SST models fitted to 1984–2013 predict kelp canopy in central and southern California and Washington in 2017–2025, but northern California canopy stayed at 9% of prediction

Question: does sea surface temperature (SST) alone predict floating kelp canopy loss in the 2014-2016 northeast Pacific heatwave and the recovery after it? Data: Santa Barbara Coastal LTER Landsat kelp canopy (30 m pixels, quarterly, 1984 to mid-2026, Baja California Sur to Washington) and NOAA OISST v2.1 monthly SST. Method: we grouped pixels into coastal segments of about 25 km in 7 latitude regions and used log annual maximum canopy. We fitted segment models with region-specific SST anomaly slopes on 1984-2013 only, selected the model by leave-one-year-out cross-validation, and then compared observed with predicted canopy for 2014-2025. Years were the units of inference. Results (64 segments): in northern California (37.8-42 N, bull kelp), canopy in 2017-2025 was 9.4% of the SST prediction (log residual -2.37, year bootstrap 95% interval -2.65 to -2.03), below all 22 nine-year windows of 1984-2013 cross-validated residuals; all 4 segments were below their own 5th percentile. In northern Baja California, canopy followed the model in 2017-2019 but fell to 7% of prediction in 2023-2025, below all training windows. In southern and central California and Washington, 2017-2025 residuals (-0.13, -0.06, +0.02) were inside training variation. In 2014-2016, giant kelp regions lost less canopy than the model predicted, but northern California lost more. Five sensitivity analyses (segment filter, pixel coverage, July-September canopy) gave the same conclusions for these regions. Oregon and Baja California Sur had one segment each, and their results were not robust. Limitations: SST is a proxy and the model explains little year-to-year variation; nutrients, waves and grazers are not in the model, so the study shows where canopy departs from temperature control but not the cause.

giant kelp · bull kelp · Macrocystis pyrifera · Nereocystis luetkeana · marine heatwave · Landsat · sea surface temperature · OISST · recovery · hysteresis · California Current · out-of-sample prediction · regime shift