Multivariate statistical methods – principal component analysis (PCA) and hierarchical cluster analysis (HCA) – are applied to identify geochemically distinct groundwater groups in the territory of Latvia. The main processes observed to be responsible for groundwater chemical composition are carbonate and gypsum dissolution, fresh and saltwater mixing and ion exchange. On the basis of major ion concentrations, eight clusters (C1–C8) are identified. C6 is interpreted as recharge water not in equilibrium with most sediment forming minerals. Water table aquifers affected by diffuse agricultural influences are found in C3. Groundwater in C4 reflects brine or seawater admixture and gypsum dissolution in C5. C7 and C2 belong to typical bicarbonate groundwater resulting from calcite and dolomite weathering. Extremely low Cl⁻ and SO₄²⁻ are observed in C8 and described as pre-industrial groundwater or a solely carbonate weathering result. Finally, C1 seems to be a poorly defined subgroup resulting from mixing between other groups. This research demonstrates the validity of applying multivariate statistical methods (PCA and HCA) on major ion chemistry to distribute characteristic trace elements in each cluster even when incomplete records of trace elements are present.

Estimated natural background levels (NBLs) are needed to assess groundwater chemical status according to the EU Groundwater Directive. They are commonly derived for different substances by applying statistical methodologies. Due to the complexity of the sea water intrusion process, some of those methods do not always provide appropriate assessment of chloride NBLs. This paper analyzes the applicability of different NBL estimation methods in five EU coastal aquifers with significant differences in available datasets and hydrogeological settings. A sensitivity analysis of results to different constraints was performed to remove samples with anthropogenic impacts. A novel statistical approach combining different methods to identify the range of chloride NBLs is proposed. In all pilots the estimated NBLs were below 85 mg/L and fitted well with previous studies and expert judgment, except Campina del Faro aquifer (the maximum being 167.5 mg/L). Although this approach is more time consuming, it provides a more robust solution.

Groundwater pollution by agrochemicals such as nitrogen fertilizers can cause complex biogeochemical transformations to take place in groundwater-dependent ecosystems. To explore the interaction between nitrogen load and groundwater-dependent, spring-fed ecosystems, a study was conducted in Latvia in an area of suspected high nitrate (NO₃⁻) vulnerability due to its geological settings. A map of NO₃⁻vulnerability along the margins of the carbonate aquifer in Latvia is presented. The map is based on a conceptual model that was developed during an extensive case study involving hydrological, hydrochemical, and habitat investigation of springs discharging from a karst aquifer and spring-fed ecosystems. Areas that should be prime targets for restricting fertilizer application are highlighted on the map. Although the case study revealed increased nitrogen pollution (up to 51 mg L⁻¹, standard deviation of 9 mg L⁻¹, in the springs discharging from the karst aquifer), no clear evidence of adverse effects due to NO₃⁻pollution on the groundwater-dependent ecosystems using biotic indicators was found, highlighting the resilience of spring-fed ecosystems against high nitrogen inputs. In the case study, downstream groundwater-dependent ecosystems retained 70% of the reactive nitrogen during the vegetation season, but only a small proportion during the cold season. Thus, NO₃⁻pollution can be partly mitigated by restoring wetlands along valley slopes where natural groundwater discharge takes place. The conceptual model developed for groundwater NO₃⁻vulnerability is applicable to other areas in the Baltic region and other places with similar climatic and geological conditions.

This paper presents the results of the 2022 Groundwater Time Series Modelling Challenge, where 15 teams from different institutes applied various data-driven models to simulate hydraulic-head time series at four monitoring wells. Three of the wells were located in Europe and one was located in the USA in different hydrogeological settings in temperate, continental, or subarctic climates. Participants were provided with approximately 15 years of measured heads at (almost) regular time intervals and daily measurements of weather data starting some 10 years prior to the first head measurements and extending around 5 years after the last head measurement. The participants were asked to simulate the measured heads (the calibration period), to provide a prediction for around 5 years after the last measurement (the validation period for which weather data were provided but not head measurements), and to include an uncertainty estimate. Three different groups of models were identified among the submissions: lumped-parameter models (three teams), machine learning models (four teams), and deep learning models (eight teams). Lumped-parameter models apply relatively simple response functions with few parameters, while the artificial intelligence models used models of varying complexity, generally with more parameters and more input, including input engineered from the provided data (e.g. multi-day averages). The models were evaluated on their performance in simulating the heads in the calibration period and in predicting the heads in the validation period. Different metrics were used to assess performance, including metrics for average relative fit, average absolute fit, fit of extreme (high or low) heads, and the coverage of the uncertainty interval. For all wells, reasonable performance was obtained by at least one team from each of the three groups. However, the performance was not consistent across submissions within each group, which implies that the application of each method to individual sites requires significant effort and experience. In particular, estimates of the uncertainty interval varied widely between teams, although some teams submitted confidence intervals rather than prediction intervals. There was not one team, let alone one method, that performed best for all wells and all performance metrics. Four of the main takeaways from the model comparison are as follows: (1) lumped-parameter models generally performed as well as artificial intelligence models, which means they capture the fundamental behaviour of the system with only a few parameters. (2) Artificial intelligence models were able to simulate extremes beyond the observed conditions, which is contrary to some persistent beliefs about these methods. (3) No overfitting was observed in any of the models, including in the models with many parameters, as performance in the validation period was generally only a bit lower than in the calibration period, which is evidence of appropriate application of the different models. (4) The presented simulations are the combined results of the applied method and the choices made by the modeller(s), which was especially visible in the performance range of the deep learning methods; underperformance does not necessarily reflect deficiencies of any of the models. In conclusion, the challenge was a successful initiative to compare different models and learn from each other. Future challenges are needed to investigate, for example, the performance of models in more variable climatic settings to simulate head series with significant gaps or to estimate the effect of drought periods.