FMA Challenge 2: Difference between revisions
Content deleted Content added
| (33 intermediate revisions by 3 users not shown) | |||
Line 1:
= Introduction =
The FMA Challenge
This Wiki page assumes an intermediate to advanced user level. If you haven't already completed FMA Challenge 1, please see
This model specifically shows:
*Tidal influences on floodplains and riverine systems;
*Variation in results due to 2D cell resolution and whether or not the model should contain a 1D component;
Line 12 ⟶ 13:
*The differences between two different 2D solvers, TUFLOW classic and TUFLOW GPU.<br>
Data for this model is provided
*[https://www.tuflow.com/Download/TUFLOW/Demo_Models/FMA_Challenge_Model_2_QGIS.zip QGIS Data Download]
*[https://www.tuflow.com/Download/TUFLOW/Demo_Models/FMA_Challenge_Model_2_MapInfo.zip MapInfo Data Download]
*[https://www.tuflow.com/Download/TUFLOW/Demo_Models/FMA_Challenge_Model_2_ArcGIS.zip ArcGIS Data Download]
=Relevant Tutorials=
It may be useful to revisit some of the following tutorials:<br>
*General 2D modelling - <u>[[
*2D topography modification - <u>[[
=Model Setup=
This section provides an overview and discussion of the model domain setup.
It is at your discretion which GIS package, text editor and method of model simulation to use (batch mode or within the text editor). All files required to setup and run the models are available within the download package. You have the choice of running with shape file or mif for usage in ArcGIS/QGIS or Mapinfo respectively.<br> <br>
This model does not require a licence dongle to run.
==Grid Setup and Cell Size==
Line 37 ⟶ 41:
As with FMA Challange 1, variable model cell sizes have been simulated using the powerful event and scenario functionality of TUFLOW.
The grids investigated
{| align="center" class="wikitable" width="50%"
Line 49 ⟶ 53:
| 30m (grid) || 80,842
|}
=== Note on Runtimes ===
There is a series of batch files provided with the demo model that give the choice of running only the 15m and 30m grid resolutions or all three. Due to the size of the model and large number of cells for the CPU only 10m case, it may take up to 24hrs of 'real time' to run on your computer. If you don't want to run the 10 m case then just run either 'run_all_001.bat' or '_run_all_CPU_001.bat'. Further details are provided in the Readme.txt within the runs directory.
Current benchmarching of runtimes for the CPU vs GPU respectively has found that the GPU simulations are typically 20-30 times faster than the CPU for this demo model. For models with a much larger number of cells >10 million the benefit of the GPU increases dramatically with CPU/GPU runtime speed increases over 100. Further detail on the hardware benchmarking conducted on FMA Demo Model 2 is provided [[Hardware_Benchmarking | here]].
==Simple Manning’s n / Topography Test==
Line 55 ⟶ 65:
The topography for these runs did not include any special provision for incorporating the levee and road embankments. The elevation sampling of the DEM was simply based on inspecting the DEM at the models’ elevation points. At this stage no 3D breaklines along the embankment crests were used to ensure the crest of the elevation was correctly represented by the models’ elevations.
▲Additional simulations were carried out using TUFLOW that included the NLCD Land Cover GIS layer, and 3D breaklines of the levee and road embankments visible in the DEM and aerial photo. This scenario (we envisage) would probably be representative of the present day conditions, and not representative of the 1964 flood event conditions.
The NLCD GIS layer was read directly by TUFLOW and Manning’s n values assigned to the different land covers as tabulated below. The water/river n value of 0.02 is considered at the lower end of typical n values for rivers based on numerous 2D model calibrations. Typically for the lower reaches of coastal rivers the in-bank n value is 0.022 to 0.025 with an acceptable range considered to be (0.02 to 0.03). In upper reaches, away from tidal influences, n values tend to be higher (0.025 to 0.04) for perennial rivers. Possible reasons for the low n value could be due to the river bathymetry accreting since 1964 (we assume the bathymetry provided is based on more recent surveys), or the inflows maybe high.
Line 105 ⟶ 113:
The NLCD land cover provided a poor representation of the in-bank river section when compared with the DEM and aerial photography. The river section was therefore digitized to replace the NLCD representation in the river. A general comment overall was that the NLCD land cover was not always in agreement with the aerial photo and somewhat poor in its accuracy and resolution.
For the levee and road crests, GIS breaklines were digitized along the crests, and elevations sampled from the DEM (refer ''2d_zsh_ridges_001''). Ideally, ground surveys of these hydraulic controls would be available to ensure the crest is accurately modeled. TUFLOW reads the GIS breaklines and interpolates the crest elevation to the nearest 2D cell elevations to ensure the embankment height is correctly modeled.
In general the comparison with the high water marks tends to be higher overall, but the distribution of flow has improved with less water travelling to the southern tidal entrance. In a few areas it is worse such as in the images below. In the images the red values are the 1964 HWMs, and the black the calculated peak levels. In these areas the presence of the embankments has a strong influence on the calibration results. To test this influence a further scenario removing two critical embankments was carried out as discussed further below.
Line 134 ⟶ 142:
==Challenges==
No unexpected “challenges” for Challenge 2, but it is worth noting the challenges that arise in calibration exercises. Sometimes there is an expectation that models should be able to reproduce recorded levels to within a very small tolerance. In these situations there needs to be a good understanding/appreciation by both the client and the
===Are Manning’s n Values the Same for 1D and 2D models?===
Line 143 ⟶ 151:
<ol>
<li>Recorded levels can themselves be quite uncertain. They should preferably be rated as to their accuracy and the type of flood mark noted with the recording. For example, a water mark in the wall of a house is a reliable, accurate mark of the peak water level; while a debris level is simply an indication that the flood was at least this high (debris marks may not be at the peak of the flood). Scanning through the high water marks in this study would indicate some inconsistencies. For example, in the northern end of the study area the 5.35 HWM in the image below is upstream of the 5.66 HWM. Either the 5.35 is low and/or the 5.66 is high. If the accuracy of the HWMs can’t be established, then as a generalization, the modeler should err on the high side in case the flood mark was not recorded at the flood peak.<br>
[[File:Critical_Embankment5.jpg|600px]]
<li>The
<li>Levees, roads, railways and other embankments can have a major influence on the flood behavior. Those embankments that overtop should be ground surveyed along their crests and the crest height correctly represented in the model’s elevations. Often for calibration events the height or presence of the embankments is not known accurately and this needs to be taken into account in the ability of a model to reproduce flood marks.
Line 158 ⟶ 168:
==Caveats==
In summary, the TUFLOW 15m No Embankments simulation provided the best calibration
=Conclusion=
In this challenge, we explored a tidally influenced floodplain and riverine system. From this, we gained a better understanding of how cell resolution affects results and when it is suitable to neglect 1D components, how to use a looped batch file to run through scenarios, and an understanding of comparison between the
Congratulations on finishing Challenge 2! Please click [[Main_Page | here]] to return to the main page.
| |||