Quantifying changes in fish population stability using statistical early warnings of regime shifts

This data package describes long-term trends in metrics describing population stability and used as statistical early warnings of regime shifts in 29 fish species that inhabit the San Francisco Bay-Delta in central California, USA. Metrics used in this study include spatial synchrony, temporal coefficient of variation (CV), and lag-1 temporal autocorrelation. Trends were measured using ordinary least squares linear regression. These derived data were developed from abundance (as CPUE) time series based on three long-term fish monitoring studies included in https://doi.org/10.6073/pasta/a29a6e674b0f8797e13fbc4b08b92e5b; the Fall Midwater Trawl Survey, Delta Juvenile Monitoring Program, and Bay Study. Selected data were from fall months (September to December) in 1980-2023, from midwater trawl and beach seine surveys for which sampling effort (e.g., tow volume) was recorded. Data on fish exceeding maximum length thresholds for age-0 fish were discarded, except for white sturgeon, where the maximum length threshold corresponded to approximately 10 years of age, the onset of reproductive maturity. Observations from different sampling stations were aggregated into 10 sub-regions (South San Francisco Bay, Central San Francisco Bay, San Pablo Bay, Napa River, Suisun Bay, Delta Confluence, South Delta, North Delta, San Joaquin River, Sacramento River, and midwater trawl samples and beach seine samples were considered separately because the methods sample distinct habitat types. Combinations of sub-region and sampling method were considered distinct spatial units. EWI metrics were measured in 5-year rolling windows to permit assessment of changes over time. The temporal CV and lag-1 autocorrelation were measured on individual spatial unit time series, ignoring windows with >1 year of missing data. The coefficient of variation divides the standard deviation by the mean. Lag-1 autocorrelation was measured as Pearson correlation. Spatial synchrony was measured across spatial units, ignoring spatial units with >1 year of missing data, and ignoring rolling windows where <3 spatial units had sufficient data. Spatial synchrony was measured as the mean of pairwise Spearman correlations. Trends in EWI metrics were measured only when there were at least 5 rolling window measurements spanning at least 10 years.

本数据集描述了栖息于美国加利福尼亚州中部旧金山湾-三角洲区域的29种鱼类的种群稳定性相关指标的长期变化趋势，此类指标可作为种群状态转变（regime shifts）的统计预警信号。本研究采用的指标包括空间同步性（spatial synchrony）、时间变异系数（temporal coefficient of variation, CV）以及滞后1阶时间自相关（lag-1 temporal autocorrelation）。指标趋势通过普通最小二乘线性回归（ordinary least squares linear regression）计算得到。

本派生数据集基于三项长期鱼类监测研究的丰度（以单位捕捞努力量渔获量（Catch Per Unit Effort, CPUE）表示）时间序列构建，相关原始数据收录于https://doi.org/10.6073/pasta/a29a6e674b0f8797e13fbc4b08b92e5b；三项研究分别为秋季中层拖网调查（Fall Midwater Trawl Survey）、三角洲幼鱼监测计划（Delta Juvenile Monitoring Program）以及海湾研究（Bay Study）。本次选取的数据为1980-2023年秋季（9月至12月）的中层拖网与海滩围网（beach seine）调查数据，且所有采样均记录了采样努力量（如拖网体积）。

体长超过0龄鱼最大体长阈值的个体数据均被剔除，但白鲟（white sturgeon）除外——白鲟的最大体长阈值对应约10年鱼龄，即其生殖成熟起始年龄。研究将不同采样站位的观测数据聚合为10个子区域：南旧金山湾、中旧金山湾、圣巴勃罗湾、纳帕河、苏森湾、三角洲汇流区、南三角洲、北三角洲、圣华金河、萨克拉门托河；由于中层拖网与海滩围网的采样方法对应不同的生境类型，故将两类样本分开处理，子区域与采样方法的组合被视为独立的空间单元。

预警指标（Early Warning Indicators, EWI）通过5年滑动窗口进行计算，以实现随时间变化的趋势评估。针对单个空间单元的时间序列，计算时间变异系数与滞后1阶自相关，当滑动窗口内缺失数据时长超过1年时则忽略该窗口。变异系数为标准差与均值的比值，滞后1阶自相关以皮尔逊相关系数（Pearson correlation）进行计算。空间同步性的计算覆盖所有空间单元，但会剔除缺失数据时长超过1年的空间单元，以及有效空间单元数量不足3个的滑动窗口；空间同步性以两两斯皮尔曼相关系数（Spearman correlations）的均值表示。仅当滑动窗口的有效测量次数不少于5次且时间跨度至少为10年时，才会计算预警指标的长期变化趋势。