Abstract Climate Impact Response Functions (CIRFs) can be useful for exploring potential risks of system failure under climate change. The performance of a water resource system can be synthesized through a CIRF that relates climate conditions to system behavior in terms of a specified threshold of deliveries to demands or environmental flow requirements. However, in highly regulated water resource systems this relationship may be quite complex, depending on storage capacity and system operation. In this paper we define a CIRF for these types of systems through a multivariable logistic regression (LR) model where a binary variable (system response) is explained by two continuous variables or predictors (precipitation and temperature). The approach involves generating multivariate synthetic inflow time series and relating them to specific climate conditions. Next, these inflows are used as inputs in a water management model, and the outcome is coded as a binary variable (failure or its absence) depending on selected vulnerability criteria. To identify the time span before the failure event in which climate variables are relevant, we characterized drought development stages through relative standardized indices. Mean values of precipitation and temperature for the selected time span are computed and used as explanatory variables through a LR model, which is validated using data from several climate models and scenarios. Results show that the predictive capacity of LR models is acceptable, so that they could be used as screening tools to detect challenging climate conditions for the system which would require adaption actions.

中文翻译：

---

为高度管制的水资源系统开发气候影响响应功能

摘要 气候影响响应函数 (CIRF) 可用于探索气候变化下系统故障的潜在风险。水资源系统的性能可以通过 CIRF 进行综合，该 CIRF 将气候条件与系统行为相关联，具体取决于需求或环境流量要求的指定阈值。然而，在高度管制的水资源系统中，这种关系可能非常复杂，具体取决于存储容量和系统运行情况。在本文中，我们通过多变量逻辑回归 (LR) 模型为这些类型的系统定义 CIRF，其中二元变量（系统响应）由两个连续变量或预测变量（降水和温度）解释。该方法涉及生成多变量合成流入时间序列并将它们与特定气候条件相关联。接下来，这些流入量被用作水资源管理模型的输入，结果根据选定的脆弱性标准被编码为二元变量（失败或不存在）。为了确定与气候变量相关的失败事件之前的时间跨度，我们通过相对标准化的指数来描述干旱发展阶段。计算选定时间跨度的降水和温度平均值，并通过 LR 模型将其用作解释变量，该模型使用来自多个气候模型和情景的数据进行验证。结果表明LR模型的预测能力是可以接受的，