The question of how much time and resources are required to implement, run and postprocess stochastic groundwater simulations is key for the purpose of making stochastic groundwater modelling a common practice and profitable activity. Three important factors are presented below:
1) Do groundwater modellers need to considerably modify their existing modelling workflows to perform stochastic simulations?
Chances of successfully implementing stochastic groundwater models decrease if important modifications to the existing modelling workflows are required. These modifications will increment the risk of failing to deliver modelling results in reasonable times, which could cause even major repercussions if the main results expected by the client are exclusively associated to the stochastic component of the model.
A more convenient approach is to modify the existing modelling workflows as little as possible, and to think of the stochastic simulations as a complement to the traditional deterministic modelling results. In this way, consulting companies can offer stochastic groundwater modelling as an additional task or as a continuation of traditional modelling projects, instead of a new and uncertain type of project that conditions the development of a probably expensive numerical model around stochastic simulations.
2) Are stochastic computational times compatible with commercial projects?
Stochastic simulations take time, which can be considerably reduced by the use of parallel and distributed computing. It is concluded that any stochastic modelling workflow has to include the option of performing parallel simulations. That said, there is an additional critical factor to consider: where in the traditional modelling process should the stochastic modelling begin? This question is intrinsically related to the initial question of how much existing modelling workflows need to be modified to perform stochastic simulations.
A convenient option is to start the stochastic modelling process in the latest stages of the modelling project, ideally after finishing the calibration and deterministic predictive simulations. This approach requires minimum changes to the existing modelling workflows, taking the stochastic simulation process out of the iterative modelling loop of "conceptual model - calibration - sensitivity analyses - predictive simulations", which might be completely incompatible with stochastic modelling times. Alternatively, it allows to implement stochastic simulations using previously developed models, which opens new business opportunities not only with current but also with past clients.
This approach has the additional benefit of starting the stochastic simulations after finishing the deterministic simulations, which are considered to be the traditional main deliverable. If the stochastic results are consistent with the deterministic results, then the traditional deterministic approach was adequate and the stochastic results contribute to demonstrate the robustness of the conceptual and numerical model, adding quantitative estimations of the model uncertainty which could ultimately allow to readjust any adopted safety factors. A properly developed model will very likely follow this route. If the stochastic results lead to different conclusions than the deterministic results, then the stochastic simulations can be considered as a key contribution to the project, and a revision of the deterministic and stochastic components of the model has to be made to understand the differences.
Postponing the stochastic modelling process to the latest stages of the project has the inconvenient of starting to condition the stochastic results to decisions taken during the model construction and calibration. My personal opinion is that the practical benefits of the stochastic modelling approach described in the previous paragraphs exceed this type of disadvantages, especially considering the current limited use of stochastic groundwater modelling within the industry. Independent of this, more advanced stochastic modelling methodologies that try to address this specific issue exist, like the utilization of stochastic multi-model approaches. These modelling techniques are however more expensive.
3) Do modellers need to develop their own pre and postprocessing modelling routines?
Although it is many times interesting and beneficial to develop in-house processing routines, having access to a robust, easy to use and widely available stochastic modelling software could make an important difference both from a technical and business perspective. Developing adequate in-house codes takes time, especially if these codes need to be flexible enough to be compatible with diverse modelling objectives and configurations, and efficient enough to process up to tens or even hundreds of gigabytes of stochastic modelling results in reasonable times.
All of this said, the main challenge for relying exclusively on in-house routines is that clients will very likely prefer the utilization of validated, easy to use and widely available software packages, that do not tie their projects to a specific consultant. Furthermore, one can argue that the existence and continuous improvement of these type of easy to use - widely available software is what will end up making stochastic groundwater modelling a common practice in the future.
1) Do groundwater modellers need to considerably modify their existing modelling workflows to perform stochastic simulations?
Chances of successfully implementing stochastic groundwater models decrease if important modifications to the existing modelling workflows are required. These modifications will increment the risk of failing to deliver modelling results in reasonable times, which could cause even major repercussions if the main results expected by the client are exclusively associated to the stochastic component of the model.
A more convenient approach is to modify the existing modelling workflows as little as possible, and to think of the stochastic simulations as a complement to the traditional deterministic modelling results. In this way, consulting companies can offer stochastic groundwater modelling as an additional task or as a continuation of traditional modelling projects, instead of a new and uncertain type of project that conditions the development of a probably expensive numerical model around stochastic simulations.
2) Are stochastic computational times compatible with commercial projects?
Stochastic simulations take time, which can be considerably reduced by the use of parallel and distributed computing. It is concluded that any stochastic modelling workflow has to include the option of performing parallel simulations. That said, there is an additional critical factor to consider: where in the traditional modelling process should the stochastic modelling begin? This question is intrinsically related to the initial question of how much existing modelling workflows need to be modified to perform stochastic simulations.
A convenient option is to start the stochastic modelling process in the latest stages of the modelling project, ideally after finishing the calibration and deterministic predictive simulations. This approach requires minimum changes to the existing modelling workflows, taking the stochastic simulation process out of the iterative modelling loop of "conceptual model - calibration - sensitivity analyses - predictive simulations", which might be completely incompatible with stochastic modelling times. Alternatively, it allows to implement stochastic simulations using previously developed models, which opens new business opportunities not only with current but also with past clients.
This approach has the additional benefit of starting the stochastic simulations after finishing the deterministic simulations, which are considered to be the traditional main deliverable. If the stochastic results are consistent with the deterministic results, then the traditional deterministic approach was adequate and the stochastic results contribute to demonstrate the robustness of the conceptual and numerical model, adding quantitative estimations of the model uncertainty which could ultimately allow to readjust any adopted safety factors. A properly developed model will very likely follow this route. If the stochastic results lead to different conclusions than the deterministic results, then the stochastic simulations can be considered as a key contribution to the project, and a revision of the deterministic and stochastic components of the model has to be made to understand the differences.
Postponing the stochastic modelling process to the latest stages of the project has the inconvenient of starting to condition the stochastic results to decisions taken during the model construction and calibration. My personal opinion is that the practical benefits of the stochastic modelling approach described in the previous paragraphs exceed this type of disadvantages, especially considering the current limited use of stochastic groundwater modelling within the industry. Independent of this, more advanced stochastic modelling methodologies that try to address this specific issue exist, like the utilization of stochastic multi-model approaches. These modelling techniques are however more expensive.
3) Do modellers need to develop their own pre and postprocessing modelling routines?
Although it is many times interesting and beneficial to develop in-house processing routines, having access to a robust, easy to use and widely available stochastic modelling software could make an important difference both from a technical and business perspective. Developing adequate in-house codes takes time, especially if these codes need to be flexible enough to be compatible with diverse modelling objectives and configurations, and efficient enough to process up to tens or even hundreds of gigabytes of stochastic modelling results in reasonable times.
All of this said, the main challenge for relying exclusively on in-house routines is that clients will very likely prefer the utilization of validated, easy to use and widely available software packages, that do not tie their projects to a specific consultant. Furthermore, one can argue that the existence and continuous improvement of these type of easy to use - widely available software is what will end up making stochastic groundwater modelling a common practice in the future.
