Hawai'i Institute of Marine Biology, Kaneohe, Hawaii, USA.
High Meadows Environmental Institute, Princeton University, Princeton, New Jersey, USA.
Ecol Appl. 2024 Jun;34(4):e2961. doi: 10.1002/eap.2961. Epub 2024 Mar 24.
Ecological forecasts are becoming increasingly valuable tools for conservation and management. However, there are few examples of near-real-time forecasting systems that account for the wide range of ecological complexities. We developed a new coral disease ecological forecasting system that explores a suite of ecological relationships and their uncertainty and investigates how forecast skill changes with shorter lead times. The Multi-Factor Coral Disease Risk product introduced here uses a combination of ecological and marine environmental conditions to predict the risk of white syndromes and growth anomalies across reefs in the central and western Pacific and along the east coast of Australia and is available through the US National Oceanic and Atmospheric Administration Coral Reef Watch program. This product produces weekly forecasts for a moving window of 6 months at a resolution of ~5 km based on quantile regression forests. The forecasts show superior skill at predicting disease risk on withheld survey data from 2012 to 2020 compared with predecessor forecast systems, with the biggest improvements shown for predicting disease risk at mid- to high-disease levels. Most of the prediction uncertainty arises from model uncertainty, so prediction accuracy and precision do not improve substantially with shorter lead times. This result arises because many predictor variables cannot be accurately forecasted, which is a common challenge across ecosystems. Weekly forecasts and scenarios can be explored through an online decision support tool and data explorer, co-developed with end-user groups to improve use and understanding of ecological forecasts. The models provide near-real-time disease risk assessments and allow users to refine predictions and assess intervention scenarios. This work advances the field of ecological forecasting with real-world complexities and, in doing so, better supports near-term decision making for coral reef ecosystem managers and stakeholders. Secondarily, we identify clear needs and provide recommendations to further enhance our ability to forecast coral disease risk.
生态预测正成为保护和管理的越来越有价值的工具。然而，几乎没有考虑到广泛的生态复杂性的近实时预测系统的例子。我们开发了一种新的珊瑚疾病生态预测系统，该系统探索了一系列生态关系及其不确定性，并研究了预测技巧如何随着较短的前置时间而变化。这里介绍的多因素珊瑚疾病风险产品利用生态和海洋环境条件的组合来预测中太平洋和西太平洋以及澳大利亚东海岸的珊瑚礁上出现白综合征和生长异常的风险，并且可以通过美国国家海洋和大气管理局珊瑚礁观察计划获得。该产品每周根据分位数回归森林对 6 个月的移动窗口进行预测，分辨率约为 5 公里。与前一代预测系统相比，该预测在预测 2012 年至 2020 年保留的调查数据中的疾病风险方面表现出更高的技能，对于预测中等到高疾病水平的疾病风险，改进最大。大部分预测不确定性来自于模型不确定性，因此预测精度和准确性不会随着前置时间的缩短而显著提高。这一结果是因为许多预测变量无法准确预测，这是整个生态系统中常见的挑战。每周的预测和情景可以通过与最终用户群体共同开发的在线决策支持工具和数据资源管理器进行探索，以提高对生态预测的使用和理解。该模型提供了近乎实时的疾病风险评估，并允许用户改进预测并评估干预情景。这项工作通过引入现实世界的复杂性来推进生态预测领域的发展，并在此过程中更好地支持珊瑚礁生态系统管理者和利益相关者的短期决策。其次，我们确定了明确的需求并提出了建议，以进一步提高我们预测珊瑚疾病风险的能力。
