Impacts of marine heatwaves on top predator distributions are variable but predictable

The study addresses how short-term extreme ocean warming events—marine heatwaves (MHWs)—affect the distributions of highly mobile marine top predators. While long-term warming impacts are well-studied, episodic events like MHWs can cause catastrophic ecosystem and socio-economic outcomes. Prior work has noted species redistribution under long-term warming and El Niño, but there is limited multi-species, multi-event understanding for recent unprecedented MHWs. Because conditions during MHWs can resemble end-of-century averages, their impacts can preview future redistribution. Observational evidence suggests poleward/vertical shifts, but data are often patchy and single-species/event focused. The authors aim to quantify and compare distributional responses across 14 predator species and four major North Pacific MHWs (2014, 2015, 2019, 2020), and to assess transboundary jurisdictional shifts across the US, Mexico, Canada, and high seas. The Northeast Pacific is an ideal testbed due to extreme MHWs overlapping biodiversity hotspots and major foraging grounds where many predators are sensitive to climate perturbations.

The paper situates MHWs within broader climate extremes that impact ecosystems and human well-being. It notes recent record-setting MHWs (e.g., 2012 Northwest Atlantic, 2015–16 Tasman Sea, 2013–16 Northeast Pacific) and highlights that previous studies often relied on opportunistic sightings, surveys, or tagging that document poleward or vertical shifts but lack standardized, multi-species, multi-event comparisons. The literature underscores variability in MHW drivers and characteristics, implying species responses may vary among events. The authors argue for statistical, multivariate species distribution models (SDMs) to interpolate across space, time, and taxa and to capture complex environmental influences beyond temperature alone during MHWs.

• Data: Telemetry data (2000–2010) for 14 top predators (seabirds, mammals, turtles, tunas, sharks), including Tagging of Pacific Predators (TOPP) and private datasets. Additional telemetry sources included BirdLife’s Seabird Tracking Database and other programs. Species were grouped geographically (Northern, Coastal, Southern) based on August–October observations.
• Environmental covariates: Primary productivity, oxygen (200 m), sea surface temperature (SST) and spatial SD of SST, sea level anomaly, eddy kinetic energy, mixed layer depth, chlorophyll-a, and day of year. Variables resampled to 0.25° resolution. Anomalies during MHWs were calculated relative to August–October 2000–2020 means.
• Modeling: Boosted regression trees (BRTs; binomial) modeled probability of presence as a function of environment. Pseudo-absences were generated 1:1 with presences, matched by date. BRT settings: bag fraction 0.6, tree complexity 3, learning rate 0.0001–0.00001 ensuring ≥2000 trees. Models produced daily habitat suitability (0–1) for 2000–2020 within a minimum convex hull of training data. Variable importance was assessed.
• Core habitat classification: Continuous suitability reclassified to daily binary core habitat using species-specific thresholds defined as suitability ≥ median (50th percentile) of predictions at true presences. The 50% threshold was chosen to avoid complete loss during extreme events.
• Impact metrics (computed for August–October of each MHW year, relative to 2000–2020 Aug–Oct mean): (1) Displacement distance and direction of the core habitat center of gravity (km and degrees), (2) Range extent change via interquartile range product (north–south × east–west) indicating compression (negative) or expansion (positive), (3) Core habitat area change (km²; percent change). Metrics were averaged over each event.
• Jurisdictional analysis: Daily core habitat area was calculated within EEZs (US, Mexico, Canada) and high seas (Marineregions.org v11). For each species and MHW, Aug–Oct average anomalies (percent change vs baseline) were computed per jurisdiction, and largest gains/losses summarized.
• Validation and uncertainty: Extensive validation with independent datasets (e.g., fisheries observers, surveys, citizen science, tagging) and multiple cross-validation approaches. Temperature-only models were compared to multivariate models. Sensitivity assessed by fitting 20 replicate BRTs to different 75% subsamples and propagating to metrics (means, SEs, CVs). All analyses in R 4.0.4.

• Species responses to MHWs were highly variable across species and events, with impacts ranging from near-total habitat loss (e.g., bluefin tuna, 2015) to two-fold habitat increases (e.g., California sea lion, 2019).
• Direction and magnitude of displacement differed within and among MHWs. In 2019–2020, sooty shearwaters had large displacement distances (536 km in 2019; 721 km in 2020), while most species had relatively lower displacement.
• Coastal species shifted northwest during 2014–2015, but several (bluefin and albacore tunas, blue and mako sharks) shifted southeast in 2019–2020, likely tracking emergent cool-water refugia along the southern US and Mexican coasts.
• Range dynamics: Sooty shearwaters and California sea lions showed reversal between events—range compression in 2014–2015 (22–64% decrease) and large range expansion in 2019–2020 (335–377% increase). Elephant seals and sea lions lost habitat in 2014 (29% and 65% decrease, respectively) but gained habitat in 2019–2020 (elephant seals +71–46%; sea lions +117–158%).
• Habitat area changes in 2019–2020: While most species lost habitat, elephant seals, salmon sharks, California sea lions, and yellowfin tuna gained habitat.
• Jurisdictional shifts: The US EEZ gained predator habitat during each MHW (largest +10% in 2015; smallest +2% in 2020). Mexico lost habitat in 2014, 2015, 2020 (largest −8% in 2015). Canada and the high seas generally gained and lost habitat, respectively. Notable species-level shifts included 31% (yellowfin) and 22% (albacore) of habitats shifting into US waters (2015), 39% of white shark habitat shifting into the US EEZ (2019), 8% of leatherback turtle habitat into the US EEZ (2019), and 14% of salmon shark habitat into the Canadian EEZ (2020).
• Environmental drivers: Most severe impacts occurred where temperature anomalies were highest. Differences among MHWs in anomaly depth, oxygen solubility, and productivity contributed to species-specific responses (e.g., deeper warm anomalies in 2014–2015 linked to oxygen decline; shallower anomalies in 2019–2020 had weaker biogeochemical effects).
• Model performance: Multivariate models outperformed temperature-only models on novel validation data (median AUC 0.8 vs 0.6; t-test p < 0.05). Independent datasets corroborated predicted shifts, though performance varied with data domain alignment.

The standardized, multi-species, multi-event framework reveals that MHW impacts on top predator distributions are diverse yet predictable. Species exhibited markedly different displacement, range, and habitat area responses across events, reflecting event-specific physical and biogeochemical conditions and species-environment relationships. The most severe impacts aligned with regions of strongest thermal anomalies, but responses could not be explained by temperature alone; incorporating oxygen, productivity, and other variables improved predictive skill and ecological inference. Cross-jurisdictional habitat shifts were substantial and dynamic, implying that governance must adapt rapidly across national boundaries during extreme events. The operationalization of daily species distribution predictions demonstrates feasibility for real-time dynamic ocean management, and the results highlight the value of early warning systems that forecast ecological responses to anticipated MHWs.

This study provides a comprehensive, standardized assessment of how multiple marine heatwaves affect the distributions of 14 top predators in the Northeast Pacific, demonstrating highly variable but predictable responses. By leveraging multivariate boosted regression tree models trained on extensive telemetry, the authors quantify displacement, range changes, habitat area changes, and transboundary shifts. They show that management cannot assume consistent impacts across events and species, but can exploit predictability to respond in near real-time. As a proof of concept, they operationalized daily distribution predictions (Top Predator Watch) suitable for integration into dynamic ocean management. Future directions include developing ecological early warning systems that forecast species distributions in response to MHWs, continued observation to update models in real time, and coordinated transnational management to address episodic redistributions and associated human-wildlife conflicts.

The correlative SDMs capture species-environment associations but do not explicitly model species traits (physiology, movement syndromes, life histories) or behaviors (e.g., reproductive migrations), nor ecological processes such as prey availability, interspecific interactions, population structure, or site fidelity. Core habitat thresholds may influence estimates during extreme events. Validation data sets varied in spatial domain and type relative to telemetry, affecting performance for some species. The models estimate fundamental niches within the North Pacific training domain and may not capture all context-specific drivers outside this domain.