|
Hazlett-Kincaid,
Inc. Hydrogeology Today™ |
||
|
Optimization Send Your Letters New Talent at Hazlett-KincaidAn internet-based electronic
publication of Hazlett-Kincaid. Copyright © 2001. Hazlett-Kincaid, Inc.
All rights reserved. Contact Us: ph: 717.859.1413 Hazlett-Kincaid, Inc. Comments or Questions? |
There is a lot of talk about optimization these days. Much of the buzz is over the fact that the groundwater and fate and transport modeling processes can be greatly enhanced by optimization, with the end results often orders of magnitude better than using hand calibration. An Introduction to OptimizationFor the un-annointed among you, let's start with some
plain-English definitions. Now, groundwater modelers traditionally calibrate their models to either heads, pressures, concentrations, or temperatures. In reality, calibration can be performed with respect to any measured spatial variable. Calibration is a trial-and-error process where parameters are tweaked until model variables match measured values (acceptable error is less than 10% of measured values). The problem with this approach is that the "answer" one achieves is not necessarily the best result nor is it unique; numerous combinations of parameter values are likely to produce the same desired result. Therefore, even though models may calibrate well to current data, over a generation, modelers have learned that many models are not valid in the long term. I have previously written about the ever-increasing computing power on the desktop available to the average user. Optimization is part of an emerging trend taking advantage of this easy-access power, where it is used to automatically (auto-magically!) calibrate your model. This is accomplished through brute-force computation. By brute-force, I mean that the computer(s) is tasked with finding and trying all the possible parameter combinations that will produce a calibrated model result. To imagine how this occurs, think about a ball rolling around a smooth surface with lots of relief. If you released the ball at different points on the surface, it would roll around and eventually come to rest. The computer effectively releases the ball from every possible point. The places where the ball comes to rest are called minima (just a fancy term for low points). These minima represent calibrated solutions to the problem. However, the keen observer will note that, as I mentioned above, there are many minima on the surface and therefore more than one "best" answer to the problem. Optimization codes systematically seek out the lowest point on the surface. Some optimization codes, such as PEST (Parameter Estimation) tend to get stuck in local low points. As such, PEST is referred to as a local optimization routine and it doesn't always produce the best results (though it is likely better than most hand calibrations). Other routines seek the absolute lowest point on the surface; the so-called global minima. This approach will always yield the absolute measurably-best calibration of the model, often orders of magnitude better than hand calibration. How does this benefit you? Well, for starters, optimization takes much of the guesswork (art) out of model making. The trial-and-error hand calibration is performed automatically, optimally, and much more rapidly by a computer. For large models with many parameters and variables, this can lead to a huge time savings. As we know, time is money. Second, and probably most importantly, because optimization results are mathematically measurable, they endure close scrutiny much better than hand calibrations, wherein the experience and expertise of the modeler is the overriding factor for the quality of the model, rather than mathematical certainty. For a more in-depth look at the science and applications
of optimization, I suggest the following books, which you can purchase
through our Amazon.com Associates program: Welcome our new additions!We are proud to announce the addition of two new employees at our main office in Akron, PA: Luke Blair and Bob Libutti. Luke comes to us from the USGS in Reston, VA, where he was a GIS analyst for the GIS lab of the Eastern Earth Surface Process Team as well as their EarthVision (EV) modeling expert. Luke has an M.S. in Geology from Vanderbilt University and a B.A. from the College of Wooster. Though educated as a geologist, he is also a skilled cartographer and geographer who possesses the combination of computer savvy and practical experience we value most. He'll be putting those skills to good use, beginning work immediately on several of our projects. Bob will be a senior majoring in geology at Lafayette University (Easton, PA) in the Fall. Bob has hit the ground running since joining us several weeks ago, having worked extensively on one of our public outreach projects, the upcoming museum exhibit at the Reading Public Museum, and on revamping many of our digital promotional and web materials. We are looking forward to both Luke and Bob contributing greatly to the success of HKI with their valuable skills and we welcome them aboard!§(back to top) Send Your LettersI am opening our newsletter to contributors. If you have written interesting articles somehow related to groundwater and would like to have the article reach our numerous subscribers, email it to me, making sure it is in HTML format if any images are included. If I like what I read, I'll put it in one of our newsletters with links back to your company web site and an email link to you. I reserve editorial over site of all submitted materials. Please limit the submissions to no more than 2 images and 200 words or less.§(back to top) |
|
