We use cookies to make your experience of our website better. You can delete or block cookies, but some parts of this website won't work without them. By using this website you accept our use of cookies.

11/12/2018 16:21:17

Answers to the questions posted during the webinar are now available! View the recording here

Please see responses from Konrad Adams (KA) and Jon Lloyd (JL) below:

Mass Balance Questions

Q: If working on a bigger model with a inappropriately high mb error, how would you determine which part causes the mb issues?

JL: When looking at large scale models identifying where errors are occurring can be difficult. Simple checks of flow continuity are always really useful; ask whether you can explain any gains/losses in flow down the model, if you can't, maybe this is your mass balance problem. Also use the spatial diagnostics from the .zzd to pinpoint areas of non-convergence.

Q: Can you say something about how you calculate mass balance across 1d/2d links for both FMP1d/2d and FMP1d/Tuflow models.  It seems like people often check the 1d and 2d domains individually, but not the whole model.

(combined response with Q below)

Q: Is there a plan to have mass balance reporting for linked 1D-2D models for the while linked system? Whole system meaning FM(1D) linked with Tuflow(2D) combined reporting. Because volume can get lost or gained at the boundaries between the two

KA: Yes. The example below which I had shown was from a FM-TUFLOW model

Figure 1.png


The “lat. link inflow” is flow which has crossed the 1D-2D link boundary from 2D to 1D, so the total inflow into the 1D component is boundary inflow + lat link inflow. Similar/reverse is true for the outflow

Note that the “(…of which linked)” row is included within the boundary inflow (or outflow, as appropriate) in the preceding line, noting that the linked flow through a TUFLOW SX connection, for instance, is via an HTBDY unit in FM, and is included within the boundary flow.

At present, we don’t take the volume from TUFLOW, or any boundary information, so don’t do a combined mass balance (though the linked flow value should be comparable with what TUFLOW reports). I shall look into the possibility of doing so. For FM 1D-2D, we do however do this in a similar summary displayed in the 2D .log file – the 1D, 2D and combined 1D-2D initial and final volumes, boundary flow and linked flow are all displayed. Note that the combined link flow (1D linked inflow + 2D linked inflow) should ideally be zero.

Q: Could you please touch base on the mass balance and the precision in which the model is being run in?

KA: In certain situations, running models in single precision can lead to mass balance errors that would not occur when running a model at double precision. This is particularly the case when running models with large reservoirs, where a small increase in water level in reservoirs with very large surface areas can be missed when running your model in single precision. The model will see the appropriate changes in inflow or outflow but volume of the reservoir may not increase (as the volumetric increase in storage from a 0.001m rise in water levels is very small in comparison to the total volume in the reservoir). If you encounter this in your model then try running your model in double precision mode to see if this helps to resolve your mass balance errors.

Q: Can you explain "cumulative error" and how this is presented in the Flood Modeller diagnostics?

KA: This would be for TUFLOW output. The volume error is consistent in how it’s calculated – it is always the difference between net inflow volume and increased storage volume. The difference between cumulative and instantaneous error is that the former is since model start, and the latter only applies to the last time step/interval. The other difference is what it’s given as a percentage of; for TUFLOW (and it does this for ESTRY [TUFLOW 1D], 2D and both), it’s as a proportion of the larger of system volume and inflow+outflow volume. In Flood Modeller (we call it “mass error”), it’s either as a proportion of the max system volume (#1, default), or as a proportion of the total boundary inflow or outflow (#2). Because the “what it’s compared with” is reasonably arbitrary, or can depend on what one is modelling as to which makes more “sense”, I think it’s very important to know/understand the volume numbers themselves, which both FM and TUFLOW also supply.

Q: Where next for Flood Modeller mass error reporting, will there be any spatial diagnostics or output file for the time variant data?

KA: What seems to be a popular suggestion is to report the mass balance error on a spatial resolution scale (i.e. not just the overall model). As mentioned, this would help drill down where the main causes of any issues would lie.

Also, a combined FM1D-TUFLOW error looks like it would be beneficial to modellers.

These suggestions are being considered for future development.

Q: What is an acceptable mass balance error?

KA:  This is a question that gets asked a lot. I can’t give you an exact number and say that this is an acceptable mass balance error as it needs to be looked at in context with other model outputs. It is a good top level value to check, however you can have a poor model with less than 1% mass balance error but you can also have a model that has a high mass balance error without there being a significant problem (for example when running a model with multiple tidal cycles, where the error per tidal cycle may be small but over time the cumulative error becomes large).

You can read more about mass balance error in a previous ‘Tip of the month’ here

Case Study Questions

Q: Have manhole units been tested for Model X?

JL: These weren't tested. But the orifice units were actually allowing flow into the 2D domain and so there was no need to replace with manhole units. If they were active, this would be a good sensitivity test though!

Q: For Model Y, which ME diagnostic are we looking at?

JL: The ME diagnostic was as a proportion of max. system volume. As opposed to the Mass Balance [2] which is reported as a proportion of the total inflow boundary.

Q: [For model Y] Was the event 1 based on a design flow or on the actual event?

JL: All four events were calibration using real-time data.

Privacy & Terms