Scholarlink

<p>Drought risk emerges from the joint action of hydroclimatic forcing, land-surface memory, and the <i>event-time organisation</i> of dry conditions. Beyond deficit magnitude, practical risk assessment depends on when dry spells initiate, how onsets cluster, and how persistence controls recovery and the likelihood of compound and cascading impacts. These features are expected to intensify in many regions under non-stationary forcing, yet are only partially represented by accumulation-based drought indices and by continuous time-series diagnostics. We develop a stochastic, event-based framework in which <i>real dry-spell onsets</i> (the first dry month following a non-dry month) are treated as a point process evolving in a slowly varying environment. Onset hazard is modelled via a conditional-intensity decomposition that separates an exogenous baseline component—capturing slow modulation by climate drivers and trends—from an endogenous history-dependent component—capturing memory-amplified clustering consistent with persistence pathways such as circulation regime persistence, storm-track organisation, land–atmosphere coupling, and soil-moisture feedbacks. A compact diagnostic emerges naturally from this decomposition: a bounded, dimensionless <i>endogeneity (memory) index</i> quantifying, locally in time, the fraction of onset propensity attributable to event-history amplification rather than baseline forcing. Using monthly CHIRPS precipitation (1981–2024, quasi-global coverage) and a fully reproducible pipeline across contrasting hydroclimates, we show how event-time diagnostics complement standard continuous indicators by (i)&#xa0;isolating baseline-driven versus memory-driven contributions to changing onset hazard, and (ii)&#xa0;providing a physically interpretable pathway to monitor evolving dry-spell initiation risk under non-stationarity. A controlled synthetic experiment demonstrates that classical anomaly-based diagnostics—including rolling variance, lag-1 autocorrelation, and, more broadly, any method that operates on the continuous anomaly trajectory—cannot uniquely determine the temporal organisation of the induced onset process, establishing a precise sense in which the event-time framework is complementary rather than redundant. The framework is designed for risk-oriented use: it is threshold-transparent, extensible to multiple hazard definitions, and compatible with dependence-aware uncertainty quantification.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A stochastic hazard decomposition for dry-spell onset dynamics under non-stationary hydroclimatic forcing

  • 期刊论文

  • 发表日期：

    2026年6月30日

查看全文
  • Mauricio Herrera-Marín

摘要

Drought risk emerges from the joint action of hydroclimatic forcing, land-surface memory, and the event-time organisation of dry conditions. Beyond deficit magnitude, practical risk assessment depends on when dry spells initiate, how onsets cluster, and how persistence controls recovery and the likelihood of compound and cascading impacts. These features are expected to intensify in many regions under non-stationary forcing, yet are only partially represented by accumulation-based drought indices and by continuous time-series diagnostics. We develop a stochastic, event-based framework in which real dry-spell onsets (the first dry month following a non-dry month) are treated as a point process evolving in a slowly varying environment. Onset hazard is modelled via a conditional-intensity decomposition that separates an exogenous baseline component—capturing slow modulation by climate drivers and trends—from an endogenous history-dependent component—capturing memory-amplified clustering consistent with persistence pathways such as circulation regime persistence, storm-track organisation, land–atmosphere coupling, and soil-moisture feedbacks. A compact diagnostic emerges naturally from this decomposition: a bounded, dimensionless endogeneity (memory) index quantifying, locally in time, the fraction of onset propensity attributable to event-history amplification rather than baseline forcing. Using monthly CHIRPS precipitation (1981–2024, quasi-global coverage) and a fully reproducible pipeline across contrasting hydroclimates, we show how event-time diagnostics complement standard continuous indicators by (i) isolating baseline-driven versus memory-driven contributions to changing onset hazard, and (ii) providing a physically interpretable pathway to monitor evolving dry-spell initiation risk under non-stationarity. A controlled synthetic experiment demonstrates that classical anomaly-based diagnostics—including rolling variance, lag-1 autocorrelation, and, more broadly, any method that operates on the continuous anomaly trajectory—cannot uniquely determine the temporal organisation of the induced onset process, establishing a precise sense in which the event-time framework is complementary rather than redundant. The framework is designed for risk-oriented use: it is threshold-transparent, extensible to multiple hazard definitions, and compatible with dependence-aware uncertainty quantification.