News: Forecasting of Hurricane Isaac

2012/08/29 – Computerworld
Supercomputers help New Orleans prepare for Hurricane Isaac
Computing advances since Katrina have helped the city plan better on the storm surge, for one

Computer-World

About the time of Katrina, the computer models “were much coarser and had minimum resolutions of only 100-200 meters,” said Casey Dietrich, a post-doctoral researcher at the Institute for Computational Engineering and Sciences at University of Texas in Austin.

Dietrich has been running compute models at the Texas Advanced Computing Center at the University of Texas to assess the impact of the storm surge on Texas.

Emergency planners in both states take the data generated by the university researchers and incorporate it into geographic information systems.

“They can look down at neighborhood scale and say ‘on this street along the levy we’re going to have water this high,’ and plan accordingly,” Dietrich said.

Comparing the capability today with that at the time of Katrina, Dietrich said: “I think we have a very strong understanding of how hurricane wave storm develop and how they can threaten a coastal environment.”

Also see local coverage by the Institute for Computational Engineering and Sciences.

Performance of the Unstructured-Mesh, SWAN+ADCIRC Model in Computing Hurricane Waves and Surge

JSC2012Coupling wave and circulation models is vital in order to define shelf, nearshore and inland hydrodynamics during a hurricane. The intricacies of the inland floodplain domain, level of required mesh resolution and physics make these complex computations very cycle-intensive. Nonetheless, fast wall-clock times are important, especially when forecasting an incoming hurricane.

We examine the performance of the unstructured-mesh, SWAN+ADCIRC wave and circulation model applied to a high-resolution, 5M-vertex, finite-element SL16 mesh of the Gulf of Mexico and Louisiana. This multi-process, multi-scale modeling system has been integrated by utilizing inter-model communication that is intra-core. The modeling system is validated through hindcasts of Hurricanes Katrina and Rita (2005), Gustav and Ike (2008) and comprehensive comparisons to wave and water level measurements throughout the region. The performance is tested on a variety of platforms, via the examination of output file requirements and management, and the establishment of wall-clock times and scalability using up to 9,216 cores. Hindcasts of waves and storm surge can be computed efficiently, by solving for as many as 2.3E12 unknowns per day of simulation, in as little as 10 minutes of wall-clock time.

JC Dietrich, S Tanaka, JJ Westerink, CN Dawson, RA Luettich Jr, M Zijlema, LH Holthuijsen, JM Smith, LG Westerink, HJ Westerink (2012). “Performance of the Unstructured-Mesh, SWAN+ADCIRC Model in Computing Hurricane Waves and Surge.” Journal of Scientific Computing, 52(2), 468-497, DOI:10.1007/s10915-011-9555-6.

Example Input Files for SWAN+ADCIRC

Updated 2023/02/24: Corrected the fort.26 file for SWAN v41.31 to remove erroneous WCAP input.
Updated 2021/12/13: Adding example files for SWAN v41.31 with ST6 physics.
Updated 2016/09/18: Adding example files for use with older versions of SWAN.
Updated 2016/07/31: Minor changes to the SWAN control file to be compatible with SWAN v41.10.

Many new users of SWAN+ADCIRC, after reading the instructions on how to compile and run the coupled models, have asked for example input files to test their implementation. This page provides an example application of Hurricane Gustav (2008), using files that were available already on the Internet for other applications.

The entire set of example input files is: Gustav-EC95d-v41.31.zip.

The individual files are also linked and described below. They have been modified to be compatible with the latest versions of the codes, and to reflect our latest settings for SWAN and ADCIRC. These files are extremely coarse, both in the mesh resolution and the wind field, but they are a good starting point for new users of the coupled models. For more information about our high-resolution hindcast of Gustav, please see our article in Monthly Weather Review.

Continue reading →

FigureGen v.49

Updated 2013/09/30: Added a link to a published application of FigureGen.

Updated 2012/08/27: Real-time coupling now controlled by the -DSLOWREAD flag.

FigureGen is a Fortran program that creates images for ADCIRC files. It reads mesh files (fort.14, etc.), nodal attributes files (fort.13, etc.) and output files (fort.63, fort.64, maxele.63, etc.). It plots contours, contour lines, and vectors. Using FigureGen, you can go directly from the ADCIRC input and output files to a presentation-quality figure, for one or multiple time snaps.

This program started from a script written by Brian Blanton, and I converted it to Fortran because I am more familiar with that language. It now contains code written by John Atkinson, Zach Cobell, Howard Lander, Chris Szpilka, Matthieu Vitse, and others. But, at its core, FigureGen behaves like a script, and it uses system calls to tell other software how to generate the figure(s).

This example depicts hypothetical oil transport in the northern Gulf of Mexico. The oil spill is represented by Lagrangian particles and initialized with the observed conditions of 29 June 2010, but then the wind forcing of Hurricane Ike (2008) is applied. Oil is pushed into the marshes along the entire coastline of southern Louisiana.

Cover_Particles_Ike_Wind_NGOM

Continue reading →

Surface Trajectories of Oil Transport along the Northern Coastline of the Gulf of Mexico

CSR2012After the destruction of the Deepwater Horizon drilling platform during the spring of 2010, the northern Gulf of Mexico was threatened by an oil spill from the Macondo well. Emergency responders were concerned about oil transport in the nearshore, where it threatened immediately the fishing waters and coastline from Louisiana to Florida. In this region, oil movement was influenced by a continental shelf with varying width, the protruding Mississippi River delta, the marshes and bayou of southern Louisiana, and the shallow sounds and barrier islands that protect the coastline. Transport forecasts require physics-based computational models and high-resolution meshes that represent the circulation in deep water, on the continental shelf, and within the complex nearshore environment.

This work applies the coupled SWAN+ADCIRC model on a high-resolution computational mesh to simulate the current velocity field on the continental shelf, nearshore and marsh areas during the time that oil was visible on the surface of the Gulf. The SWAN+ADCIRC simulations account for the influence of tides, riverine discharge, winds and wind-driven waves. A highly-efficient Lagrangian particle transport model is employed to simulate the surface trajectories of the oil. The transport model accounts for dispersion and advection by wind and currents. Transport is evaluated using two week-long sequences of satellite images. During both periods, the SWAN+ADCIRC current fields alone appeared to be more successful moving the oil than when direct wind forcing was included. In addition, hypothetical oil transport is considered during two hurricane scenarios. Had a hurricane significantly impacted the areas, depending on its track, oil would have moved farther into the marshes of southern Louisiana or farther along the shelf toward Texas than actually occurred during the spill.

JC Dietrich, CJ Trahan, MT Howard, JG Fleming, RJ Weaver, S Tanaka, L Yu, RA Luettich Jr, CN Dawson, JJ Westerink, G Wells, A Lu, K Vega, A Kubach, KM Dresback, RL Kolar, C Kaiser, RR Twilley (2012). “Surface Trajectories of Oil Transport along the Northern Coastline of the Gulf of Mexico.” Continental Shelf Research, 41(1), 17-47, DOI:10.1016/j.csr.2012.03.015.

Controlling Errors with Limiters on Spectral Propagation Velocities

Updated 2016/07/31: This post is now outdated. SWAN has been updated to improve its treatment of the spectral propagation velocities, so these limiters are not needed. Please see this post.

Updated 2012/12/26: Added link to published manuscript.

Updated 2012/11/19: Changes to reflect our accepted submission to Ocean Modelling.

As we have gained experience with the coupling of SWAN and ADCIRC, we have noticed that SWAN can focus wave energy due to excessive refraction in regions with coarse mesh resolution. Wave energy can become focused unrealistically at a single mesh vertex, causing the wave properties to become non-physical. In deep water, the significant wave heights can become 150m or larger. In shallow water, the peak wave periods can become 30s or larger, as the energy is pushed into the lowest-discretized frequency bin.

We have developed a few work-around solutions to this problem (Part 1 and Part 2). These solutions have enabled the wave refraction process in the region of interest, and disabled it elsewhere in the computational domain. For example, by enabling selectively the refraction in the northern Gulf of Mexico, we can obtain the following hindcast of the significant wave heights during Hurricane Gustav (2008).

Maximum significant wave heights (m) during Gustav (2008).

Maximum significant wave heights (m) during Gustav (2008).

However, a more robust solution would be the limiting of the spectral propagation velocities, especially the directional turning rate, based on the Courant-Friedrichs-Lewy (CFL) condition. We have implemented recently these limiters in SWAN+ADCIRC. On this page, the limiters are introduced and tested on idealized and realistic applications.

It should be noted that these limiters are not a replacement for increased mesh resolution. The SWAN solution will always be better when the mesh is improved to represent the bathymetric gradients in the region of interest. However, when it is not feasible to increase the mesh resolution, then these limiters can control the largest SWAN errors without affecting the solution elsewhere.

Continue reading →