TUFLOW General Discussion

From Tuflow
Revision as of 14:00, 30 May 2022 by Chris Huxley (talk | contribs) (Created page with " =Frequently Asked Questions (FAQ)= == Why are model results developed in an older release different to a newer release? == If comparing a Classic model with HPC, also check t...")
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

Frequently Asked Questions (FAQ)

Why are model results developed in an older release different to a newer release?

If comparing a Classic model with HPC, also check the Will TUFLOW HPC and TUFLOW Classic results match? page in addition to this answer.
In addition to the above, there are reasons why model results would be different between different TUFLOW releases, whether it is the Classic or HPC solver as follows:

  • General improvements and fine-tuning of the solution scheme, especially for the more complex hydraulic physical terms and situations such as: sub-grid turbulence representation; treatment of shocks (e.g. hydraulic jumps); and transitioning between sub-critical and super-critical flow on steep slopes.
  • Some new functionality can cause a significant change in results. For example:
    • Sub-Grid Sampling (SGS) applied to an existing model that used a too coarse cell resolution in high flow areas of highly variable topography (relative to the 2D cell size). SGS will greatly improve the model's ability to convey water accurately in these situations with vastly improved results.
    • New default sub-grid turbulence scheme in the 2020 release of TUFLOW HPC that is cell size independent and allows modellers to use cell sizes much smaller than the flow depth across all scales from flume to large rivers. For more information on differences between Smagorinsky scheme (HPC releases up to 2020) and the new Wu turbulence scheme (2020 onwards) see here.
  • Changes to the default settings and values, e.g.:
    • different default eddy viscosity formulation and/or coefficients,
    • improved data pre-processing approaches such as sampling materials on cell mid-sides instead of cell centres,
    • and many others.
    • For backward compatibility the Defaults == command is available to run old models on new releases to replicate past results (note, sometimes full backward compatibility cannot be catered for due to different code compiler and updates that can't be reverted, especially for several releases earlier).
  • New features that use GIS attributes previously reserved (i.e. unused). If these attributes were not populated with the recommended “reserved” value (usually 0 or blank), then they can cause unpredictable results in later releases.
  • Bug fixes noting that most bug fixes are input/output related and rarely affect the model's hydraulic calculations.
  • Change in timestepping can also produce a small change in results. HPC uses the Runge-Kutta 4th order integrator, which is usually fairly insensitive to time step provided the model is running stably. However when a region is filled by flow that only just overtops an embankment, a 10 mm difference in water levels upstream of the embankment can create a much larger difference in levels downstream. Hence small differences in time-stepping (along with many other aspects of model setup) can trigger local differences in model results.
  • Model orientation (if changed) could also mean slight change in results. This is mostly given by interpolating values from different calculation points. Every cell has nine calculation points. Based on the model origin, all or most of the calculation points would have different topography elevation sampled, which translates to slightly different results.
  • If using 1D channel, possibly different cells have been selected as HX boundary and might have different elevations. This can be reviewed in 1d_to_2d check file.

Generally, there should not be substantial differences as the fundamental equations being solved are unchanged and TUFLOW Classic and HPC solvers have always solved all the physical terms using a 2nd order spatial approach. The one exception is the turbulence (eddy viscosity) representation, which is the most complex and challenging to solve of all the physical terms (many 2D schemes simply omit this term). If significant differences (>10% of depth change across the whole model) are observed then it’s most likely due to the first four dot points above. To identify in which release(s) the significant changes occurred, the model can be run with the latest build and for past releases. The changes for each release are documented in their release notes. Past releases and release notes are all available here. Once the exact release where the changes occurred is tracked down, individual features can be turned off to narrow down the cause.

The recommendation is usually for new or reworked models to use the newest build to take advantage of the latest features and enhancements, some level of calibration might be required for reworked models. The new TUFLOW executable is not different from the previous ones in the meaning that any existing model should be re-calibrated if there are available calibration data. However, particularly if a model is already calibrated, using prior builds of TUFLOW or winding back default settings using Defaults == command is considered reasonable for established models that are to be used for minor tasks and an update of the model would not be cost effective.

What is the difference between single and double precision and when should I use them?

For each release of TUFLOW, both single and double precision engine versions are available. In the executable filename the single precision version of TUFLOW will include iSP and iDP for the double precision version. The double precision won't make the results twice as good, it stores numbers as 8-byte real numbers (15-17 significant figures) as opposed to 4-byte (6-9 significant figures).

When to use single or double precision depends on the solution scheme:

  • TUFLOW Classic uses water level as the primary variable within the hydrodynamic solver and double precision is typically necessary when using direct rainfall or with model elevations above 100m. In these cases, numerical precision (rounding errors) can cause mass conservation errors. For example, a small rainfall (e.g. 1mm/hr) converted to metres / second (~2.78e-7) may be lost through numerical precision and result in accumulated mass balance error, specially for longer model run time.
  • TUFLOW HPC uses cell-averaged water depth as the primary variable within the solver, rather than using water surface elevation as the primary variable and computing water depth on the fly from surface elevation minus bed elevation. This means that precision issues associated with applying a very small rainfall and/or modelling high elevations are not applicable in HPC. Unless testing shows otherwise, the single precision version of TUFLOW should be used for all HPC simulations. An error message will be triggered if TUFLOW HPC is used with double precision unless HPC DP Check == OFF is specified within the TCF. The need to use double precision with HPC could occur when the coupled ESTRY 1D engine requires the use of the double precision solver to achieve better stability in 1D with more significant numbers. This usually happens with carved 1D channels within the 2D domain, either the 1D channel itself or the boundary links between 1D and 2D domain are causing the mass error. On rare occasions, models with higher elevations and small QT inflow would also require to run in double precision, because QT boundaries have hidden 1D node and as such are solved in the 1D ESTRY engine.

The single precision version of TUFLOW uses significantly less memory (RAM) and is about 20% faster for TUFLOW Classic and four times faster for HPC. Unless required otherwise, the single precision version of TUFLOW is recommended. A good step in the model development is to run the model with both the single and the double precision and if the results / mass balance are similar then the single precision version is sufficient.

Advanced reasoning to use single precision with TUFLOW HPC:

  • HPC is an explicit finite volume scheme which is mass conserving to numerical precision.
  • The HPC scheme uses 4th order time integration, which means the simulation completes in fewer time steps compared to 1st or 2nd order time integration schemes.

I am running existing and developed case and see differences away from the model changes. Why?

Any geometry changes between models, no matter how small, will affect results, sometimes to a greater degree than that occurring in the area of change. For example, a few millimetres increase in water level can determine whether or not overtopping of an embankment occurs, and this can consequently cause even larger impacts on the downstream side of the embankment. Furthermore, these changes can be compounded by subsequent changes in timestepping when using the adaptive timestepping option (the default in TUFLOW HPC), especially at fringes of the flood extent, where cells are constantly wetting and drying. Modellers and reviewers should be judicious and pragmatic when assessing which impacts are real and which are numerical noise.
Suggestions:

  • Use the latest TUFLOW HPC release available.
  • Check that cell size is appropriate to the modelling exercise.
  • Use depth varying manning's n (lower manning's n for shallow water depths), specifically for direct rainfall models.
  • Set appropriate Map Cutoff Depth for the modelling task. e.g. direct rainfall models might have higher values to avoid undesirable noise on the wet/dry interface.

Why seemingly identical models can produce non-identical results?

Generally speaking single path numerical solvers such as those used for hydraulic modelling should be able to produce the same numerical results twice to the last bit of every binary number calculated and output. However, this can become difficult with parallel computations as the order in which a list of single or double precision numbers are summated can produce slightly different rounding errors and thereby produce very slightly different results. For the vast majority of models TUFLOW Classic, TUFLOW HPC and TUFLOW FV will reproduce numerically identical results when run on the same CPU/GPU. Occasionally this might not be the case when identical simulations are run on different CPUs/GPUs due to hardware differences.
Prior to 2020-10-AB release, the new boundary method introduced in TUFLOW HPC 2020-01-AA release for inflowing HT and QT boundaries (refer see Section 6.1 of the 2020 Release Notes) can in rare situations be affected by bitwise reproducibility when parallelised. When this issue occurs, very slight numerical differences can occur throughout the model, noting that they will be of a much smaller magnitude than those that occur when carrying out impact assessments, but will cause undesirable numerical noise in the impact mapping.

Do I need TUFLOW licence to create TUFLOW inputs and view results from TUFLOW models?

No, a TUFLOW licence is only needed to run simulations. All TUFLOW inputs and outputs use free open formats that are read and editable by third party software, for example QGIS and Notepad++:

How closely would TUFLOW results match other hydraulic software?

Different software will give different results for the simple reason that they all include different calculation assumptions. Understanding what those assumptions are and how they influence results will be important for the sensitivity testing. TUFLOW, like all hydraulic modelling software, needs to be applied appropriately and models should be calibrated to real world events if calibration data are available.

Can TUFLOW's 1D engine be used for modelling complex pipe hydraulics?

TUFLOW can model complex pipe hydraulics with a level of accuracy similar or better than industry peers. There are a few notable features that place TUFLOW ahead of other software:

  • TUFLOW's 1D solution accurately accounts for both non-pressurised and pressurised flows within the pipe network.
  • TUFLOW's treatment of pipe junction losses is one of the most sophisticated. The default method (Engelund) will adjust losses dynamically every timestep of the simulation based on the hydraulic conditions at that time and the following:
    • changes in pipe size
    • expansion/contraction if there is a manhole at the pipe junction
    • variation in pipe approach and exit angles at junctions
    • variation in pipe approach and exit elevation at junctions
  • Alternative loss methods to Engelund are also available, such as Fixed losses. The Fixed method conforms with some industry guidelines, such as the Qld Urban Drainage Manual (QUDM). Fixed losses are not set as the default as this generally requires the modeller to manually enter appropriate values at every manhole, whereas the Engelund approach in TUFLOW, which is based on that in MIKE Urban with several improvements developed in conjunction with Gold Coast City Council’s infrastructure team, provides an excellent automatic approach with no or minimal user input beyond the pipe and manhole geometry. The other advantage of the Engelund approach is that it is dynamic and adjusts losses according to the flow conditions, whereas the Fixed approach assumes the same energy loss coefficient for all flow regimes. TUFLOW also allows having a mix of different methods in the one model, for example, there may be a special manhole where the Fixed or other approach needs to be applied.
  • There are numerous pit inlet options, from automatic capture rates to manually defined depth-discharge relationships. In all cases the 2D cell water depth at the inlet influences the amount of flow entering the pit, and as such the 1D underground pipe network.
  • The 2020 TUFLOW release offers sub-grid topography sampling to process all elevations within the cell into a depth/volume relationship for its calculations. This approach ensures much more accurate water depth estimations at pit inlets, even if the 2D cell resolution is much larger than the geometry of the drain at the inlet. This in turn translates to more accurate representation of the pit inflow, and as such flow through the entire pipe network. No other 1D/2D stormwater drainage modelling software offers this functionality. The new Quadtree functionality also allows the user to model key flowpaths, such as road drains, in high resolution.
  • The 2D overland approach used by TUFLOW ensures any above ground inundation is defined by the model topography. This approach avoids any engineering judgement flow path definition mistakes which the 1D overland software suffer from.


Up
Go-up.png Back to TUFLOW Modelling Guidance
Retrieved from "https://wiki.tuflow.com/w/index.php?title=TUFLOW_General_Discussion&oldid=27787"