University of Khartoum

Prediction of Environmental Changes and Chemical Impacts Based on Artificial Neural Networks and GIS

Prediction of Environmental Changes and Chemical Impacts Based on Artificial Neural Networks and GIS

Show full item record

Title: Prediction of Environmental Changes and Chemical Impacts Based on Artificial Neural Networks and GIS
Author: Babikir M. Eljunaid, Ammar
Abstract: The objective of this study is to evaluate the potentiality of using Artificial Neural Networks (ANNs) for predicting the environmental changes in Greater Khartoum area based on remote sensing and Geographic Information System (GIS) techniques. This includes the prediction of the chemical impacts on Khartoum North Industrial Area. One of the objectives of this research work aims to use intelligent and expert systems to model the real world for the purpose of detecting such changes and impacts without involving in sophisticated Digital Image Processing (DIP) and GIS processes. The primary data used in this study is a set of visible and infrared spectral bands acquired by the American Landsat Imaging System in 1996 and 2000. A vector map (produced in 2005) was used as an ancillary data to extract the chemical impacts on Khartoum North Industrial Area. To achieve the research objectives, learning patterns indicating the environmental parameters in the study area as well as Khartoum North Industrial Area were generated based on digital image processing and GIS techniques. The produced patterns were used to train a General Regression Neural Network (GRNN) model for predicting future environmental changes in Khartoum North Industrial Area. Further training session was carried out to predict the chemical impacts on the study area. A wide range of procedures were carried out to gain the knowledge base upon which the shell can teach itself. Again digital image processing and GIS techniques were used to assess the general performance of the shell. The General Regression Neural Network model, as tested, was found to be able to recognize and predict environmental changes. The model was trained for approximately 39 minutes (00:38:58) with 3000 training patterns. The best smoothing factor achieved during the training session was found to be 0.744141 with a minimum mean squared error of ± 0.026936. When the trained shell was applied to 6000 new production (test) patterns, the model was able to process the whole set of patterns with reasonable results. The highest percentage of the error in all sets was found to be 0.054% (out of the total classified area) while the average was 0.031%. In all sets of production patterns the number of pixels of the predicted image was found to be equal to the corresponding value of the original image. This indicates that the basic geometrical characteristics for iv both images are the same. Thus the shell processing does not affect the geometry of the predicted image. Further tests show that the GRNN model was able to predict chemical impacts on Khartoum North Industrial Area with reasonable error. The whole set of test patterns (127 patterns) has been processed. When the predicted images (test patterns) compared with the actual images that produced by DIP techniques, 89.52% of the total area was correctly predicted. This is equivalent to 9.0 Km2 of the total area (10.156 Km2) covered by the test patterns. Although chemical impacts due thermal effects were originally detected based on digital image processing and GIS techniques, very sophisticated procedures have been applied to achieve this objective. Therefore, artificial neural networks, as tested, provide a real solution for end users without being involved in such sophisticated processes.
Description: 138page
URI: http://khartoumspace.uofk.edu/handle/123456789/9301
Date: 2015-04-23


Files in this item

Files Size Format View

This item appears in the following Collection(s)

Show full item record

Share

Search DSpace


Browse

My Account