Quantifying groundwater recovery after drought: a comparative modelling study in the Emilia-Romagna multi-layered aquifer system

Extreme hydroclimatic events are increasingly challenging groundwater security, especially in intensively exploited regions; yet, the recovery dynamics of aquifer systems after prolonged drought events remain poorly quantified. This study investigates the response and recovery times of the multi-layered aquifer system in Emilia-Romagna region (north-eastern Italy), one of the country’s most populated and productive areas. We analyze how climatic stressors and anthropogenic pressures shape groundwater head decline and post-drought rebound, comparing the results provided by a numerical groundwater flow model implemented in MODFLOW 6, and by a random forest algorithm.

Both models are calibrated over the years 2010-2018 to reproduce the historical evolution of groundwater heads across the regional aquifer system. A scenario analysis is then carried out from 2019 to 2050, imposing a set of drought conditions characterized by reductions in precipitation and varied groundwater abstractions. These scenarios represent short- and long- duration low-recharge periods with different levels of stress intensity, enabling a systematic exploration of aquifer system’s behaviour under combined climatic and anthropogenic forcing.

Groundwater recovery is assessed through the analysis of groundwater heads simulated by both modeling approaches. This study provides quantitative insights into the resilience of the regional multi-layered aquifer system to extreme hydroclimatic events and aims at clarifying the respective roles of climate variability and groundwater exploitation in shaping future groundwater security. In particular, the goals are to quantify (i) the mean recovery time following each drought scenario as a function of its duration and intensity, (ii) the relative contribution of abstraction changes to driving groundwater decline and delaying recovery, and (iii) the sensitivity of recovery times to different input variables. Finally, we aim to assess the extent to which the random forest algorithm can replicate the physics-based model under unseen future scenarios, identifying conditions in which data-driven approaches may complement or, in specific context, substitute numerical groundwater models.

Results show that recovery times are strongly dependent on the imposed precipitation reduction and are often markedly influenced by pumping regimes, which can exert a dominant control on the system. The random forest model accurately reproduces system dynamics under conditions similar to the calibration period but shows reduced reliability under extreme scenarios. Overall, the results highlight the need to carefully account for both climatic variability and human-driven pressures when evaluating future groundwater resilience, and underscore the value of integrating complementary modeling approaches to improve groundwater management strategies under increasing hydroclimatic uncertainty.