- Peter Anderson
- Kurt Fedra
Urban Environmental Management: towards the City of Tomorrow
- Vladimir Krasnopolsky
Neural Networks for Environmental Scientists and Engineers
Genetic AlgorithmsPeter G. Anderson
Professor, Computer Science Dept
Rochester Institute of Technology
Rochester, New York 14623-5608
A genetic algorithm (GA) is an indirect method for rapidly searching for good solutions for hard problems. The problems that GAs are particularly suited for are those that have no straight-forward algorithmic solution-generators but do have methods for evaluating how good a proposed solution is. Such problems are typified by scheduling and lens design.
GAs are patterned on nature's evolution ("survival of the fittest") or selective breeding. A GA maintains a large population of trial solutions to the problem, selects some with higher fitness, and recombines their components to form new solutions. Over time, the population contains more and more solutions with higher and higher fitness.
This is a short, elementary course to introduce GAs to participants with a computer programming background. GA programs, tools, and applications will be provided.
Urban Environmental Management: towards the City of TomorrowDr. Kurt Fedra
Environmental Software & Services GmbH
Kalkgewerk 1, PO Box 100, A-2352 Gumpoldskirchen
Tel:+43 2252 633 05
The session will bring together results from a number of European projects dealing with the urban environment, and in particular issues of air quality and transportation.
Emphasis will be on the use of models, GIS and expert systems, for analysis and decision support, as well as the Internet as a central communication and dissemination structure.
Neural Networks for Environmental Scientists and EngineersDr. Vladimir Krasnopolsky
Environmental Modeling Center NWS/NCEP/NOAA (SAIC)
5200 Auth Rd.
Camp Spring, MD 20746
tel. 301-763-8000 ext. 7262
Home page: polar.wwb.noaa.gov/omb/people/kvladimir
This tutorial targets experts in environmental sciences and engineering (satellite remote sensing, satellite data analysis and applications, meteorology, oceanography, hydrology, environmental numerical modeling, etc.) who are interested in an introduction to neural network techniques and in getting a jump start in using neural network techniques in their fields. This tutorial will also be beneficial for experts in numerical modeling and in assimilating data in numerical models.
The tutorial contains a brief introduction into neural networks (multiplayer perceptrons). This introduction is beneficial for those without any previous knowledge of neural networks. For other participants it will serve as a brief and useful reminder that organizes knowledge into a system and introduces a unified terminology. The main goal of this introduction is to demonstrate that neural networks are a generic and efficient tool for modeling continuous mappings.
The second part of the tutorial presents a broad class of neural network remote sensing applications - solutions of forward and inverse problems in remote sensing. The remote sensing forward and inverse problems are introduced. It is shown that both forward and inverse problems can be considered as continuous mappings and therefore, neural networks can be used to solve these problems. Two practical neural network applications (implemented at NCEP/NOAA) illustrate this part of the tutorial. The first is a neural network fast forward model for direct assimilation of microwave satellite data into numerical weather prediction models. The second application is a multi-parameter retrieval algorithm, which retrieves multiple atmospheric and oceanic parameters from microwave satellite data.
The third part of the tutorial presents another broad class of neural network applications - fast parameterization of subgrid processes in numerical models. The calculation of subgrid processes (effects with the scales smaller than an integration step of the numerical model) usually takes significant time in numerical models. Parameterizations of such processes can be considered as continuous mappings, and therefore, fast neural network parameterizations can be developed which help to significantly improve the accuracy and performance of the models. Three practical neural network applications illustrate this part of tutorial. The first one is a fast neural network parameterization of the equation of state of seawater. This neural network parameterization is currently used in the NCEP ocean circulation model and data assimilation system. The second application is a fast neural network parameterization of nonlinear wave-wave interaction, which is currently tested in the global ocean wave model at NCEP. The third application is a fast neural network parameterization of the radiation physics in the ECMWF numerical weather forecast model.
In the fourth and last part of the tutorial an outline of a generic neural network approach is presented. This outline is designed to facilitate the use of neural network techniques in applications by people with little or no previous experience in developing neural network applications. Important practical questions and tips about selecting neural network architecture, building training, test and validation data sets, estimating required quantity and quality of the data, training strategies, etc. are discussed.
Table of Contents
- Briefly about Environmental Sciences
- Brief Review of NN Applications in Environmental Sciences.
- NNs as a Generic Tool for Modeling Continuous Mappings
- NNs in Remote Sensing
- Forward and Inverse Problems in Remote Sensing
- NN Fast Forward Models and NN Multi-Parameter Retrieval Algorithms
- Example 1: NN Fast Forward Model for Direct Assimilation of Satellite Microwave Data in Numerical Models.
- Example 2: NN Multi-Parameter Retrieval Algorithm to Derive Atmospheric and Oceanic Parameters (surface wind speed, moisture concentrations, and sea surface temperature) from Satellite Microwave Data.
- NNs in Numerical Modeling
- Using NNs for Fast Forward and Inverse Parameterization of Subgrid Processes in Numerical Models
- Example 3: NN Forward Parameterization - Equation for Seawater Density in Oceanic Global Circulation Model
- Example 4: NN Inverse Parameterization - Equation for Seawater Salinity in Oceanic Data Assimilation System
- Example 5: NN Indirect Parameterization - Nonlinear Energy Transfer in Wind Wave Models
- Example 6: Fast NN Radiation Parameterization for Numerical Weather Prediction Models
- An Outline of a Generic Approach
- Problem Analysis
- Data Analysis
- Do You Really Need NN for Solving This Problem?
- NN Training
- NN Validation