Hydrological cycle during droughts: large-scale analyses for process understanding and modelling

Droughts strongly affect the environment and human activities with long-term and far-reaching impacts that will increase in the next decades under global changes. Thus, we need an in-depth understanding of drought processes and their robust modelling to cope with drought risk. For hydrologists, recurring challenges include predicting the impacts of precipitation (P) deficits in the form of soil moisture, streamflow (Q), or groundwater deficits. Water stored in catchments and evapotranspiration (ET) regulate drought evolution, that is the propagation of P deficits through the hydrological cycle and the subsequent recovery. Yet, analyses explicitly considering the joint contribution of storage and ET to drought evolution across different hydroclimatic regimes are rare. Furthermore, many hydrological models poorly simulate Q during droughts, but previous studies have rarely assessed model performances during droughts in multi-variable and spatially-distributed evaluations. This PhD thesis aimed to answer two main research questions: (i) do storage changes and ET affect drought evolution across climates and landscapes?; (ii) does a distributed hydrological model properly represent Q, ET, and storage during droughts? I performed a large-sample data-based analysis of Q, ET, and changes in the subsurface storage (in soil and groundwater) over the period 2010-2019 for 102 Italian catchments to answer the first question. To address the second question, I evaluated Q, ET, and storage simulations from the process-based distributed hydrological model Continuum over the Po river basin (northern Italy) during recent droughts, including the severe 2022 event. From the large-sample data-based analysis, I found that annual subsurface storage changes represented on average 11% of annual P across the study catchments, and mostly buffered Q deficits during drought years and their recovery. ET, instead, both buffered and aggravated Q deficits, and it had a decoupled response to P. These results revealed the prominent role of subsurface storage in driving the evolution of annual droughts. From model evaluation, I showed worse model performances in simulating Q for severe than for moderate droughts (mean KGE across the 38 study sub-catchments = 0.55±0.25 during moderate droughts and 0.18±0.69 in 2022) and I linked them to a degraded simulation of ET, rather than storage, especially in the human-affected croplands (mean r = -0.03 and nRMSE = 1.8 across the croplands in 2022). By calibrating the model during a moderate drought, I showed similar model performances during the severe event (mean KGE = 0.18±0.63), which further point to specific human-water processes during this event. Therefore, I delineated possible ways forward for model improvement during severe droughts, such as an enhanced consideration of human interference, especially in ET. The findings of the thesis provided a consistent picture of the different role ET and storage have in drought evolution and in our modelling capabilities, coherently with recent literature, also on multi-year droughts. Moreover, these results emphasized the need for a holistic approach across the hydrological cycle for process understanding and model evaluation during droughts, with the ultimate goal of improving drought modelling for water resources management, disaster risk reduction, and climate change impact assessments.