Rising catch variability preceded historical fisheries collapses in Alaska

Statistical indicators such as rising variance and rising skewness in key system parameters may provide early warning of ‘‘regime shifts’’ in communities and populations. However, the utility of these indicators has rarely been tested in the large, complex ecosystems that are of most interest to managers. Crustacean fisheries in the Gulf of Alaska and Bering Sea experienced a series of collapses beginning in the 1970s, and we used spatially resolved catch data from these fisheries to test the predictions that increasing variability and skewness would precede stock collapse. Our data set consisted of catch data from 14 fisheries (12 collapsing and two non-collapsing), spanning 278 cumulative years. Our sampling unit for analysis was the Alaska Department of Fish and Game statistical reporting area (mean n for individual fisheries ¼ 42 areas, range 7–81). We found that spatial variability in catches increased prior to stock collapse: a random-effects model estimating trend in variability across all 12 collapsing fisheries showed strong evidence of increasing variability prior to collapse. Individual trends in variability were statistically significant for only four of the 12 collapsing fisheries, suggesting that rising variability might be most effective as an indicator when information from multiple populations is available. Analyzing data across multiple fisheries allowed us to detect increasing variability 1–4 years prior to collapse, and trends in variability were significantly different for collapsing and non-collapsing fisheries. In spite of theoretical expectations, we found no evidence of pre-collapse increases in catch skewness. Further, while models generally predict that rising variability should be a transient phenomenon around collapse points, increased variability was a persistent feature of collapsed fisheries in our study. We conclude that this result is more consistent with fishing effects as the cause of increased catch variability, rather than the critical slowing down that is the driver of increased variability in regime shift models. While our results support the use of rising spatial variability as a leading indicator of regime shifts, the failure of our data to support other model-derived predictions underscores the need for empirical validation before these indicators can be used with confidence by ecosystem managers.