Ecological forecasts for marine resource management during climate extremes

Forecasting weather has become commonplace, but as society faces novel and uncertain environmental conditions there is a critical need to forecast ecology. Forewarning of ecosystem conditions during climate extremes can support proactive decision-making, yet applications of ecological forecasts are still limited. We showcase the capacity for existing marine management tools to transition to a forecasting configuration and provide skilful ecological forecasts up to 12 months in advance. The management tools use ocean temperature anomalies to help mitigate whale entanglements and sea turtle bycatch, and we show that forecasts can forewarn of human-wildlife interactions caused by unprecedented climate extremes. We further show that regionally downscaled forecasts are not a necessity for ecological forecasting and can be less skilful than global forecasts if they have fewer ensemble members. Our results highlight capacity for ecological forecasts to be explored for regions without the infrastructure or capacity to regionally downscale, ultimately helping to improve marine resource management and climate adaptation globally.

Methods
Summary
We configure two existing resource management tools, originally configured to use observed (historical) ocean temperatures, to a forecasting system and conduct a retrospective forecast to test their skill. We first conducted a retrospective forecast using global forecasts (73 ensemble members) across the full historically available period (1981-2020) – termed the ‘Global’ model. Global forecasts of monthly sea surface temperature were obtained from the North American Multimodel Ensemble (NMME; Table S1; https://www.cpc.ncep.noaa.gov/products/NMME/).

We then compared the performance of three forecast configurations: First, we used global forecasts (73 ensemble members) across a reduced historical period (1981-2010) - termed the ‘Global Full Ensemble’. Second, we used forecasts regionally downscaled (3 ensemble members) to the CCE for the same reduced historical period (1981-2010) - termed the ‘Downscaled Ensemble’. Third, we used a reduced subset of the global forecasts (3 ensemble members) for the same reduced historical period (1981-2010) - termed the ‘Global Reduced Ensemble’.  

All forecasts are compared to SST observations, extracted from a CCE regional reanalysis. This reanalysis is based on the Regional Ocean Modeling System (ROMS) and covers the west coast of the U.S. (30-48˚N, 134-115.5˚W) with 0.1 degree (~10 km) horizontal resolution and 42 terrain-following vertical levels.

Case Study 1: Habitat Compression Index 
The Habitat Compression Index (HCI) is a regionally resolved measure of cool thermal habitat along the U.S. West Coast; the index presented here monitors surface water conditions off California (35-40°N). The HCI is used to assess the degree to which upwelling habitat (indicated by cool water) is compressed against the coast, as nutrient-rich upwelled waters attract whales seeking enhanced foraging opportunities. The HCI was calculated as the number of grid cells with SST lower than a monthly SST threshold within 150 km of the coastline. The HCI was normalized by the total number of grid cells of the 150 km domain to scale values from 0 to 1.  Monthly SST thresholds are the mean monthly SST from 1981-2010 from the coast to 75 km offshore. Low HCI values represent high compression, or reduction of cool thermal habitat, and are the primary interest to resource managers tasked with mitigating whale entanglement risk. The long-term mean of the HCI is used to identify a high compression event (i.e. values below the mean.

Case Study 2: TOTAL Tool
The Temperature Observations to Avoid Loggerheads (TOTAL) tool monitors anomalously high SST in the Southern California Bight (31-34°N, 120-116°W) as an indicator of turtle bycatch risk and to recommend potential implementation of a fishery closure. TOTAL was calculated as the six-month rolling mean of SST anomalies in the Southern California Bight domain. The spatial closure is potentially enacted during three months of the year (June, July, August) based on SSTA of the preceding six months. If SSTA exceeds a threshold, calculated as the minimum monthly anomaly value preceding three historical closure periods (Aug 2014, Jun-Aug 2015, & Jun-Aug 2016), a closure is recommended. 

Skill assessment
Forecast skill of each management tool was assessed by comparing observed and forecast values using three metrics: (1) correlation coefficient, which provides a statistical measure of the strength of a linear relationship between observed and forecast values; (2) forecast accuracy, which indicates the fraction of correct forecasts; and (3) the Symmetric Extremal Dependence Index (SEDI) which has several properties that make it well suited to quantifying skill for rare events. Details and equations for metrics are described in the manuscript.

天气预报已日趋普及，但随着人类社会面临新型且不确定的环境状况，生态预报的迫切需求日益凸显。在极端气候事件期间对生态系统状况进行预警，可为前瞻性决策提供支撑，但当前生态预报的应用仍较为有限。本研究展示了现有海洋管理工具向预报配置转型的可行性，并可提前12个月生成具有预报技巧的生态预报。该类管理工具通过海洋温度异常值来缓解鲸类渔具缠绕和海龟兼捕问题，研究证实，此类预报可对极端气候事件引发的人-野生动物冲突发出预警。此外，本研究还表明，区域降尺度预报并非生态预报的必需手段，若集合成员数较少，区域降尺度预报的技巧甚至可能低于全球预报。研究结果凸显了，对于缺乏区域降尺度基础设施与能力的地区而言，可探索开展生态预报，最终助力全球海洋资源管理与气候适应工作。

研究方法
方法概述
我们将两款原本基于观测（历史）海温运行的现有资源管理工具改造为预报系统，并通过回溯预报测试其预报技巧。首先，我们利用全球预报（73个集合成员）在完整历史可用时段（1981-2020年）开展回溯预报，该配置被称为"Global"模型。月度海表温度（Sea Surface Temperature, SST）全球预报数据来自北美多模式集合（North American Multimodel Ensemble, NMME；附表S1；https://www.cpc.ncep.noaa.gov/products/NMME/）。

随后我们对比了三种预报配置的性能：其一，在缩短后的历史时段（1981-2010年）使用全球预报（73个集合成员），该配置被称为"Global Full Ensemble"；其二，在相同缩短的历史时段（1981-2010年）使用针对加州洋流生态系统（California Current Ecosystem, CCE）降尺度的预报（3个集合成员），该配置被称为"Downscaled Ensemble"；其三，在相同缩短的历史时段（1981-2010年）使用全球预报的缩减子集（3个集合成员），该配置被称为"Global Reduced Ensemble"。

所有预报均与从CCE区域再分析资料中提取的SST观测值进行对比。该再分析资料基于区域海洋模式系统（Regional Ocean Modeling System, ROMS），覆盖美国西海岸（30°N~48°N，134°W~115.5°W），水平分辨率为0.1°（约10 km），包含42层地形追随垂直层。

案例研究1：栖息地压缩指数
栖息地压缩指数（Habitat Compression Index, HCI）是一种沿美国西海岸刻画冷水栖息地的区域解析指标；本研究使用的该指标监测加州外海（35°N~40°N）的表层海水状况。HCI用于评估上升流栖息地（以冷水为表征）向海岸挤压的程度，因为富含营养物质的上升流水体可吸引鲸类前来觅食。HCI的计算方式为：海岸线150 km范围内海表温度低于月度阈值的网格单元数量。随后将该数值归一化至150 km海域范围内的总网格单元数，使结果取值范围为0~1。月度海表温度阈值为1981-2010年期间海岸至75 km近海海域的平均月度海表温度。HCI低值代表冷水栖息地被高度挤压或缩减，这正是负责缓解鲸类缠绕风险的资源管理者重点关注的指标。利用HCI的长期平均值可识别高压缩事件（即数值低于平均值的情况）。

案例研究2：TOTAL工具
避免蠵龟兼捕温度观测工具（Temperature Observations to Avoid Loggerheads, TOTAL）通过监测南加州湾（31°N~34°N，120°W~116°W）内异常偏高的海表温度，作为海龟兼捕风险的指示因子，并为实施渔业休渔提供建议。TOTAL的计算方式为南加州湾海域内海表温度异常（Sea Surface Temperature Anomaly, SSTA）的6个月滑动平均值。根据前6个月的SSTA，每年6、7、8月这三个月可能实施空间休渔。若SSTA超过阈值（该阈值为2014年8月、2015年6-8月及2016年6-8月这三次历史休渔期前的最小月度异常值），则建议实施休渔。

预报技巧评估
我们通过三种评估指标对比观测值与预报值，以此评价各管理工具的预报技巧：（1）相关系数：用于定量表征观测值与预报值之间线性关系的强度；（2）预报准确率：代表正确预报的占比；（3）对称极端依赖指数（Symmetric Extremal Dependence Index, SEDI）：该指标具备多项特性，非常适合用于量化极端罕见事件的预报技巧。各指标的详细定义与计算公式详见本文手稿。