Jump to content
GMS, SMS, and WMS User Forum
sukark

Calibrating & validating a Tuflow model

Recommended Posts

Hello everyone!!

I am trying to learn and know how to calibrate and validate a 2D TUFLOW model, after the simulation is run. I have daily averaged discharge and water level data for 4 months which i used to run the simulation. Now my aim is to calibrate and validate this model. But how to do this??I am new to this world and don know much and have idea about doing this properly. So i need your suggestions, informations and support.

Thanks

Share this post


Link to post
Share on other sites

Hi Sukark,

There are some tools in SMS that you can use to obtain numbers to compare with. The primary thing you need from SMS is to extract time-series data for the places you have you calibration data. This is done using the observation coverage. Profile plots will give plots of datasets at specific times along an arc, time-series plots will give plots through time of water surface at a point or flowrates across an arc. There is a tutorial on the observation coverage but unfortunately, it is geared towards steady state models. However, you may want to look at it as it does have a small section on time-series plots.

You can create transient measurements and put the observation data into them and show them on the time-series plot. You can also get the data into a spreadsheet program for comparison. To do this right-click on the plot and choose the command that is something along the lines of "view data." You can copy/paste using Ctrl-C/Ctrl-V to get the data into your spreadsheet software. Once there, you can create plots or analysis using spreadsheet tools.

I hope this helps.

Hello everyone!!

I am trying to learn and know how to calibrate and validate a 2D TUFLOW model, after the simulation is run. I have daily averaged discharge and water level data for 4 months which i used to run the simulation. Now my aim is to calibrate and validate this model. But how to do this??I am new to this world and don know much and have idea about doing this properly. So i need your suggestions, informations and support.

Thanks

Share this post


Link to post
Share on other sites

Hello Rusty!!

Thanks a lot for your suggestions and information. I went through the Observation coverage tutorial u have mentioned and found it to be informative and useful. You have mentioned "unfortunately, it is geared towards steady state models".But could you please elaborate in detail what's the difficulty for following this tutorial as my guide for my model calibration?

I look forward to your response

Thanks

Sudhir

Share this post


Link to post
Share on other sites

How can i create transient measurement? I have one water level data measured in the river from another location between upstream and downstream boundary. how can i use that data to create transient measurement?

Share this post


Link to post
Share on other sites

Create a point, select it, right click and choose attributes. In the top part of the dialog where measurements are specified, you need to click on the toggle that says "trans" or something like that. That will put a button in the bottom spreadsheet for each point location. Clicking on this button will allow you to specify a curve of measured data for the location.

How can i create transient measurement? I have one water level data measured in the river from another location between upstream and downstream boundary. how can i use that data to create transient measurement?

Share this post


Link to post
Share on other sites

Hello Rusty!!

Is it worth to run the simulation for a day assuming it as a steady state flow and follow the tutorial for model calibration. But as the flow behaviour in reality is not steady. what do you suggest??

Share this post


Link to post
Share on other sites

Hi,

I would suggest using the best BC information that you have, extracting curve data using time-series observation plots, copy/paste this data to a spreadsheet program (right-click on plot and choose view values), add a series of points that represents the average daily value in the middle of the 24-hour period it represents and do a qualitative estimate of how well your values match. I would also use the spreadsheet program to compute average values for the same 24-hour periods and see how well they line up.

I wouldn't recommend using steady state for calibration when your measured values were not extracted from a steady-state problem.

Cheers,

Rusty

Hello Rusty!!

Is it worth to run the simulation for a day assuming it as a steady state flow and follow the tutorial for model calibration. But as the flow behaviour in reality is not steady. what do you suggest??

Share this post


Link to post
Share on other sites

Hello Rusty!!

Thanks a lot for your suggestions. It has really helped me to move forward and find answers to my confusions and queries. I have some questions for you that i have problems and confusion with:

1. Actually I am trying to evaluate the performance of different resolution DEMs for flood modelling. So I have high resolution DEM (2*2m LiDAR), 10*10m and 25*25m contour interpolated DEMs. With 10*10m DEM the simulation ran without problem and didn't take significantly long time even when i put one month daily averaged discharge and water level data for my BC. But with 2'2m LiDAR DEM, the software didn't respond nicely. While converting my elevation into elevation grid, it didn't respond. I converted it into 5*5m DEM in Arc GIS and ran the simulation for shorter timestep (6-7 hours of a day) but even then it is taking longer time, more than 1.5 hour to build the model. what could be the reason for that?? also the solution to overcome it.

2.Is it enough if i run the simulation just for 6-7 hours period of time of a day and calibrate it. Or more longer time period data need to be used for boundary condition? Does the longer time/shorter time have any effect on the simulation results??

3. In which kind of situation, is it possible or logical to use steady-state condition for building model?? (just my quiriosity)

I look forward to your answers

Thanks

Sudhir

Share this post


Link to post
Share on other sites

Hi,

I will answer your questions below as best that I can.

Hello Rusty!!

Thanks a lot for your suggestions. It has really helped me to move forward and find answers to my confusions and queries. I have some questions for you that i have problems and confusion with:

1. Actually I am trying to evaluate the performance of different resolution DEMs for flood modelling. So I have high resolution DEM (2*2m LiDAR), 10*10m and 25*25m contour interpolated DEMs. With 10*10m DEM the simulation ran without problem and didn't take significantly long time even when i put one month daily averaged discharge and water level data for my BC. But with 2'2m LiDAR DEM, the software didn't respond nicely. While converting my elevation into elevation grid, it didn't respond. I converted it into 5*5m DEM in Arc GIS and ran the simulation for shorter timestep (6-7 hours of a day) but even then it is taking longer time, more than 1.5 hour to build the model. what could be the reason for that?? also the solution to overcome it.

- There is a limit to the size of model you can build in SMS. If you are using the 32-bit version, it is easy to create data that takes up more RAM than you can access using a 32-bit process. The 64-bit version can get a hold of more RAM but eventually things will just go so slow that it isn't practical. Keep in mind that for TUFLOW a 10m grid will have data points every 5m. TUFLOW stores elevations at corners, mid-sides, and cell centers. When you decrease the cell size, you will also need to decrease the timestep. The rule of them is that the timestep in seconds should be the cell size in meters divided by two. Decreasing the cell size can have an enormous impact on runtimes. If you decrease your cell size by half, you increase the number of cells by 4 (twice as much in each direction). After decreasing the timestep, you will increase your runtime by a factor of 8. I'm not entirely sure what you can do to overcome this other than to try looking at a smaller area.

2.Is it enough if i run the simulation just for 6-7 hours period of time of a day and calibrate it. Or more longer time period data need to be used for boundary condition? Does the longer time/shorter time have any effect on the simulation results??

- That is a difficult question to answer because it will likely be site specific. It depends upon mostly upon the size of the model but some on how much the flow changes and how fast of a flow you are interested in. The important thing to know is how long does it take for an effect on one end of the model to affect the other. I would try running for a few hours at a low flow rate and switch to a higher flow over the period of 15 minutes or so. See how long it takes for that effect to reach the downstream boundary.

3. In which kind of situation, is it possible or logical to use steady-state condition for building model?? (just my curiosity)

- I have kind of an extreme position on this that may not be representative of others. I would do steady state for flume experiments where you do hold flows constant for a long time, controlled flow situations where you are interested in specific flowrates (like below a DAM), or if I am told I have to. In every other case I would use transient boundary conditions because these reflect reality and help evaluate the effect of storage. Some US government agencies will require steady state flows for some analysis to purposely remove the effect of storage.

I look forward to your answers

- I hope they were helpful.

Thanks

Sudhir

Share this post


Link to post
Share on other sites

Hello Rusty!!

Many thanks for your answers which which provided me good insight about my queries. Today I spent my whole day trying to calibrate my model for 10m DEM. I have created observation coverage as suggested by you for one location and looked how well the calibrated WL fits with observed WL. I changed Mannings n value and calibrated it several times. I got acceptable results at some points for 5 time steps out of 7. The calibrated WL for 2 time steps is not within acceptable range despite of many try. I couldn't think of any reason why these two timesteps are not matching. Can u give me some suggestions to overcome this.

I have divided my material data into four class: agricultural land, urban area, forest and channel. and most of the area is urban.

Can u give me some hints and tips about using mannings n value more effectively? ome other information/suggestions useful at the stage.

I look forward to your response.

Sudhir

Share this post


Link to post
Share on other sites

Hi Sudhir,

I'm glad things are improving.

There are lots of reasons calibration could be proving difficult. It is odd that you can calibrate well for most time-frames and not for other times. Some ideas to look at:

1. The bathymetry data may not be completely right. DEM data often doesn't include under the water surface and is therefore of little use.

2. The quality of your boundary condition data may be suspect.

3. The quality of your calibration data may be suspect. This depends upon how it is collected. I have seen bad calibration data particularly when dealing with high water marks.

If you are comfortable with the above, the things you can change include roughness and eddy viscosity formulations.

Could you post graphics of your domain (elevation contours), your calibration locations, boundary condition curves, and your resulting plots showing the computed data you are looking at? If they are a hassle to upload to the forum, send them to support@aquaveo.com.

Rusty

Hello Rusty!!

Many thanks for your answers which which provided me good insight about my queries. Today I spent my whole day trying to calibrate my model for 10m DEM. I have created observation coverage as suggested by you for one location and looked how well the calibrated WL fits with observed WL. I changed Mannings n value and calibrated it several times. I got acceptable results at some points for 5 time steps out of 7. The calibrated WL for 2 time steps is not within acceptable range despite of many try. I couldn't think of any reason why these two timesteps are not matching. Can u give me some suggestions to overcome this.

I have divided my material data into four class: agricultural land, urban area, forest and channel. and most of the area is urban.

Can u give me some hints and tips about using mannings n value more effectively? ome other information/suggestions useful at the stage.

I look forward to your response.

Sudhir

Share this post


Link to post
Share on other sites

Hello Rusty!!

Thanks for your answer. I think in my case reliability of data should not the problem because I had really high quality (1m*1m) bathymetric data collected by side scanning sonar for river bathymetry. I converted it into the corresponding resolution (2m, 10m & 25m) of DEMs i have and combined this data with respective DEMs to get river geometry. Also regarding the boundary condition, the upstream discharge was obtained from finnish environment institute which has its observation station few kilometers up from my study area where there is a dam for hydropower plant. And the water level data at 2 calibration point is also from real time Water level observation station at those locations from finnish environment institute. I have discharge and observed water level information (at downstream BC, calibration point 1 and calibration point 2 -please refer to the graphics) on hourly basis for 2010 starting from january 1st to August 9. I think the crucial thing for me is not the data but some guidance about how to process and use these data effeciently. So can you please give some idea on following things:

1. What are the most important things one should look at before using any hydrological data for using it as boundary conditions? May be i haven't been able to look at the data critically.

2. Regarding my downstream BC, I have created BC arc at two locations as the river is branched. But i have water level data for one side only. I have been assuming the same water level at other side also. Can it be done or I have done mistake here?? And also, should the boundary arc should be drawn exactly at the same point of water level measurement or some up or down doesn't matter.

3. Out of the hydgolocial data, what is the best way to separate it into boundary condition data for building model and validation data?

4. The way i have created my river geometry is i have converted my DEMs and Sonar data into different resolutions and converted those into ESRI shapefile. In SMS i did mapping->shapes to feature objects and then feature objects->map to scatter. Then right click on default coverage and choose type->models->tuflow->grid extents. This is how i have creatd TUFLOW grid. Is this the correct way? or there is other smarter and efficient way to do this.

I have attached the related files and graphics in your mail as it said error here (rjones@aquaveo.com. Plz find the attachment

I would be really happy and grateful to have your critical comments on these confusion as you are the only guide/teacher for me at the moment. I will appreciate your help from the bottom of my heart.

Thanks

Sudhir

Share this post


Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

Loading...

×
×
  • Create New...