Advancing drought impact early warning systems for vegetation: root-zone soil moisture simulations and ensemble forecasts for southeastern Australia

Abstract

Drought ranks among the costliest natural hazards, exerting far-reaching consequences on natural ecosystems and human societies, particularly when severe soil moisture deficits induce vegetation deterioration (VD) stress. Timely and skilful forecasts of vegetation drought impacts should enable more proactive and effective drought preparedness, management and mitigation. Predicting vegetation response to drought stress however remains challenging due to complex interactions between soil moisture deficits and plant health, compounded by varying response times and magnitudes across diverse ecosystems. To address this gap, this thesis advances the quantitative understanding of soil moisture-vegetation relationships across spatial and temporal scales, serving to develop a skilful early warning framework for multi-categorical vegetation deterioration risks. The thesis encompasses three core research objectives: 1. examining the lower tail distribution of soil moisture and vegetation anomalies to identify critical thresholds for multi-level VD risk prediction; 2. establishing and refining innovative prediction frameworks-specifically integrating a copula-based joint probability approach and a 'threshold event-matching' method-to capture both mild and severe VD events; and 3. evaluating the extended skilful lead time and practical early warning horizon of the integrated framework in real-world drought risk management. By focusing on the growing season (May-October) and utilising soil moisture historical simulations and ensemble forecasts from the Australian Water Resources Assessment Landscape (AWRA-L) model, this thesis delivers a systematic approach for anticipating vegetation stress across large spatial domains. Chapter 3 develops and tests a copula-based statistical framework, using simulated root-zone soil moisture anomalies to predict multiple severity levels of vegetation deterioration as observed in satellite-derived vegetation indices. In Chapter 4, an innovative threshold event-matching approach is introduced, using in-situ soil moisture observations to capture rare but high-impact VD events. This method improves predictability, particularly in agricultural regions, by pinpointing optimal soil moisture anomaly thresholds that correspond to substantial vegetation stress and providing lead times of approximately two to six weeks. Building on these insights, Chapter 5 integrates the copula-based model and the threshold 'event-matching' framework. Copula-based modelling tends to underestimate severe VD events, whereas the event-matching framework can overestimate mild-moderate categories; combined, however, they better capture the full spectrum of multi-categorical VD risks. Subsequently, Chapter 6 applies the copula-based joint probability modelling and threshold event-matching approaches as an integrated framework to provide ensemble soil moisture forecasts, quantifying its ability to provide actionable early warning signals at lead times of one to six months. Finally, Chapter 7 consolidates the findings and emphasises the framework's operational relevance for drought-prone agricultural systems. It outlines future research priorities, including enhancing spatial scalability and linking forecasts to economic cost-loss analyses. Collectively, the thesis presents an operational predictive system that advances impact-based drought early warning. By integrating soil moisture anomalies with multi-categorical vegetation deterioration risks, it enables improved seasonal forecasting of drought impacts on vegetation. Through the application of ensemble soil moisture forecasts, the framework supports proactive decision-making and resource allocation for drought preparedness and mitigation. While its immediate application targets southeastern Australian agricultural systems, the methodologies, insights and modelling strategies developed here have broader relevance for other dryland regions worldwide. Show more