Application of the modflow groundwater numerical model to hydrogeological volcanic units

Abstract

The work of study presented can be retraced in the following main phases: Comprehensive work of examination and interpretation of several field campaign data and literature data: well log stratigraphies, geological maps. From a detailed starting approach, in which the comprehensive volcanic sequence has been analyzed and each deposit’s hydraulic behaviour has been considered, to a simplified re-organization with the definition of main hydrogeological complexes. Conceptual model building: geological and hydrogeological framework has been investigated, a volumetric reconstruction of the aquifer has been made, were areas with different hydraulic behaviour were identified and media type defined (porous media and fractured). Then the identification of the sources and sinks: summarised in the elaborated water budget. Some issues linked to the conceptual model have to be highlighted: - The evaluation of the geologic heterogeneity effect on the hydraulic behaviour at the model scale. Volcanic deposits are characterized by a high heterogeneity which conditions groundwater flow patterns. So, which value could have an equivalent (effective) hydraulic conductivity, which can represent the integrated effect of a heterogeneous media? A more detailed analysis of pumping test; including data coming from other zones of the latium volcanic complex should be useful. Heterogeneity plays a key part in the uncertainty estimation (Section 4.7.1); it has to be improved for the proper evaluation of the two key parameters, correlation lenght λ and the σ (standard deviation of lnK). - There was applied a methodology, suggested by Sonnenborg & Henriksen (2005), to determine the uncertainty of observational head data in relation to the model. The purpose of quantifying the uncertainty of observational head data (σ obs) is to achieve a measure of how accurately the model can reproduce the data. This should in principle be an objective criterion for how data weighted approximation and more importantly, how data of different types must be relatively weighted (important when using objective functions). - It was revealed that the component due to geologic heterogeneity on the whole observational head uncertainty value, played a major role for such performance assessments and that the lack of experimental data suggested that this aspect should be further analysed in more depth when doing future model analysis of volcanic aquifers. From the conceptual model a numerical groundwater flow model (mathematical model) was set up using a finite-difference code MODFLOW2000 and in a first step the graphical interface Visual Modflow 2009-1®. In this first two dimensional, steady state numerical groundwater model, the aquifer system was represented by a one layer model describing the groundwater flow systems in 2 dimensions. Bracciano Lake area together with the underlying aquifer was represented as a simplified single aquifer system, with a higher hydraulic conductivity for the lake area compared to the surrounding areas. Results from calibration and tentative validation of this simplified model did not satisfy the initial performance criteria and objectives identified, e.g. it was not possible to evaluate lake-aquifer interactions (see chapter 1) in the required details with this first model. Next step consisted in setting up a new mathematical model based on a numerical integrated groundwater flow and lake model by use of Groundwater Vistas®5.41 (ESI), which allowed incorporation of the implemented lake package in this software, thus it was possible explicitly to include Lake Bracciano and its interaction with the underlying volcanic aquifer within this package. In order to keep the model as simple as possible, the whole aquifer was divided in two layers having the same thickness, except in Lake Bracciano area. In the lake area the upper layer was defined by the lake depth, whereas the lower layer was comprised by the aquifer. Calibration and tentative validation tests were carried out on the VMF model. PEST was used for the inverse calibration, with head targets. The methodology allowed for a comparison of the residual statistics (difference between simulated and observed head). Hereby, head observational uncertainty when referring to model scale and other target values could be evaluated for three adopted quantitative accuracy Criteria (Refsgaard et al., 2010). Criteria value ranges were related to the established model detail that is a function of the objectives and the ambition level, as reported in Refsgaard et al., 2010. Ambition level can be considered as: 1) screening/basic (rough calculations), 2) intermediate and 3) high (aquifer simulation/detailed modelling). In this work of study it was decided, based on the available knowledge and data, to go for a screening model (as a first step): ”A screening model describes the regional characteristics of the ground-water-flow system without including the hydrogeologic detail or data density that would be necessary for answering site-specific questions. A calibrated screening model can be used with confidence to simulate a regional ground-water-flow system, but with less confidence to simulate local-scale flow. A screening model is a tool that can be used to improve the overall understanding of the hydrology of a basin by testing alternative conceptual models of the ground-water-flow system. Additionally, a screening model can be used to highlight areas where more hydrogeologic or water-quality data are needed”. (Feinstain et al., 2005) Calibration carried out using the Groundwater Vistas 5.41® MODFLOW interface was initially focused on a sensitivity analysis to obtain an acceptable value of the lakebed hydraulic conductivity. A second step was the “autosensitivity” analysis (Groundwater Vistas tool) that permits to determine in a detailed analysis which parameters are sensitive to the available targets and which parameters are closely linked. Then PEST was used to calibrate the model using automatic calibration (inverse routines). At the beginning the head dataset and four flux values were applied. Subsequently eight flux values were used in combination with the head data. Head targets were divided in two groups: one representing water table level nearby the lake and another representing targets all over the study area. Flux (flow in drainage systems) was then the third target group. When using inverse modelling (PEST) it is important to assign proper weights to the different targets in order to determine their relative importance in the parameter estimation process and understand by which weights the model is optimized with respect also to the objectives. Groundwater model results improved after the calibration process. One of the objectives of the Bracciano groundwater model was that it should be a useful tool for testing the current understanding and assumptions established with the conceptual model. The use of inverse modelling (PEST) here allowed for understanding about whether the constructed model for the aquifer system and the lake interaction turned out as a credible and reliable model. If that is the case, then the model can be used in future management of the water resources for the volcanic basin. Thus two tests are important for evaluating the credibility and reliability: 1) Quantitative performance e.g. the simulated and observed water balance for aquifer and lake and results of validation tests (the 3 Criteria), when running the model with data not used for calibration and 2) Qualitative performance e.g. whether the calibrated parameters are within realistic ranges, whether the simulated head distribution is credible compared to manual mapped distributions etc. During calibration process, the water budget values are constantly compared with the volumes estimation coming out from simulations to evaluate the results.The hydrogeologic system studied is characterized by the presence of four volcanic calderas, two of these still occupied by lakes (Lake Bracciano and Lake Martignano), there are two big water abstraction sites: one from lake (Paolo aqueduct) and the other fed by drains on the northwest side of the Lake Bracciano (Traiano aqueduct). At the same time the area is exposed to a continuous exploitation and dewatering from effect of the several public and private pumping wells from the groundwater aquifer put in action in the last twenty years. The management of this complex system is challenging, and could surely benefit of the predictive capacities that a calibrated and validated model would enable and provide. It should be an interest and preference of the public administration to implement a stronger tool as a groundwater model, since a quantitative understanding of the whole system and its interactions are needed in order to exploit the groundwater aquifer in a sustainable way. Also for the different water users it is important to manage their own wells based on a better understanding of the whole system, and how their groundwater abstraction and land use affects the recharge, water balance of the aquifer, interaction with lake, wetlands near the lake and runoff from the area. Lections learned with this work of study are: - The use of the “lake package” to study aquifer-lake interactions appears to be promising and needed in order to properly address the conceptual model and the objectives defined. - The MODFLOW model (final Groundwater Vistas model) could be a “basic” model that could be further improved if running a new model cycle in order to arrive to a starting model useful for the water resources administrator (with a special focus to Lake Bracciano, which constitute a site of environmental and economical interest); to reach an intermediate level useful for credible and reliable predictive simulations of the regional Bracciano aquifer-lake system for practical water management purposes - The actual model should be improved by introducing a “zonation” of values of the lakebed hydraulic conductivity, considering influencing factors as the wave energy, the slope of the bottom surface, etc. This would probably require additional field investigations of the groundwater – lake interactions - Simulation in transient conditions should be considered. This would enable more possibilities for validating the model, because time series of head and flow including the storage terms in groundwater and lake could be explicitly included. Lake budget data should be inserted at a time scale relevant (e.g. monthly): so rain, evaporation, runoff and lake elevation could be elaborated with this transient model approach. It could be another way to improve Bracciano groundwater model and to test also the input data (the calculation of monthly net precipitation and groundwater recharge to the model). A three or more layer model could be considered, with the aim of representing the presence of the aquitards, perched water tables etc. There are several challenges, and feasibility of different model codes with respect to describing ‘perched water tables’, unsaturated zone, overland flow, river and drainage flow etc. needs to be further evaluated before an high ambition model level could be met.