3DVAR Tutorial

Welcome to the 3DVAR tutorial page!

1. Introduction

Data assimilation is the technique by which observations are combined with an NWP product (the first guess or background forecast/analysis) and their respective error statistics to provide an improved estimate (the analysis) of the atmospheric (or oceanic, Jovian, whatever) state. Variational (VAR) data assimilation achieves this through the iterative minimization of a prescribed cost (or penalty) function. Differences between the analysis and observations/first guess are penalized (damped) according to their perceived error. The difference between three-dimensional (3DVAR) and four-dimensional (4DVAR) data assimilation is the use of a numerical forecast model in the latter.

This 3DVAR tutorial is recommended for people who are

• Familiar with the following components of the MM5 modeling system: TERRAIN, REGRID, INTERPF, MM5 and GRAPH. If you wish to review what these programs do then visit the documentation on the MM5 web-page;
• Potential users of 3DVAR who want to learn how to run 3DVAR by themselves;
• New users who plan on coming to the NCAR 3DVAR tutorial - for you we recommend that you try this tutorial before you come to NCAR whether you are able or unable to register for practice sessions, and this will hopefully help you to understand the lectures a lot better;
• Users who are looking for references to diagnostics , namelist.3dvar , etc - look for 'Miscellanies' and 'Trouble Shooting' sections on each page.

If you are a new 3DVAR user, this tutorial is designed to take you through 3DVAR-related programs step by step. If you have used 3DVAR before, you may find useful information here too. If you don't know anything about 3DVAR, you should first read the 3DVAR Overview, and, for more detail, the 3DVAR Technical Note.  Also view the 3DVAR tutorial presentations: Online 3DVAR_OBSPROC Presentations and Online MM5 3D-Var Presentation taken from the tutorial held at NCAR. The 3DVAR web-page contains links to many more sources of information.

2. Goals Of This 3DVAR Tutorial

In this 3DVAR tutorial, you will learn to run the various 3DVAR-related programs. We will supply you with sample observation files and MM5 format first guesses, and a low-resolution global background error statistics (GBES) file. In your own work, you will need to create these yourselves via a decoder (to convert your observations into the necessary LITTLE_R format) and the MM5 programs TERRAIN, REGRID, INTERPF and MM5. The 3DVAR system uses the output of either INTERPF (in “cold-start” mode”) or MM5 (in “warm-start” mode) as its first guess. More important thing is that you should find the tuning coefficients (listed in namelist.3dvar) suitable for your applications by using GBES file provided here, or If possible, a new background error statistics file should be created off-line for your applications.

In this tutorial, you will only be learning how to run 3DVAR with MM5. To learn how to use the same 3DVAR system with the WRF model (using BUFR observations and a WRF format first guess), please see WRF 3DVAR. A tutorial to help you know WRF 3DVAR system…..

Here, you will learn how to run 3DVAR’s observation preprocessor as well as 3DVAR itself for two cases (single observation and in an operational scenario). We suggest that you start by running through the 3DVAR-related programs at least once using these cases. This serves two purposes: First you can learn how to run the programs, and second you can test whether your computer is adequate to run the entire modeling system. After you have done this tutorial, you can try

• Running some of the other, more computationally intensive, case studies (not yet available).
• Experimenting with some of the many namelist variables (WARNING: It is impossible to test every code upgrade with every permutation of computer, compiler, number of processors, case, namelist option, etc so please modify only those you understand!). Good examples include withholding observations and tuning background error coefficients.
• Running with your own domain. Hopefully, our test cases will have prepared you (and us!) for the variety of ways in which you may wish to run 3DVAR.

As we are going through the tutorial, you will download program tar files and data to your local computer, compile and run on it. Do you know what machine you are going to use to run 3DVAR-related programs? What compilers do you have on the machine?

Running 3DVAR requires a Fortran 90 compiler. We currently support the following platforms: IBM,Compaq (DEC Alpha), and PC/Linux (with Portland Group compiler). Please let us know if this does not meet your requirements, and we will attempt to add other machines to our list of supported architectures as resources allow. Although we're interested to hear of your experiences modifying compile options, we do not yet recommend making changes to the configure file used to compile 3DVAR.

3. Tutorial Schedule

We recommend you follow the tutorial in the order of the sections listed below. As mentioned above, this tutorial assumes you know how to run the MM5 programs TERRAIN, REGRID, INTERPF and GRAPH.

a) Download Test Data – In the 3DVAR tutorial, you will be running the 3DVAR observation preprocessor (3DVAR_OBSPROC) as well as 3DVAR itself. This page describes how to access test datasets to run both 3DVAR_OBSPROC and 3DVAR.

b) The 3DVAR Observation Preprocessor (3DVAR_OBSPROC) – In this part of the 3DVAR tutorial you will download, compile and run the 3DVAR observation preprocessor (3DVAR_OBSPROC). The purpose of 3DVAR_OBSPROC is to perform gross quality control and to reformat from LITTLE_R format to the format used by 3DVAR (currently a form of ASCII but ultimately BUFR).

c) Setting up 3DVAR – In this part of the 3DVAR tutorial you will download and compile 3DVAR.

d) 3DVAR – Run Pseudo Single Observation Test (PSOT) – In this section, you will learn how to run 3DVAR using only a single “pseudo” (or “bogus”) observation.

e) 3DVAR – Run Case Study – In this section, you will learn how to run 3DVAR using observations and a first guess from an actual operational system.

f) 3DVAR – Diagnostics – 3DVAR produces a number of diagnostics file that contain useful information on how the assimilation has performed. This section will introduce you to some of these files, and what to look for.

g) Updating lateral boundary conditions – Before running a forecast, you must first modify the tendencies within the lateral boundary condition files to be consistent with the new 3DVAR initial conditions. This section shows you how.

h) Production of the background error statistics via the NMC-method (by clicking "NMC-Method Source Code").– The “NMC-method” calculates a climatological estimate (be) of background error for your domain. This is an off-line calculation, done if you are confident enough to calculate your own background error statistics, have a set of MM5 forecasts to difference for your domain, and are willing to work to tune and improve your results!

OK, if you’re still not quite sure what has happened at this point, perhaps the presentations from the NCAR tutorial will help to answer your questions – see the latest at June 2004 3DVAR Tutorial Presentations (to come). Alternatively, look to the 3DVAR Frequently Asked Questions (to come) for guidance.

Once you are able to run all these programs successfully, and have spent some time looking at the variety of diagnostics output that is produced, we hope that you'll have some confidence in handling the 3DVAR system programs when you start your cases. Good luck!

---

For comments, send email to mesouser@ucar.edu

Last Modified: May 2004