Managing fragile island freshwater resources requires identifying pumping strategies that trade off the financial cost of groundwater supply against controlling the seawater intrusion (SWI) associated with aquifer pumping. In this work, these tradeoffs are investigated through a sensitivity analysis conducted in the context of an optimization formulation of the groundwater management problem, which aims at minimizing the groundwater supply operation cost associated with groundwater pumping and desalination treatment, subject to constraints on SWI control, as quantified by the water table drawdown over the well (∆s), the reduction in freshwater volume (∆FV) in the aquifer, or the salt mass increase (∆SM) in the aquifer. This study focuses on a simplified two‐dimensional model of the San Salvador Island aquifer (Bahamas). Pumping strategies are characterized by the distance of the pumping system from the shoreline (WL), the abstraction screen depth (D) and overall pumping rate (Q), constituting the decision variables of the optimization problem. We investigate the impacts of pumping strategies on the operation cost, ∆s, ∆FV and ∆SM. Findings indicate increasing D or decreasing WL reduces ∆s, ∆FV and ∆SM, thus preserving the aquifer hydrogeologic stability, but also leads to extracting saltier groundwater, thus increasing the water treatment requirements, which have a strong impact on the overall groundwater supply cost. From a financial perspective, groundwater abstraction near the island center and at shallow depths seems the most convenient strategy. However, the analysis of the optimization constraints reveals that strategies where the pumping system approaches the island center tend to cause more severe SWI, highlighting the need to trade off groundwater supply cost against SWI control.

In groundwater pumping optimization (GPO), offline-trained data-driven surrogates can be used to replace numerical-intensive simulators in order to save computing time. The traditional offline training approach involves building surrogates prior to optimization, fitting training datasets that cover the input space uniformly or randomly, which can prove inefficient due to the potential oversampling of low-gradient areas and under-sampling of high-gradient areas. This study proposes an offline machine-learning (ML) algorithm that ranks candidate training points by scoring them based on their distance to the closest training point and on the local gradient of the surrogate estimate and then choosing the highest-rank point. This method is applied to develop surrogates for solving a two-objective GPO problem formulated on a three-dimensional (3D) island aquifer, using hydrogeological conditions representative of San Salvador Island, Bahamas. The objectives are to minimise the supply cost (f_OC) resulting from groundwater pumping and desalination and maximise fresh groundwater supply (Q_p), subject to constraints on seawater intrusion (SWI) control expressed in terms of aquifer drawdown Δs at pumping locations and aquifer salt mass increase ΔSM. Gaussian Process (GP) is the technique applied to construct surrogates of objectives and constraints, alongside the estimation of uncertainties. Using GP models, it is possible to estimate the probability of “Pareto optimality” for each pumping scheme by Monte Carlo simulation. Pareto optimal pumping schemes (POPS) are then characterized by a probability of occurrence, which can be verified by numerical simulation. The GP training strategy's effectiveness in generating POPS is compared to traditional training approaches, showing that such a strategy can efficiently identify reliable POPS.

Ensuring a steady supply of drinking water is crucial for communities, but predicting how much water will be needed is challenging because of uncertainties. As a part of Battle of Water Demand Forecasting (BWDF), this study delves into the application of Long Short-Term Memory (LSTM) networks for water demand forecasting in a city situated in the northeast of Italy. The focus is on forecasting the demand across ten distinct District Metering Areas (DMAs) over four distinct stages. To enhance the performance of the LSTM model, an evolutionary optimization algorithm is integrated, aiming to fine-tune the model’s hyper-parameters effectively. Results indicate the promising potential of this approach for short-term demand forecasting.

Seawater intrusion (SWI) in coastal aquifers is influenced by both spatial distributions of hydraulic properties and pumping rates. The heterogeneity of parameters such as hydraulic conductivity (K) may have a significant impact on the vulnerability of these systems to SWI and the sustainability of groundwater supply.In this work vulnerability is assessed in relation to indicators such as: (a) the salt concentration of extracted groundwater (C); and (b) the total dissolved mass (TDM) of salt in the aquifer. This study delves into a modified version of the classical two-dimensional Henry's problem, featuring a single pumping well situated in proximity of the coastline. The well pumping rates are expressed as fractions of the total flow rate entering the aquifer from the inland boundary. The fraction range from a minimum of zero (no pumping) to a maximum of five.To investigate the impact of aquifer heterogeneity a series of stochastic simulations are conducted using the popular variable density flow model SEAWAT. Heterogeneity in the K spatial distribution is modelled as a geostatistical log-normal process characterized by an exponential covariance function, with variance values ranging from 0.25 to 1.0, horizontal correlation scales from 10 m up to 2000 m, and a single vertical correlation scale of 10 m. This approach allows for investigating the effects of the K heterogeneity, the layering features of the aquifer system, and the intensity of groundwater pumping on C and TDM.A detailed examination of model results reveals several interesting outcomes. Most notably, the TDM indicator is rather sensitive to the K heterogeneity, whereas the C indicator is much less affected by it. In terms of TDM, selected cases are singled out and discussed. In general, lower K values exhibit minimal SWI even with large pumpage, whereas high K values lead to pronounced SWI even at low pumping. Increased variance and horizontal correlation correspond to varied SWI behaviour. The TDM spread remains generally low due to relatively low variance. With an increase in horizontal correlation, the spread becomes more pronounced for low pumping rates and less pronounced for large pumping rates.