Enter Zoom Meeting

HS3.8

EDI
Rapid, reproducible, and robust hydrosystem modeling for decision support: worked examples and open-source software tools

To provide support for resource management decision making, computational modeling workflows in hydrosystem simulation need to be efficient, reproducible, and robust with regard to the risk of unwanted outcomes. Unfortunately, each of these three attributes is difficult to achieve in practice; aspirations to simultaneously achieve all of them are truly lofty. Too often, modeling analyses are inefficient, the workflow is largely opaque and unknown, and the important simulated outcomes lack the context of uncertainty and/or risk.

This session calls for submissions that demonstrate rapid, reproducible and/or robust modeling through worked examples and software tools (a preference for open source). The worked examples should demonstrate how the researcher aspired to be rapid, reproducible, and robust; we are interested in the process and approach as much as the results. We aim to stimulate discussion based on lessons learned and results presented, for other researchers and practitioners to build on. We particularly welcome descriptions of trials and tribulations: What was difficult? What didn’t work? How were these issues overcome?

Software tools may include:
• techniques to automate modeling workflow elements or increase efficiency, reproducibility, robustness of decision-support modeling elements.
• frameworks to build models from original data in flexible ways that may enable hypothesis testing in the form of changing discretization, process representation, and other modeling decisions.
• multi-model frameworks such as Bayesian-model selection/combination, as well as frameworks to accommodate model structural error.
• Methods for uncertainty analysis, data assimilation, and management optimization under uncertainty in the decision-support context.
• machine-learning approaches for decision support analyses.

Convener: Anneli GuthkeECSECS | Co-conveners: Jeremy White, Michael Fienen, Catherine Moore
Presentations
| Fri, 27 May, 13:20–16:40 (CEST)
 
Room 2.15

Fri, 27 May, 13:20–14:50

Chairpersons: Anneli Guthke, Dirk Eilander

13:20–13:25
Session introduction

13:25–13:35
|
EGU22-10181
|
solicited
|
Virtual presentation
Jacob Knight

A scripted development and deployment approach was used for developing the next-generation groundwater flow and land-surface subsidence model of the region surrounding Houston, Texas, USA.  The area has historically experienced substantial land subsidence resulting from groundwater use. Python scripts leveraging the FloPy and PyEMU packages were written to build and run the MODFLOW 6 model, perform very-high-dimensional parameter estimation and uncertainty analysis using PEST++, and process results. Automating these processes allowed for fast and repeated iterations through all or part of the modeling workflow for purposes including: troubleshooting input errors, testing hypotheses about the hydrologic system characteristics, evaluating the influences of structural model assumptions, and experimenting with different and increasingly complex formulations of the prior parameter distribution and likelihood functions in Bayesian sense. Automated generation and storage of processed output allowed easy comparison between iterations of the modeling workflow, and Git version control software provided a self-documented model repository with full-featured “undo” for returning to previous states of the workflow and investigating outcomes. The modeling team convened regularly (monthly to twice-weekly) to review results of the latest iteration and decide the next course of action. Model performance was improved steadily and incrementally by focusing on one new feature or problem per workflow iteration until modeling goals were met. This workflow style fostered a sense of predictability and confidence in the project outcome, a welcome departure from the “typical” numerical modeling process of panic and despair.

How to cite: Knight, J.: A worked example of an iterative, scripted approach to stochastic model development and deployment in a highly contentious decision-support setting., EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-10181, https://doi.org/10.5194/egusphere-egu22-10181, 2022.

13:35–13:45
Q&A solicited talk

13:45–13:50
|
EGU22-2978
|
ECS
|
On-site presentation
Ludovic Schorpp et al.

When modelling groundwater systems in Quaternary formation, one of the first steps is to construct a geological and petrophysical model. This  is often  repetitive because it relies on multiple manual steps which include geophysical interpretation, construction of a structural model, identification of geostatistical model parameters, facies and property simulations. Those steps are often carried out in different softwares, which makes the automation intractable or very difficult. A non automated approach requires a lot of time and is critical to update the model for integrating new available data or when some geological interpretations are modified, and to conduct a cross-validation procedure to assess the overall quality of the models. Moreover, it renders the quantification of the joint structural and parametric uncertainty tedious. 

 

To address these issues, we propose a new approach and a Python module to automatically generate realistics geological and parameter models. One of its main features is that the modelling operates in a hierarchical manner. The input data consists of a set of borehole data, surface geology, and a stratigraphic pile. The stratigraphic pile describes formally and in a compact manner how the model should be constructed. It contains the list of the different stratigraphic units and their order in the pile, their conformability (eroded or onlap), the surface interpolation method (e.g. kriging, SGS, MPS, etc.) or also the filling method (e.g. MPS, SIS, etc.). Then, the procedure is automatic. In a first step the stratigraphic unit boundaries are simulated. Secondly, they are filled with lithologies and finally the petrophysical property models are simulated inside the lithologies. All these steps are straightforward and automated once the stratigraphical pile and its related parameters have been defined. Hence, this approach is extremely flexible. It is illustrated using data from an alpine quaternary aquifer in the Upper Aare plain (south-east of Bern, Switzerland).

How to cite: Schorpp, L., Straubhaar, J., and Renard, P.: ArchPy, automated hierarchical modelling of Quaternary aquifers: an example from the Upper Aare Valley, Switzerland, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-2978, https://doi.org/10.5194/egusphere-egu22-2978, 2022.

13:50–13:55
|
EGU22-3529
|
Virtual presentation
Stephan Thober et al.

Operational hydrological modelling and forecasts are based on complex simulation workflows that include, a.o. input data acquisition, pre-processing, hydrologic simulations, post-processing, publication and dissemination of the results. Stakeholders expect regular updates of the information at specified times and in high quality. Therefore, it must be ensured that in the event of an interruption in the workflow, the error can be quickly identified and rectified. Simultaneously, practitioners have high expectations of the model results, that should profit from continuous development of the hydrologic model and other components.

The open-source mesoscale Hydrologic Model mHM (mhm-ufz.org) is a spatially distributed hydrologic model that conceptualizes dominant hydrological processes on the land surface. The unique feature of mHM is the Multiscale Parameter Regionalization (MPR) [1] that relates geophysical properties of the land (e.g., soil and land cover properties) to model parameters via transfer functions at a high spatial resolution (typically less than 250 m cell size). Subsequently, model parameters are aggregated to the spatial resolution at which the model runs are conducted (over 1 km). MPR allows seamless model application at different spatial resolutions and model parameters to be transferred in space [2]. mHM has been applied at different scales ranging from catchments to continents ([3], [4], [5]). mHM is written in Fortran programming language and is available under the GNU Lesser General Public License v3.

mHM has been in continuous development for more than a decade now. In the past year, the following technical and methodological features have been added to the model:

  • Installation via conda: mHM installation can be cumbersome because a Fortran compiler and netCDF4 library is required. We have now created a conda package (ananconda.org) for mHM that allows installing release versions of mHM.
  • Reading of hourly meteorological input files: Traditionally, mHM was designed to read daily meteorological files. However, its internal time step is hourly. As higher resolved observational datasets become available, mHM can now read hourly data. This feature is critical for flood forecasting.

Recently, mHM was applied globally at 0.1 deg grid resolution within the EU Copernicus-funded ULYSSES project. It took 36 hours to simulate 1.4 million  grid cells for 30 years of daily values at 18 compute cores (using OpenMP parallelization). Although the run time provides an acceptable CO2 footprint of  the simulations, it was challenging to organize a 51 member global hydrological forecast ensemble of six terrestrial environmental variables (Q, ET, SM, SWE, PET, GWR). We used the ecFlow workflow manager (https://confluence.ecmwf.int/display/ECFLOW) to submit the simulations to an HPC cluster. ecFlow allows to monitor the status of jobs and build complex workflows that include various tasks. Using a workflow manager like ecFlow allows creating reproducible simulation results more easily. We developed a general-purpose python package (ecPy) to interact with ecFlow functionalities for a wide range of software applications. We will present these new features and design of ecPy in this presentation.

References:

[1] https://doi.org/10.1029/2008WR007327

[2] https://doi.org/10.5194/gmd-2021-103

[3] https://doi.org/10.1002/wrcr.20431

[4] https://doi.org/10.1175/bams-d-17-0274.1

[5] https://doi.org/10.1061/(asce)he.1943-5584.0002097

How to cite: Thober, S., Samaniego, L., Müller, S., Shrestha, P., Kelbling, M., Rakovec, O., Boeing, F., Marx, A., Kumar, R., and Attinger, S.: Hyperresolution Global Operational Hydrological Modelling and Forecasting: enhancing reproducability, skill and workflows setup , EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-3529, https://doi.org/10.5194/egusphere-egu22-3529, 2022.

13:55–14:00
|
EGU22-10074
|
ECS
|
Virtual presentation
Katherine Markovich et al.

Model structural error, arising from the inevitable simplification and abstraction modelers must make of complex real-world system and processes, has been shown to produce biased predictions as a direct result of parameter compensation during data assimilation. The specter of structural error especially plagues groundwater decision support modeling, since the inverse methods underpinning calibration and uncertainty analysis favor a computationally efficient and numerically stable model, or in other words, a simpler model. This work explores sequential data assimilation (DA) as a potential coping mechanism for the structural error encountered by simpler models. Unlike traditional batch methods, sequential methods assimilate data in discrete time intervals or ‘cycles’, and simultaneously estimate model state along with model parameters in order to advance the model forward in time. We hypothesize that the estimable model states in sequential DA afford more flexible and appropriate receptacles for the noise introduced into observations from model structural error. Using a paired complex-simple model approach, we empirically evaluate the predictive outcomes of batch and sequential DA in two model error scenarios: first where error arises from coarser resolution in the simple model, and second where error arises from both coarser resolution and fixed pumping rates in the simple model. Overall, we find that both formulations perform well in both history matching and forecasting when employing a high-dimensional parameterization stance, that is, treating all properties and stresses as uncertain and adjustable during the inversion process. When uncertain parameters are removed from the inversion process, however, the data assimilation process is degraded in different ways for batch and sequential formulations. These results have implications for groundwater decision support modeling as they underscore the pitfalls of fixing parameters a priori, such as with pumping, and present a proof of concept for using adjustable model states to cope with model error in decision support modeling contexts.

How to cite: Markovich, K., White, J., and Knowling, M.: Rapid, reproducible, and wrong? Exploring sequential data assimilation as a coping mechanism for model structural error in groundwater decision support modeling, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-10074, https://doi.org/10.5194/egusphere-egu22-10074, 2022.

14:00–14:05
|
EGU22-4496
|
On-site presentation
Hélène Boisgontier et al.

Setting up spatially-distributed geoscientific models typically requires many (manual) steps to process input data and might therefore be time consuming and hard to reproduce. Furthermore, it can be hard to improve models based on new or updated (large) datasets, such as (global) digital elevation models and land use maps, potentially slowing down the uptake of such datasets for geoscientific modelling.

HydroMT (Hydro Model Tools; https://deltares.github.io/hydromt/latest/) is an open-source Python package that aims to facilitate the process of building models and analyzing model results based on the state-of-the-art scientific python ecosystem, including xarray, geopandas, rioxarray, pyflwdir, numpy, scipy and dask. The package provides a common interface to data and models as well as workflows to transform data to models and analyze model results based on (hydrological) GIS and statistical methods. The common data interface is implemented through a data catalog, which is setup with a simple text yaml file, and supports many different (GIS) data formats and some simple pre-processing steps such as unit conversion. The common model interface is implemented per model software package and provides a standardized representation of the model configuration, maps, geometries, forcing, states and results. The user can describe a full model setup including its forcing in a single ini text file based on a sequence of workflows, making the process reproducible, fast and modular. Besides the Python interface, HydroMT has a command line interface (CLI) to build, update or analyze models. 

The package has been designed with an iterative, data-centered modelling process in mind. First-order models can be setup for any location in the world by leveraging open global datasets. These models can later be improved by updating the input datasets with detailed local datasets. This iterative process enables the user to quickly get an initial model and analyze its result to then make informed decisions about the most relevant model improvements and/or required data collection and to kick-start discussions with stakeholders. Furthermore, model parameter maps or forcing data can easily be modified for model sensitivity analysis or model calibration to support robust modelling practices.

Currently, HydroMT has been implemented for several models through a plugin infrastructure. Supported models include the distributed rainfall-runoff model wflow, the sediment model wflow_sediment, the hydrodynamic flood model SFINCS, the water quality models D-Water Quality and D-Emissions and the flood impact model Delft-FIAT.

In this contribution we will present different modelling applications, including a loosely coupled flood risk model chain, with a focus on how HydroMT was used to build and analyze these models.

How to cite: Boisgontier, H., Hegnauer, M., and Eilander, D.: HydroMT: A Python package to build and analyze hydro models like a data wizard, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-4496, https://doi.org/10.5194/egusphere-egu22-4496, 2022.

14:05–14:10
|
EGU22-10686
Michael Fienen et al.

Environmental water management often benefits from a risk-based approach where information on the area of interest is characterized, assembled, and incorporated into a decision model considering uncertainty. This includes prior information from literature, field measurements, professional interpretation, and data assimilation resulting in a decision tool with a posterior uncertainty assessment accounting for prior understanding and what is learned through model development and data assimilation. Model construction and data assimilation are time consuming and prone to errors, which motivates a repeatable workflow where revisions resulting from new interpretations or discovery of errors can be addressed and the analyses repeated efficiently and rigorously. In this work, motivated by the real-world application of delineating risk-based (probabilistic) sources of water to abstraction wells in a humid temperate climate, a scripted workflow was generated for groundwater model construction, data assimilation, particle-tracking, and post-processing. The workflow leverages existing datasets describing hydrogeology, hydrography, water use, recharge, and lateral boundaries to build the model. The workflow performs ensemble-based history matching and uses a posterior Monte Carlo approach to provide probabilistic capture zones describing areas that contribute recharge to wells in a risk-based framework. The water managers can then select areas of varying levels of protection based on their tolerance for risk of potential wrongness of the underlying models. All the tools in this workflow are open-source and free, which facilitates testing of this repeatable and transparent approach to other environmental problems. The specific data are available in the United States but the tools can be applied to similar datasets worldwide.

How to cite: Fienen, M., Corson-Dosch, N., White, J., Leaf, A., and Hunt, R.: From datasets to decisions – a repeatable workflow for groundwater decision support, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-10686, https://doi.org/10.5194/egusphere-egu22-10686, 2022.

14:10–14:15
|
EGU22-10933
|
Virtual presentation
Guillermo Martínez

The great environmental challenges we face as a society require the use numerical models to evaluate potential undesirable conditions using limited data and imperfect representation of natural and man-made systems. Iterative ensemble smoother (IES) methods provide not only a computationally efficient and robust approach to evaluate the uncertainty of numerical models and the worth of adding additional observation sources but a tool to test hypotheses about modeling assumptions and conditions in problems often too complex for traditional statistical methods. This work presents an approach to test hypotheses on undesirable outcomes in complex decision support models by taking advantage of the efficient and parallel capabilities of IES methods. The model used to illustrate the approach represents a groundwater system with more than 70,000 parameters used to describe hydraulic properties and boundary conditions and around 200 observation points. The version-controlled workflow used to implement the uncertainty analysis and hypothesis testing meticulously document each step of the analysis and takes advantage of open-source libraries to manipulate model files, implement the ensemble methods, and efficiently evaluate the fate of a contaminant of concern.  After a careful evaluation of conflicts between prior parameters and observations, the prior parameter distribution is updated using the ensemble-form of the Gauss-Levenberg-Marquardt algorithm (Chen and Oliver, 2013; White, 2018) to improve the match between simulated and observed data. A series of analysis were then performed to define reasonable metrics for hypothesis testing using spatial distance metrics. Results of the analysis illustrate the tradeoff between fitting the observations and the negative outcome under consideration as well as their impact on the realism of posterior parameter distributions. The computational performance and flexibility of this approach contributes to support robust, efficient, and reproducible decision-support modeling tools for water resources.

How to cite: Martínez, G.: Computationally efficient hypothesis testing using an iterative ensemble smoother and reproducible workflows, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-10933, https://doi.org/10.5194/egusphere-egu22-10933, 2022.

14:15–14:20
|
EGU22-5510
|
ECS
|
Virtual presentation
Reproducible large-scale groundwater modelling projects using the iMOD Python package
(withdrawn)
Joeri van Engelen et al.
14:20–14:25
|
EGU22-10403
|
ECS
|
Virtual presentation
Ross Kushnereit

Groundwater models are simplified and simulated depictions of aquifer systems that represent flow and/or transport of groundwater. For decades, practitioners would construct a model with the help of a graphical user interface (GUI) for a simulator such as MODFLOW, a finite-difference groundwater flow modeling program written by the United States Geological Survey (USGS). Due to the nature of GUIs, most of the time (and therefore cost) went in to the creation of model input datasets, and the “calibration” of the model would be hastily rushed at the end of the project time window. Unfortunately, the calibration process will almost certainly reveal issues derived from early stages of the project, and in the GUI framework it can take significant time and effort to manually address issues with the upstream workflow elements.

More recently, script-driven modeling workflow tools have been made available for practitioners to use to mitigate the time and cost associated with undertaking complex groundwater modeling analyses. Tools like FloPy and pyEMU python packages for interfacing with MODFLOW and the parameter estimation software PEST and PEST++.  When used together, these tools allow for all workflow steps to be automated, including the creation of the model input datasets as well as deployment analyses like data assimilation and uncertainty analysis.  More importantly, a script-driven approach allows issues (which statistically will always occur) to be addressed cleanly and efficiently, with minimal effort and little to no loss of practitioner time.

In this presentation, we present the development and implementation of a decision-support workflow for the history matching of 1,000 geostatistical realizations of transmissivity, storage, anisotropy, and recharge using multiple simulations. This work is part of a larger risk-analysis for the Waste Isolation Pilot Plant (WIPP) project in southeastern New Mexico. The predictive focus is on estimating particle travel times of long-lived radionuclides. 

Using script-driven approaches and recent iterative ensemble smoother techniques, our team was able undertake an advanced data assimilation analysis in a few hours one afternoon using a single workstation.  Previously, a similar analysis for the WIPP project took months of practitioner time on a massively parallel super-computer.

How to cite: Kushnereit, R.: Calibrating an ensemble of 1,000 realizations for estimating the uncertainty of aquifer properties in the vicinity of a long‐lived radioactive waste repository using a script-driven approach (on a Friday afternoon), EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-10403, https://doi.org/10.5194/egusphere-egu22-10403, 2022.

14:25–14:50
Q&A

Fri, 27 May, 15:10–16:40

Chairpersons: Anneli Guthke, Dirk Eilander

15:10–15:15
Session introduction

15:15–15:20
|
EGU22-5095
|
ECS
|
Virtual presentation
Miguel Leal et al.

The relationships between flow components and stream channel features during flash floods are theoretically well known and proven under controlled environments but are rarely explored and quantified for case studies in different geographic contexts. This research is focused on the spatial relationships between water depths and flow velocities for several return periods (RT), and how these are influenced by stream channel features. Spatial analyses were performed in Geographic Information Systems (GIS) using the hydraulic modelling results (HEC-RAS/HEC-GeoRAS) in a section of a small watershed in Portugal, which is frequently affected by flash floods. This section comprises about 1000 meters of the main watercourse (Barcarena stream) and the last 350 meters of one of its main tributaries (Massamá stream).

The relationships between water depths and flow velocities are not particularly evident in the floodable areas, although correlation coefficients increase with increasing return periods (0.39 for 5-year RT; 0.50 for 100-year RT). Water depths tend to grow with increasing flow velocities and vice versa. Nevertheless, this trend changes when high values of water depth or velocity are reached, preventing higher correlations. This inversion is explained by modifications in channel geometry, morphology or slope, the presence of confluences and obstacles, and flood width/overbank flooding. Unlike what happens with the entire floodable area, strong negative correlations between water depths and flow velocities were found along the stream centrelines. Correlation coefficients of -0.78 and -0.83 (2-year RT), and -0.66 and -0.87 (100-year RT) were determined for the Barcarena and Massamá streams, respectively. The more direct relationship in the tributary can be explained by the narrower channel when compared to the main watercourse and by the limited overbank flooding. Bed slope, channel and flood width, and roughness are highly relevant on the longitudinal variations of water depths and velocities and on the location of their maximum values. The relationships between water depths and velocities can also change in result of increasing peak discharges and return periods.

The 1D hydraulic model provided good results in the definition of floodable areas, water depths and longitudinal flow velocities. Lateral velocities are correctly represented in straight sections or in mildly curved bends, which are present in most of the study area, but there are errors in the sharp bends and at the confluence of the Barcarena and Massamá streams. The lack of hydrometric data compromises the calibration and validation of the velocity results. The non-existence of LiDAR elevation data in the floodplains and the lack of elevation data along the stream channels compromise the quality of the DSM. However, it was possible to overcome the lack of elevation data along the stream channels by including the position of the thalwegs in the DSM through the Topo to Raster tool of ArcMap. This guarantees the transversal and longitudinal variations of elevation, improving the hydrologic modelling results in areas with scarce or no elevation data along the channels. The obtained results demonstrate the usefulness of GIS to represent hydraulic modelling results and perform spatial analysis for flood events and other natural hazards.

How to cite: Leal, M., Reis, E., and Pinto Santos, P.: Flash flood spatial analysis using hydraulic modelling and Geographic Information Systems, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-5095, https://doi.org/10.5194/egusphere-egu22-5095, 2022.

15:20–15:25
|
EGU22-4195
|
Virtual presentation
Wolfgang Schmid et al.

Setting groundwater allocation limits requires an understanding of recharge fluxes to the aquifer system. Very often rainfall percolation through the subsurface represents the critical recharge flux. In this groundwater limit setting context, recharge estimates are often established as a component of the groundwater flow model history matching process. Typically, there are many recharge models available, and the basis for selecting any particular model is often confusing.

Some of these recharge models are numerical solutions of variably saturated pressure head and flow and represent the full complexity of the soil-vegetation-atmosphere transfer of water. Such models require many parameters that may not be measured or verified, and/or are computationally expensive. This can make the history matching process and predictive uncertainty analyses difficult.

Simpler model representations of the recharge processes are also available, either through upscaling (e.g., by lumping together different soil profiles, with different vegetation) and/or by simplification of the recharge estimation method. Such simplifications may involve empirical equations to derive gross recharge, single bucket-type root zone water balance calculations, or solving net recharge with the help of analytical solutions of flow or pressure heads and linear approximations of gross recharge or evapotranspiration from groundwater as function of the groundwater head. These simpler models often have a greater utility (i.e., they are quicker to run and are more numerically stable) but may be accompanied by additional ‘simplification’ induced uncertainty.

Regardless of the method used, the uncertainty and bias of these recharge predictions can be high.  The uncertainty of groundwater model predictions underpinning the setting of allocation limits can also be high. However, the performance of a recharge model in terms of how it impacts the reliability of the predicted impacts relevant to the groundwater allocation limit, is currently not considered. This study addresses this issue, exploring the costs and benefits of recharge models of varying complexity, in the context of setting groundwater abstraction limits. This is demonstrated using a synthetic, but realistic case study in Western Australia.

We adopt a paired complex-simple model analysis workflow, and implement it using the Flopy-PyEMU Python-based scripting framework. This workflow is then used to explore the performance of more complex and simpler models within the groundwater allocation management context by measuring each model’s bias and uncertainty. We compare a cell-by-cell Richards’ equation-based recharge model, with a series of simpler recharge contender models. This scripted workflow supports the efficient deployment of the paired complex-simple model stochastic analysis and interpretation of its outputs.

How to cite: Schmid, W., Moore, C., Hall, J., Dawes, W., Silberstein, R., Guzman, S., Siade, A., and Nelson, R.: Recharge model performance in the context of setting groundwater allocation limits, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-4195, https://doi.org/10.5194/egusphere-egu22-4195, 2022.

15:25–15:30
|
EGU22-5826
|
ECS
|
|
Virtual presentation
Cécile Coulon et al.

Numerical models and optimization algorithms can be valuable tools for decision-making in coastal and island aquifers, where pumping wells are threatened by salinization. Yet, the implementation of pumping optimization under uncertainty remains limited in practice, because of long simulation times and challenges associated with uncertainty propagation through series of models. A method was developed to optimize pumping rates in an island freshwater lens considering parameter, observation, and climate uncertainty. It was implemented in an island aquifer in the Magdalen Islands (Québec, Canada). A seawater intrusion model with rapid simulation times was developed using MODFLOW-SWI2. The iterative ensemble smoother algorithm implemented by PESTPP-IES allowed for history matching and nonlinear uncertainty quantification. The model predictive uncertainties were coupled with climate uncertainties, including recharge uncertainty (derived from various global circulation models and emission scenarios) and sea-level rise uncertainty. Using PESTPP-OPT, the pumping rates in the freshwater lens were then maximized while avoiding the risk of well salinization and considering parameter, observation, and climate uncertainty. Results of the pumping optimization were compared with estimates of water demand uncertainty. This study used widely available, model-independent software and could be used to support groundwater management decision-making in other insular or coastal areas.

How to cite: Coulon, C., Lemieux, J.-M., Pryet, A., and Gatel, L.: Optimization of pumping rates in an island freshwater lens considering parameter, observation, and climate uncertainty, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-5826, https://doi.org/10.5194/egusphere-egu22-5826, 2022.

15:30–15:35
|
EGU22-8976
|
Virtual presentation
Sreekanth Janardhanan et al.

Groundwater use for irrigation, stock and domestic purposes from shallow unconfined aquifers is rarely metered in most parts of the world despite significant increase in the rate of use in the recent decades. Most aquifers systems are poorly characterized and monitored rendering assessment of groundwater balance and informed management decision making difficult.

Advances in automated data collection through remote sensing and other technologies in the recent years, makes available pertinent data sets that can indirectly inform groundwater balance from which groundwater recharge and discharge can be estimated. Assimilating such data sets using traditional means would require simulating the physics across multiple domains including climate, surface water, unsaturated and saturated zones and may be untenable in most decision-making contexts.

Recent advances in Artificial Intelligence and Machine Learning techniques (AI/ML) have created the opportunity to estimate groundwater balance components probabilistically by considering the correlation and causal relationships with other climatic, hydrological and geospatial variables for which data sets are readily available, for example estimates of actual evapotranspiration from remote sensing data. Such machine learning-based models need not necessarily be underpinned by the explicit solutions of governing equations pertaining to the physical processes involved across multiple domains. These ML-based models can be used either independently or in combination with physically based models for retrospective or predictive assessments of groundwater balance components and quantification of recharge and discharge components including historical pumping rates from an aquifer.

This study develops ML-based groundwater balance estimators using machine learning based on suitable supervised learning algorithm for a selected unconfined aquifer system that spans across 16 districts in northwest Bangladesh region. The study uses daily precipitation data, evapotranspiration estimates using Moderate Resolution Imaging Spectroradiometer (MODIS) data, interpolated river stages, and weekly observed water levels from monitoring bores to train, test and validate a Deep Neural Network model implemented using PYTORCH.

Simulated groundwater levels obtained using the trained and tested ML models are used to estimate long-term groundwater storage changes in region and are compared to estimates from a numerical groundwater MODFLOW model developed and history-matched using Flopy and PEST++ frameworks. Both the ML and MODFLOW models are implemented for a rectangular grid with 1500 m × 1500 m cells.  The workflow scripted using PYTORCH and Flopy libraries enabled the ready comparison of ML and numerical models’ outputs and evaluate the applicability of ML models for groundwater balance simulation.

 

How to cite: Janardhanan, S., Pagendam, D., MacKinlay, D., Pena-Arancibia, J., and Mainuddin, M.: Groundwater balance estimators using Machine Learning , EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-8976, https://doi.org/10.5194/egusphere-egu22-8976, 2022.

15:35–15:40
|
EGU22-5966
|
Virtual presentation
Alexandre Pryet et al.

The assimilation and prediction of concentration data is often impeded by the computation time of groundwater transport models based on the resolution of the advective-dispersive equation. This is unfortunate because such data is often rich in information and the prediction of concentration values is of great interest for decision making.  Particle tracking may be used as an efficient alternative under a series of simplifying assumptions, which are often reasonable at groundwater sinks. A rapid transport model allows the use of assimilation and optimization methods requiring many model calls.  We developed a Python package to facilitate the use of the USGS MODFLOW6 and MODPATH7 models to simulate the transfer of tracer or contaminant concentrations to a groundwater sink (typically a pumping well or a drain). The approach requires the identification of one or a series of sources of tracer/contaminant such as a contaminated stream or area in the model domain. The package handles particle seeding around the sink and estimation of the concentration of water withdrawn from the sinks. Both “strong” and “weak” sources can be considered. Concentrations are computed with a mixing law from the particle endpoints and velocities. We investigated the best practice to obtain robust derivatives with this approach, which is essential for all methods based on the linearized version of the model. We provide a step-by-step workflow from model construction to parameter estimation, linear uncertainty analysis, and chance-constraints optimization with the PEST suite. The interest and practical details of the approach are illustrated on a well field vulnerable to a stream, and a parametric analysis is provided in order to evaluate the impact of key numerical parameters on the presented results.

How to cite: Pryet, A., Matran, P., Cousquer, Y., and Roubinet, D.: Particle tracking as a vulnerability assessment tool for drinking water production, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-5966, https://doi.org/10.5194/egusphere-egu22-5966, 2022.

15:40–15:45
|
EGU22-10013
Timofey Samsonov et al.

River hydrograph separation is one of the most important operations applied to the streamflow data. Numerous separation techniques and and their software implementations have been developed so far. In operational practice of Russian hydrological organizations and research institutes an event-based approach is commonly used for the hydrograph separation. Different meteorological events such as temperature transition through zero and rains are recognized in meteorological data, and then the corresponding changes in river hydrograph are identified, which eventually helps to attribute each peak in hydrograph with corresponding genetic component. The base flow component is traditionally defined according to Kudelin’s approach, taking into consideration different schemes of surface-ground water runoff interaction. In contrast, the most widespread separation approach in Western school is filtering-based. Lyne-Hollick, Maxwell, Boughton, Jakeman, Chapman and some more sophisiticated filters can be applied to separate the flow into quick and base. Results of two approaches are quite different, especially in terms of the baseflow component. In current study we present the updated open-source grwat R package, which puts both worlds together. It contains both the genetic event based and filtering-based hydrograph separation approaches with the ability to mix them together. In particular, applying the filtering-based separation inside the detected genetic events provides curve of the baseflow well corresponding to tracer-based studies. The second novelty of the package is the intellectual procedure for determination of the second-order events that complicate the freshet (seasonal) flood, such as rain floods. Finally, the updated package contains the internal spatial database of hydrograph separation parameters which is obtained over the European territory of Russia through experimental work. This database allows automated selection of the optimal separation parameters based on the location of the river gauge supplied by package user. The database can be extended to other regions of the world through collaborative work of package users.

The study was supported by the Russian Science Foundation grant No. 19-77-10032

How to cite: Samsonov, T., Rets, E., and Kireeva, M.: Region-specific multiple-approach separation of river hydrograph using the GrWat R package, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-10013, https://doi.org/10.5194/egusphere-egu22-10013, 2022.

15:45–15:50
|
EGU22-8764
|
Virtual presentation
Davíd Brakenhoff et al.

In recent years, meteorological droughts over Northwestern Europe caused severe declines in groundwater levels with significant damage to groundwater-dependent ecosystems and agriculture. One possible solution to reduce the declines in groundwater levels is to temporarily lower the extraction rates of nearby well fields used for drinking water production. The effectiveness of such measures depends on the magnitude and time of the response of the groundwater system to changes in groundwater extraction, which is salient information for decision makers. The response of the groundwater system is commonly quantified using numerical groundwater models that are time-consuming to develop and can be difficult to calibrate. In this research, a quick data-driven approach is proposed, based on time series analysis, that serves as a complement to more traditional groundwater modeling approaches. 

A scripted workflow was developed using Pastas, an open-source Python module for Transfer Function Noise modeling. The approach was applied to 243 monitoring wells in an area of the Netherlands, a country where summer droughts can cause serious problems, even though the country is better known for problems with too much water. For each monitoring well, the best model structure and relevant hydrological forcings (rainfall, evaporation, river stages, and extraction rates of well fields) were selected iteratively. Model selection was performed through split-sample testing and diagnostic checking. The accepted model for each monitoring well represents an independent estimate of the contribution of different hydrological forcings and processes to the groundwater response and is based exclusively on observed data. The modeled responses to the pumping rates of the well fields were used to determine the feasibility of reducing extraction rates to control heads during droughts.

How to cite: Brakenhoff, D., Vonk, M., Collenteur, R., and Bakker, M.: Application of time series analysis to explore the feasibility of reducing extraction rates to mitigate groundwater declines during summer droughts: a case study in the Netherlands, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-8764, https://doi.org/10.5194/egusphere-egu22-8764, 2022.

15:50–15:55
|
EGU22-10694
|
ECS
|
Virtual presentation
Felix Bruckmaier et al.

The Food and Agriculture Organization (FAO) AquaCrop model has demonstrated its ability to accurately simulate the growth of various crops. However, the quality of simulation results depends on the calibration of the model, which in turn requires field observations of model inputs and parameters. This limits the utility of AquaCrop in data-scarce regions, such as the Global South. A user-friendly method to analyze parameter sensitivities and model output uncertainties could facilitate the assessment of the model output reliability. The AquaCrop version provided by FAO, however, is run through a standalone graphical user interface (GUI) and therefore does not allow for systematic calibration. Besides, the user cannot customize parameter-specific features like irrigation scheduling.

This work presents a tool that enhances the MATLAB-based open-source application of AquaCrop, AquaCrop-OS (AOS), with the following functionalities: A Bayesian modeling feature is designed to calibrate the AOS model considering input data uncertainty, while the MATLAB toolbox Sensitivity Analysis For Everybody (SAFE) is integrated to automate sensitivity and uncertainty analysis. Irrigation schedules may now also be created dynamically and depending on different simulated parameters like the rooting depth. The user can distinguish between different environmental stresses, either by cause or affected variable. Every functionality is supplemented with intuitive graphics. The tool will be released under an open-source license on GitHub. A standalone executable version with a GUI will cater for non-MATLAB users.

The AOS model is calibrated on field data from an experimental agricultural plot in the Ganga River basin in Kanpur, India, for two wheat cropping seasons between 2018 and 2019. The proposed tool is used to quantify the uncertainty in the model input data and parameters and their effects on model outputs.

How to cite: Bruckmaier, F., Adla, S., Disse, M., and Tripathi, S.: Modified AquaCrop-OpenSource tool for data-scarce regions, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-10694, https://doi.org/10.5194/egusphere-egu22-10694, 2022.

15:55–16:00
|
EGU22-12971
|
ECS
|
|
Virtual presentation
Seelam Naga Poojitha and Vinayakam Jothiprakash

Water distribution network (WDN) is an essential infrastructure for conveying potable water to communities. Constituting different components, it has a complex structure involving significant financial investments for its design. During its life span, the failure of WDN partially or entirely is an inevitable consequence of the network's hydraulic or mechanical uncertainties. Therefore, the WDN design problem naturally involves a tradeoff between the reliability and cost aspects. The present study formulates a reliability-based hybrid metaheuristic optimization model for its robust design. Primarily, the proposed framework is composed of three components, an optimization algorithm, a simulation model, and a reliability assessment model. 
A novel hybrid technique in a combination of differential evolution (DE) and krill herd algorithm (KHA), DE-KHA, is used as the optimization algorithm. The DE-KHA is the computationally efficient algorithm for effortlessly tackling the WDN design problems by balancing the exploration and exploitation features. EPANET 2.0 hydraulic simulator that performs the hydraulic analysis of WDN is used as a simulation model. The hydraulic characteristics of the network, such as flow-through pipes, unit headloss, the actual and total pressure head at the demand nodes, and the demands delivered to the demand nodes, are assessed using EPANET 2.0. The reliability model evaluates the network's performance under mechanical uncertainties. The mechanical failures are the scenarios of networks component failure, which in the present study are considered network pipe outages. The reliability model is based on the minimum cut set method, where the pipe failure combinations that cause the failure of the network are found explicitly considering the minimum pressure head requirement at the demand nodes.
Search for the optimal solution is progressed by the optimization technique, where the constraints of continuity and energy balance equations are explicitly taken care of by the EPANET 2.0 simulation model. Then considering the hydraulic head at the demand node, the minimum cut sets are finalized, and the network’s performance under mechanical failure scenarios is assessed using the reliability model. The DE-KHA and reliability model code is written in MATLAB and linked to EPANEt 2.0 using MATLAB-EPANET toolkit.
The application of the developed framework is validated considering Two loop Network (TLN). It is a hypothetical network studied by many researchers for validating their optimization models. TLN is made up of eight pipes, connected by seven nodes with a single reservoir of 210 m total head that feeds the entire network. Thus, it is a gravity-fed network with no pumps operated. Considering this simple case study, yet a challenging problem with 148 possible solutions in the search space, the reliability-based model is validated. The results present the computational efficiency of the model in yielding the optimal design cost of $ 419,000 with minimal computational effort. Furthermore, the algorithm proposed is efficient in exploring various alternate optimal solutions with considerable reliability, thus, presenting robust design options for TLN. Considering the efficiency of the proposed model, the study suggests it for the robust design of real-life WDNs.

Keywords: Water distribution network design; Reliability; Mechanical uncertainty; Metaheuristic algorithms 

 

How to cite: Naga Poojitha, S. and Jothiprakash, V.: Reliability-Based Hybrid Metaheuristic Optimization Model for the Design of Two Loop Network Under Mechanical Uncertain Scenarios, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-12971, https://doi.org/10.5194/egusphere-egu22-12971, 2022.

16:00–16:05
|
EGU22-11823
|
ECS
|
Virtual presentation
Syed Md Touhidul Mustafa et al.

Particularly in the Nordic region, water excess and shortage (drought) are becoming more frequent phenomena that challenge the development of agriculture and crop production. Identification of appropriate water management strategies is essential (i) to ensure sustainable water resources management for crop production and the functioning of healthy ecosystems; and (ii) to improve resilience to hydrological extremes. Integrated hydrological models offer that potential through understanding and forecasting of hydrological systems under anthropogenic and climatic influences, and providing information for improved decision-making. This study aims to develop a decision support instrument based on integrated hydrological modelling to identify appropriate management solutions and improve field- and catchment-scale water management in Nordic agriculture. The study area is Tyrnävä catchment, located in the northern part of Finland near Oulu city. Initially, the available hydro-climatological and hydrogeological data of the Tyrnävä catchment are characterized in detail. Then the hydrogeological parameters of the model are identified based on existing hydrogeological, climatic and remotely sensed data and their spatial, temporal and vertical variability. Next, a regional integrated surface-subsurface hydrological model is set up using HydroGeosphere. After successful calibration and validation using observed groundwater level, river discharge and soil moisture data, the model will be used in implementing and evaluating different management strategies (e.g., different irrigation options during droughts and controlled drainage management) for the future and their influence on the surface and groundwater systems. Uncertainty arising from different sources will be quantified using the Integrated Bayesian Multi-model Uncertainty Estimation Framework with the support of a supercomputer to improve the reliability and accuracy of the decision support instrument. Additionally, stakeholders’ involvement through local workshops is ensured throughout the modelling study, from the beginning to obtain reliable and useful decision support. Finally, based on these results, informed decisions regarding the appropriate water management can be made, which is important for sustainable water resources management for crop production and the functioning of healthy ecosystems particularly in Nordic agriculture.

How to cite: Mustafa, S. M. T., Autio, A., Haghighi, A. T., Marttila, H., Avellan, T., Schilling, O. S., Brunner, P., Scholz, M., and Klöve, B.: Integrated hydrological modelling for decision support to improve field and catchment scale water management in agriculture, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-11823, https://doi.org/10.5194/egusphere-egu22-11823, 2022.

16:05–16:10
|
EGU22-10747
|
ECS
|
Virtual presentation
Lee Chambers et al.

The low-lying coastal urban area of South Dunedin, New Zealand, is particularly susceptible to the impacts of sea-level rise, which is projected to rise by as much as 1.2 m by 2100 under high emissions scenarios.  Currently, more than 2,500 homes are < 50 cm above mean sea level and groundwater levels are typically < 1 m below the surface.  As sea levels rise, groundwater levels are also predicted to rise, increasing the probability of inland groundwater inundation (groundwater flooding) throughout South Dunedin.  It is therefore imperative to develop an improved understanding of the physical controls, and the uncertainty associated with these controls, on the occurrence and severity of the groundwater inundation hazard caused by rising sea levels.  We deploy a simple and fast-running model within a highly-parametrised Uncertainty Quantification (UQ) workflow to investigate the adequacy of steady-state-only versus transient calibration when assessing the risks of groundwater inundation.  The decision to proceed beyond a steady-state-only calibration is time-consuming and costly (often vastly so) and requires careful attention and further research in practical application.  The reduction in uncertainty of decision-relevant forecasts accrued through implementing a transient calibration procedure (or lack thereof), given existing and yet to be acquired data, is the metric by which the modelling is judged.  Firstly, the workflow involves history matching and uncertainty analysis implemented through PESTPP-IES to explore and reduce the uncertainty of decision-relevant forecasts (spatial groundwater elevation and drain fluxes).  Secondly, a paired complex-simple model analysis is used to: explore 1) the potential uncertainty reductions in decision-relevant forecasts achieved through transient calibration and 2) the potential introduction of unquantifiable bias of decision-relevant forecasts introduced by the competing calibration procedures.

How to cite: Chambers, L., Hemmings, B., Moore, C., Cox, S., Levy, R., and Knowling, M.: Decision-support modelling for an uncertain future: developing forecasts of sea level rise impacts on groundwater, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-10747, https://doi.org/10.5194/egusphere-egu22-10747, 2022.

16:10–16:35
Q&A

16:35–16:40
Session closing