and returning to zero from 2020 onward (Figure 2C). Shallower quadrats (2–4 m depth; separate author survey) showed comparable declines (GLMM: −0.17% ± 0.06% cover year⁻¹, p < 0.001). Although cover was initially more variable due to sporadic recruitment, with seasonal peaks up to 6%–8% mean cover in 2015, 2017, and 2018 (up to 30% in individual quadrats), recruitment waned after 2017 with no understorey kelp plants recorded since 2021 (Figure 2C).

3.3. Standing Stock Biomass Surveys

There was a pronounced and consistent decline in the standing stock of E. cava populations from 1980–1981 to 2023–2024 (Figure 3). In the earliest survey (1980–1981), the seasonal peak biomass reached 2.37 kg DW m⁻² (95% CI: 1.98–2.90), with a mean seasonal biomass of 1.69 kg DW m⁻² (1.55–1.85). By 1996–1997, peak biomass had declined to 1.54 kg DW m⁻² (1.44–1.91), representing a ~35% reduction relative to the 1980–1981 baseline, while mean biomass fell to 1.23 kg DW m⁻² (1.11–1.37). A dramatic collapse was evident by 2015–2016, when seasonal peak biomass was estimated as only 0.37 kg DW. m⁻² (0.24–0.50) and mean biomass was 0.27 kg DW. m⁻² (0.15–0.39). These values represent > 80% reductions compared to 1980–1981. In the most recent surveys (2023–2024), no E. cava individuals were recorded, and both peak and mean seasonal biomass were zero across all quadrats, confirming local extirpation of the population.

FIGURE 3.

Seasonal dynamics and long‐term decline of Ecklonia cava standing stock. (A) Points show individual quadrat‐level measurements of dry biomass (kg DW m⁻²), jittered slightly for clarity. Coloured lines represent seasonal trajectories fitted for each survey period using generalised additive models (GAMs) with cyclic cubic splines, and shaded ribbons denote 95% confidence intervals. For 2015–2016, the fitted line is shown as dotted between April and August (months 3.5–8.5) to reflect the absence of survey data during this interval. (B) Peak seasonal biomass for each survey period with 95% bootstrap confidence intervals, estimated from the fitted seasonal trajectories. (C) Mean seasonal biomass across the annual cycle with 95% bootstrap confidence intervals, estimated from the fitted seasonal trajectories to enable comparison among survey periods with different sampling dates. Colours correspond to survey periods (1980–1981 = green; 1996–1997 = blue; 2015–2016 = purple; 2023–2024 = orange).

3.4. Thermal Tolerance and Growth Rates

Juvenile E. cava individuals maintained high maximum photosynthetic efficiency (F _v /F _m  ≈ 0.8) at temperatures up to ~29.5°C (Figure 4A). Beyond this threshold, F _v /F _m declined rapidly to < 0.6 by 31°C and approaching 0.1 by 33°C (Figure 4A). Segmented regression identified a common breakpoint (ψ) at 29.6°C ± 0.14°C (SE). All individuals were photosynthetically non‐viable when temperatures exceeded ~32°C–33°C. Growth trajectories of juveniles differed markedly between temperature treatments (Figure 4B). At 22°C (early summer temperature), thalli increased in surface area by 45.4% ± 6.5% over the 35‐day period. At 28°C (peak summer temperature), growth was strongly suppressed, with only a 12.1% ± 4.2% increase over the same period. Linear mixed‐effects models confirmed significantly lower weekly growth rates at 28°C compared with 22°C (difference in slopes = −0.07 ± 0.02 week⁻¹, p = 0.004). At 28°C, all individuals survived, indicating non‐lethal conditions, yet growth was significantly suppressed.

FIGURE 4.

(A) Thermal limit of Ecklonia cava (as assessed by failure of Photosystem II) during a thermal ramp between 20°C and 33°C. Each point represents the F _v /F _m of a kelp replicate from an individual tank, with the temperature representing the mean temperature of that tank in the 12 h prior to the PAM measurement. Solid line is a segmented regression using the tank as a fixed effect (with 95% CI ribbon). The red line denotes the estimated thermal threshold and the shaded band its ±SE. (B) Mean relative growth (% change in surface area vs. Week 0) across weeks for each tank. Thick lines are model‐estimated means from the mixed effects model and with shading showing 95% CIs. Blue = 22°C (early summer); Red = 28°C (peak summer).

3.5. Fish Surveys on Abundance and Grazing Pressure

During the fish surveys, kelp condition within the bay changed rapidly at sites with extant kelp (Figure 5A). The kelp forest site was dominated by healthy kelp canopies early in summer 2021 but shifted to degraded, sparse plants by October 2021, with complete loss of canopy‐forming kelp thereafter. The mosaic site supported dense, healthy kelp canopies in May 2021, but shifted to degraded, sparse canopies by August 2021 and complete loss of mature plants by September 2021. At both sites, new recruits were recorded after canopy loss but did not lead to recovery. The barren site had exhibited complete loss of canopy‐formers from the start of monitoring in May 2021 and showed no recovery (or recruits) by the end of the study.

FIGURE 5.

Seasonal dynamics of kelp state, herbivorous fish abundance, and relative grazing pressure at three coastal sites from May 2021 to March 2022. (A) The proportion of transects classified as Pristine (healthy adults present; green), Impacted (thinned adults, blue), Extirpated (no adults or recruits, red), or Recruits (only new recruits and no adults; grey) in each month at the Kelp‐forest, Mosaic, and Barren sites, based on belt‐transect surveys. (B) The abundance of herbivorous fish, separated into Browser (blue) and Grazer (red) feeding groups, based on belt‐transect surveys. (C) The relative grazing pressure (bite marks h⁻¹ relative to maximum rate within that site) of herbivorous fish, separated into Browser (blue) and Grazer (red) feeding groups, based on one‐hour video assays. In both (B) and (C), points represent the mean per transect/video, solid lines indicate the site‐level monthly mean, and shaded ribbons denote the ±1 SE for each feeding group. Transect replication was six at the kelp‐forest site, eight at the mosaic site, and seven at the barren site; six one‐hour camera deployments were used during each survey where conditions allowed.

Abundance patterns from UVC counts (Figure 5B) showed browsers ( Calotomus japonicus, adult Siganus fuscescens, Girella punctata ) increasing from May and reaching late‐summer/early autumn maxima, with mean abundance of browsers peaking at 26.3 individuals 100 m⁻² in November 2021 at the kelp forest site, 22.8 in August at the mosaic site, and 26.7 in September at the barren site. Grazers ( Prionurus scalprum , juvenile S. fuscescens , Acanthurus nigrofuscus ) peaked earlier in the warm season, reaching 25.3 individuals 100 m⁻² in October at the kelp forest site, 17.5 in September at the mosaic site, and 11.6 in September at the barren site. Browser presence persisted into early spring across sites, whereas grazer abundance was concentrated in summer to early autumn. Generally, all herbivorous species were observed at all sites, despite the changes in kelp density and habitat structure.

Fish occurrence and grazing pressure appeared mismatched (Figure 5B,C). On kelp, browser activity increased from summer into autumn and early winter, peaking at the kelp forest and mosaic sites in December 2021, and at the barren site in September 2021 (Figure 5C). Grazing on kelp was dominated by Calotomus japonicus across sites (≈95.4% at the mosaic site, 85.3% at the barren site, and 69.4% at the kelp forest site), with secondary contributions from Prionurus scalprum (≈2.0% at mosaic and 10.6% at kelp forest) and Siganus fuscescens (≈11.3% at barren; juveniles ≈16.5% at kelp forest). Within the turf habitat at the barren site, which was monitored throughout in the absence of kelp, grazer herbivory peaked in July 2021, while browser grazing (on the turf) peaked later, in October 2021. Within the turf habitat at the barren site, grazing was primarily by Prionurus scalprum at about 65.6%, with Calotomus japonicus contributing about 15.7% and Acanthurus nigrofuscus about 12.6%. Grazer activity in turf algae areas did occur when both natural and transplanted kelp were present but was lower compared to the browsers, with peaks at the kelp forest site in August 2021, the mosaic site in September 2021, and at the barren site in January 2022 (Figure 5C).

Thermal conditions during grazing closely reflected each species' biogeographic affinity. The tropical rabbitfish Siganus fuscescens and surgeonfish Acanthurus nigrofuscus fed at the warmest temperatures, with lower operating bounds near 25°C and 27°C, respectively. In contrast, the subtropical and temperate species ( Calotomus japonicus , Prionurus scalprum , and Girella punctata ) still grazed under cooler conditions (≈15°C–16°C), enabling foraging later into winter/earlier after winter. During the Kuroshio Large Meander (2017–2024), the onset of suitable grazing temperatures occurred 71 ± 23 days earlier for C. japonicus , P. scalprum , and G. punctata (Wilcoxon, BH‐adjusted, W = 12, p = 0.049), and 18 ± 6 days earlier for S. fuscescens (W = 7, p = 0.048), while A. nigrofuscus showed no detectable change (5 ± 12 days earlier, p = 0.56). The end of the grazing window extended significantly for S. fuscescens (18 ± 7 days later; W = 58.5, p = 0.03) and more moderately for A. nigrofuscus (19 ± 8 days later, W = 41.5, p = 0.07), while C. japonicus , P. scalprum , and G. punctata showed a non‐significant extension (10 ± 6 days later, p = 0.21). Consequently, total grazing durations increased by 86 ± 27 days (W = 55, p = 0.02) across the subtropical and temperate species, and by 35 ± 12 days (W = 61.5, p = 0.01) for S. fuscescens , whereas A. nigrofuscus showed a smaller but statistically significant expansion (25 ± 9 days, W = 42.5, p = 0.02). These results indicate a broad extension of thermally suitable grazing conditions, with earlier onset and delayed cessation across tropical to temperate herbivores, effectively prolonging annual grazing pressure throughout the region (Figure 6).

FIGURE 6.

Suitable grazing windows for the herbivorous fish species, shown in relation to seasonal sea surface temperatures (each year shown separately). Blue indicates temperature‐suitable periods for grazing, and red indicates unsuitable temperatures. Horizontal bars denote the median start and end of the core grazing window. Upper row shows the conditions prior to the Kuroshio Large Meander (2009–2016) and the lower row shows the recent conditions during the Large Meander (2017–2024).

3.6. Distribution Within Izu

Across all coastal regions of the Izu Peninsula that historically supported seaweed forests, contemporary sea surface temperatures are now markedly elevated in 2017–2024 compared to 2009–2016 (Figure 7A) and exceed the lethal threshold of E. cava (~29°C) far more frequently each year (Figure 7B). Relative to the earlier period (2009–2016), the mean annual increase in days ≥ 29°C was 4.19 ± 0.22 days year⁻¹ (SE; n = 193 pixels; 95% CI: 3.76–4.63), with a cumulative rise of 33.53 ± 1.73 days (95% CI: 30.12–36.94) over 2017–2024. These extended exposures likely surpass the recovery capacity of E. cava populations in the region.

FIGURE 7.

(A) Difference in the mean sea surface temperature (SST, °C) in 2017–2024 compared to 2009–2016. (B) Difference in the total number of days > 29°C in 2017–2024 compared to 2009–2016 (indicating kelp lethal temperatures). (C, D) Difference in the annual thermal window suitable for grazing in 2017–2024 compared to 2009–2016 (indicating grazing pressure) for sub‐tropical fishes (C) and tropical fishes (D). (E) Illustrative distribution of seaweed forests in Izu Peninsula based on Hasegawa (2010) representing pre‐meander coverage (green stripe pattern). Points show the locations of qualitative surveys showing the absence of kelp forests both before and after the Large Meander (Blue, ‘Always absent’) and those that were present before, and have since been lost (Red, ‘Extirpated’). The historical kelp distribution in (E) are shown at a fixed distance from the coastline (2.5 km), this is to make them visually easier to see but does not represent their offshore distribution.

Simultaneously, the annual thermal windows suitable for herbivorous fish grazing lengthened markedly (Figure 7C,D). Using pre‐established thermal limits for grazing activity, the subtropical fish species grazing window expanded by 44.68 ± 0.84 days year⁻¹ (95% CI: 43.04–46.32; Figure 7C), while the tropical fish species window extended by 17.80 ± 0.29 days year⁻¹ (95% CI: 17.23–18.37; Figure 7D) between 2017–2024 and 2009–2016.

Qualitative diver re‐surveys of previous kelp forests between 2019 and 2024 from multiple Izu areas (Figure 7E) confirm that areas once covered by dense E. cava and E. bicyclis forests now show complete loss. No substantial forest remnants remain in the southern or eastern bays, and scattered individuals (if any) are likely to persist only in isolated patches.

4. Discussion

Our results document an abrupt, sustained collapse of kelp forests at a sentinel monitoring site and are suggestive of wider functional loss across the Izu Peninsula. Over four decades, the standing stock of Ecklonia cava in Shimoda declined by more than 80%, culminating in complete loss and local extirpation by 2023–2024. Parallel losses in canopy cover and the absence of recruits indicate a transition from once‐dense forests to persistent barren and turf‐dominated states. This collapse unfolded under rapidly intensifying ocean warming. Together, these results indicate that recurrent MHWs have repeatedly exceeded the species' thermal safety margin, likely contributing to catastrophic canopy loss. Beyond thermal stress, sustained herbivory by warm‐water fishes now operates most of the year; together with chronic and acute warming, this pressure may contribute to canopy loss and suppress recovery during the limited cool periods. Although these datasets cannot separate the precise contribution of each driver, their combined evidence supports an interaction between intensifying thermal stress and expanded herbivory. The Izu Peninsula therefore exemplifies how physical and biological stressors can interact to push coastal ecosystems beyond ecological thresholds, promoting a shift from kelp‐dominated reefs to structurally simplified barren states within less than a decade.

4.1. Ocean Warming, MHWs, and the Physiological Limits of Ecklonia cava

The four‐decade thermal record from Shimoda showed a steady rise in baseline temperatures prior to 2017, followed by an abrupt acceleration in warming and MHW activity during the Kuroshio Large Meander. Before 2017, coastal SST trends around Japan were generally low (typically < 0.25°C decade⁻¹), and the Izu Peninsula warmed at only 0.14°C decade⁻¹. In contrast, during the Large Meander the regional warming rate in Izu increased more than twenty‐fold, reaching 3.03°C decade⁻¹, with a parallel surge in cumulative MHW intensity. This shift reflects the influence of mesoscale circulation. When the Kuroshio heads northward toward the Izu–Tokai coast, warm offshore waters are advected into nearshore bays, producing persistent positive anomalies and suppressing seasonal cooling (Hirata et al. 2025; Sugimoto 2025). This configuration effectively transformed the Izu Peninsula into a long‐term natural experiment for sustained heat stress; a preview of future mid‐latitude thermal regimes.

Physiological assays confirmed that these thermal conditions surpassed the tolerance limits of E. cava, with physiological decline observed at ~29.5°C and total functional failure above 32°C. These thresholds coincide with in situ temperature peaks recorded during recent MHWs, indicating that ambient conditions alone can induce mortality. Even sub‐lethal temperatures (27°C–28°C) suppressed juvenile growth by > 70% relative to typical early summer conditions (22°C), constraining tissue replacement and recovery. This pattern aligns with previous observations from the same area, where optimal photosynthesis occurred at ~25°C–27°C and sharp declines were observed above ~29°C (Serisawa, Yokohama, et al. 2001), reinforcing the similarly narrow thermal operating window for local E. cava populations. Together, these responses imply a very limited thermal safety margin for juvenile E. cava in Shimoda; reduced blade turnover at these temperatures restricts tissue replacement and predisposes canopies to loss. Comparable sub‐lethal suppression preceding mortality has been documented in other kelp systems (e.g., E. radiata , Saccharina latissima), where brief exposures only a few degrees above local maxima precipitate canopy collapse (Filbee‐Dexter et al. 2020; Wernberg et al. 2016).

These recurrent heat events have compressed the seasonal window for regrowth. Historically, E. cava regenerated blades and reproductive tissue through the cool months, achieving peak biomass by early to mid‐summer. Under the new thermal regime, elevated autumn and winter temperatures delay recruitment and shorten the period before the next summer extreme, producing a temporal ‘squeeze’ that mirrors other global warming hotspots (Smale 2020; Vergés et al. 2014). The Izu case thus represents a physiological bottleneck at the species' warm distributional limit, where chronic and acute warming act in concert to drive local extirpation, consistent with projections of climate‐driven range shifts and contraction of major kelp species along the Japanese coast (Sudo et al. 2020).

4.2. Thermal Expansion of Herbivory and Loss of Seasonal Refuge

Long‐term warming has also extended the activity window of tropical and subtropical herbivorous fishes such as Siganus fuscescens , Calotomus japonicus , and Prionurus scalprum . Our surveys show that these species now occur in high densities across all habitat states (from healthy kelp forest to mosaic and barren), with feeding intensity tightly linked to temperature. This seasonal broadening of feeding activity represents a clear manifestation of temperate reef tropicalisation, where tropical and subtropical herbivores increasingly dominate and restructure formerly temperate ecosystems (Choi et al. 2024; Nakamura et al. 2013).

Bite‐rate analyses identified species‐specific lower thresholds for grazing > 25°C for tropical taxa and ~15°C–16°C for subtropical and temperate species. When projected from the pre‐meander to Large Meander conditions, both the timing and duration of grazing shifted markedly during the Kuroshio Large Meander. Among subtropical and temperate species ( Calotomus japonicus , Prionurus scalprum , Girella punctata ), the onset of suitable temperatures advanced by ~71 days, and the end of the season was delayed by ~10 days, leading to an ~86‐day expansion of their primary grazing window. For the tropical browser Siganus fuscescens , onset advanced by ~18 days and cessation was delayed by ~18 days, resulting in a ~35‐day extension of the grazing period. Although Acanthurus nigrofuscus exhibited a smaller response (~25‐day increase), these changes collectively mean that grazing now overlaps nearly the entire window in which E. cava attempts to rebuild biomass after summer die‐back. Browsers such as S. fuscescens perforate blades and reduce photosynthetic area, while C. japonicus removes whole thalli, severing stipes and causing rapid loss. Sequential feeding by these guilds results in near‐continuous biomass removal; chronic browsing through summer–autumn followed by destructive scraping in winter. This eliminates the brief cool‐season respite that historically allowed canopy recovery—a pattern consistent with warming‐driven extensions of herbivore activity in this and other western boundary current systems (Kumagai et al. 2018; Sudo et al. 2022; Vergés et al. 2016).

The expansion of herbivory reflects not only greater fish abundance but temperature‐driven behaviour. Warmer conditions lower locomotion costs, shorten digestion times, and raise activity budgets (via Q10 effects), sustaining higher daily feeding and wider movement across reefs (Clarke and Johnston 1999; Volkoff and Rønnestad 2020). Mild winters also boost juvenile rabbitfish survival, producing dense autumn cohorts that graze during the kelp reproductive season (Vergés et al. 2016; Zarco‐Perello et al. 2019). As canopies thin, encounter rates on remnant patches rise (edge concentration), exposing meristems to repeated cropping by herbivores (Andrew and Jones 1990; Barrientos et al. 2022). Superimposed on recurrent warm‐water intrusions during the Kuroshio large‐meander phase, these physiological, demographic, and structural changes drive consumer‐driven recovery failure, whereby recruits and small thalli are removed faster than they can replace lost canopy. Once kelp density drops below a threshold, per‐capita grazing intensifies, accelerating deforestation and stabilising barren states (Wernberg et al. 2016; Zarco‐Perello et al. 2021).

4.3. Compound Stressors and the Emergence of a Stable Barren Regime

The synchrony between thermal and biotic pressures explains the abrupt and persistent shift observed in the study site. Occasional reappearances of kelp recruits (at least until 2022) demonstrate that propagules remained present, but continuous grazing prevented maturation into adult stands. The absence of refuges eliminates local spore supply, while turf algae rapidly occupy exposed rock, further inhibiting recruitment (Burek et al. 2018). These self‐reinforcing processes—chronic grazing, turf proliferation, and reproductive failure—characterise regime shifts from forested to barren states (Filbee‐Dexter and Wernberg 2018). Although our observational data do not formally test for alternative stable states or tipping‐point dynamics, the persistence of barren habitat despite occasional recruitment is consistent with reduced resilience and feedbacks that can maintain degraded reef states.

Similar compound stressor dynamics are emerging throughout warming western boundary current systems, where escalating heat exposure and tropical herbivore incursions push temperate reefs beyond ecological thresholds. The Izu Peninsula exemplifies these processes, with Kuroshio‐driven thermal anomalies likely magnifying grazing impacts and contributing to the persistence of barren states. This trajectory parallels transformations in other rapidly warming boundary current‐influenced systems, including the East Australian and Agulhas systems (Krumhansl et al. 2016; Vergés et al. 2014), as well as kelp declines in western Australia associated with the poleward‐flowing Leeuwin Current (Wernberg et al. 2016), suggesting that kelp loss in this region is part of a wider Indo‐Pacific reorganisation. The functional implications are far‐reaching, as loss of kelp forest can lead to simplified habitat structure, reduced biodiversity, and diminished carbon sequestration, nutrient cycling, and shoreline protection. Consistent with this local collapse, peninsula‐wide mapping (Figure 7) revealed that nearly all areas with former kelp‐forest now experience prolonged periods above the lethal threshold of E. cava and extended herbivore activity windows, indicating that these conditions are pervasive across the Izu coastline. Critically, the intensification of thermal stress and herbivory mirrors global patterns observed throughout warming hotspots globally, such as Korea, Taiwan, southern Australia, South Africa and the Mediterranean, where kelp forests and other canopy‐forming algae have been replaced by low‐productivity turf or coral assemblages as tropical herbivores expand poleward (Filbee‐Dexter and Wernberg 2018; Krumhansl et al. 2016; Vergés et al. 2016; Zarzyczny et al. 2024). The shift thus represents not merely a change in community composition but a fundamental reconfiguration of ecosystem processes and services (Eger et al. 2023), with cascading consequences for coastal fisheries, blue carbon potential, and socio‐ecological resilience.

The persistence of barren habitat does not necessarily imply irreversible collapse, but recovery from such depauperate states can be challenging, even once ideal conditions return (Filbee‐Dexter and Scheibling 2014; Ling et al. 2015). Recovery is likely to depend on active restoration efforts that combine spore seeding with herbivory management (Eger et al. 2022), and the success of these efforts will only be maximised if they are implemented at the appropriate time and under suitable conditions (e.g., White et al. 2026). In our example, if cooler periods create temporary recovery windows, management actions that protect remnant or recovering patches, reduce grazing pressure where feasible, and target restoration during periods of lower thermal stress may increase the likelihood of kelp persistence and recovery (Eger et al. 2022). However, as the duration and frequency of such recovery windows are likely to decrease under continued warming, and MHWs themselves are becoming a persistent problem, the long term persistence of restored habitats will likely be dependent on sustained monitoring, adaptive restoration strategies (e.g., herbivory control before and during a MHW; Hopf et al. 2025) and efforts to increase the thermal tolerance of restored populations through assisted adaptations (Layton et al. 2022; Smith et al. 2023).

The processes described here are most likely to apply to warm‐edge kelp systems where canopy‐forming kelps occur close to their upper thermal limits (Shi et al. 2026) and where ocean warming extends the seasonal activity window of warm‐adapted herbivores (Vergés et al. 2016). Such conditions may be particularly relevant in boundary‐current regions and other warming transition zones, where rapid changes in thermal exposure can coincide with tropicalisation of consumer assemblages. However, the relative importance of warming, herbivory, and recruitment limitation will vary among systems depending on local thermal regimes, grazer assemblages, habitat structure, taxon‐specific thermal tolerance and life history, propagule supply, and connectivity with cooler source populations. Thus, while the Izu Peninsula provides a sentinel example of kelp decline under rapid warming and expanded grazing, these dynamics are likely to be strongest where physiological stress and sustained herbivory overlap in space and time.

5. Conclusions

The collapse of Ecklonia cava forests in Shimoda is consistent with a cascading interaction between ocean warming, intensifying marine heatwaves, and the thermal expansion of herbivory. Chronic exposure to temperatures approaching or exceeding experimentally‐derived physiological thresholds likely weakened kelp populations, while sustained grazing may have reduced opportunities for recovery, contributing to the persistence of barren reef conditions characteristic of temperate‐reef tropicalisation. This collapse is consistent with a threshold‐like response in which thermal stress and grazing pressure together reduced ecosystem resilience and promoted an abrupt shift towards a barren state. By integrating multi‐decadal SST data, long‐term kelp biomass records, herbivore behaviour, and experimental physiology, this study provides multiple converging lines of evidence linking climate forcing, biological response, and ecological outcome within a single coastal ecosystem. The resulting framework, linking canopy trajectories, thermal niches in grazing, and long‐term climatic context, offers a transferable approach for diagnosing and managing similar collapses elsewhere. Yet the prospects for rapid recovery in Izu remain limited. Given that global and Western Pacific SSTs are projected to continue rising over the coming decades, further declines of temperate kelp forests in Japan are not only plausible but already unfolding. These trajectories are likely to be compounded by other accelerating stressors, including nutrient depletion, coastal development, sedimentation, and the poleward expansion of warm‐adapted herbivores, which together will narrow the climatic space in which canopy‐forming kelps can persist. Shimoda's reefs therefore represent both a warning, in terms of the rapid loss of a foundation species once physiological limits are exceeded and biological pressures intensify, and a model, for how integrated, cross‐disciplinary science can reveal the mechanisms driving ecosystem shifts. Understanding and acting upon these mechanisms will be essential if the remaining temperate kelp forests of the Western Pacific are to persist under accelerating ocean change.

Author Contributions

Ben P. Harvey: conceptualization, methodology, investigation, formal analysis, validation, supervision, funding acquisition, visualization, project administration, resources, writing – original draft, writing – review and editing. Hiroki Takenaka: investigation, writing – review and editing. Mariko Shiel: investigation, writing – review and editing. Joshua M. Heitzman: formal analysis, writing – review and editing, investigation. Guinther Mitushasi: writing – review and editing, investigation. Layla Iijima: investigation, writing – review and editing. Shigeki Wada: conceptualization, writing – review and editing, funding acquisition, investigation. Takeo Hama: investigation. Nathan G. King: writing – review and editing. Dan A. Smale: writing – review and editing. Sylvain Agostini: conceptualization, investigation, supervision, funding acquisition, writing – review and editing, methodology, resources.

Funding

This work was supported by the Japan Society for the Promotion of Science (JSPS) Core‐to‐Core Program (Grant JPJSCCA20210006) and Grant‐in‐Aid for Scientific Research (B) (Grant 23K26924) to B.P.H., S.A. and S.W. B.P.H. was also supported by the Organization for DESIGN THE FUTURE of University of Tsukuba in Japan. Financial support was also provided through the MEXT (Ministry of Education, Culture, Sports, Science and Technology, Japan) initiative for Educational and Research Organization Reform, under the program entitled ‘Construction of a University‐wide, Cross‐disciplinary Research Promotion Platform for Solving Global‐scale Challenges.’. D.A.S. was supported by a UKRI Future Leaders Fellowship (MR/X023214/1).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Comparison of satellite‐derived sea surface temperature and in situ temperature measurements at Shimoda. (A) Relationship between daily OISST sea surface temperature and daily mean in situ temperature recorded by HOBO pendant loggers at the study site (2015‐09‐24 to 2016‐07–04). The dashed line indicates the 1:1 relationship. (B) Seasonal comparison of daily mean OISST SST and in situ HOBO temperatures across the logger deployment period. Satellite‐derived SST tracked the seasonal pattern of in situ temperature well, although OISST values were generally higher than in situ measurements, with a median difference of ~0.9°C.

Figure S2: Japan‐wide sea surface temperature and marine heatwave trends across 1982–2024. Spatial estimates of (A) sea surface temperature (SST) trends and (B) cumulative marine heatwave intensity (CI) trends across the wider Japanese shelf region from 1982 to 2024.

Figure S3: Location of fish assemblage, grazing, and kelp‐state survey sites. Map showing the three shallow rocky‐reef survey sites used for monthly fish assemblage, grazing, and habitat‐state surveys from May 2021 to March 2022. Sites represent contrasting kelp states: kelp forest, mosaic kelp/barren habitat, and barren reef.

Figure S4: Raw bite‐rate data from video grazing assays. Raw bite rates recorded during one‐hour video assays across kelp‐forest, mosaic, and barren reef sites.

Table S1: Summary of datasets used in the study, including temporal coverage, spatial scale, sampling approach, response variables, and ecological inference supported.

Data Availability Statement

The datasets generated and/or analysed during the current study, together with the code used to generate and analyse them, are available in the Zenodo repository (DOI): 10.5281/zenodo.22260295.