Please use this identifier to cite or link to this item: http://repository.aaup.edu/jspui/handle/123456789/2899
Title: Predicting the effects of weather conditions on agriculture by applying spatial data mining methods رسالة ماجستير
Authors: Eleyat, Mohammed Omer Qasem$AAUP$Palestinian
Keywords: gis,neural networks,data mining
Issue Date: 2018
Publisher: AAUP
Abstract: Weather conditions including precipitation, temperature, wind speed, solar radiation, and humidity affect plants in different ways that can make them more susceptible to disease and insect problems. Unlike the short term weather, the climate represents the average weather conditions over a long period of time, which determines what will probably grow well in a certain region. However, extreme weather conditions can kill plants and damage whole farms, which may result in huge agricultural loss. Due to the importance of the weather forecast to the scientific and technological issues and challenges of the century problems, the researcher applied spatial data mining methods to the collected weather data in order to forecast weather conditions and alert farmers to take precautions and avoid agricultural damage. Specifically, with reference to the Palestinian Meteorological Authority and Ministry of Agriculture, the collected data included statistics about precipitation, temperature, wind speed, solar radiation, humidity and the percentage of the affected crops. These data were digitized, prepared, cleaned and normalized to make ensure that the analysis and results are correct. Then, they were converted into GIS formats and ArcGIS. After that, the spatial data of the research were saved, visualized, joined and analyzed using GIS software. The data were also interpolated i.e. data were extended so that they cover all the points that are around 1000 meters apart from each other. Several algorithms, such as ordinary least square and multi perceptron neural network of data mining, were implemented to predict the percentage of the affected crops before and after interpolation. Some of these algorithms were applied to ArcGIS software and some of them to the WEKA software; some codes were programmed to convert data, sort them and apply the new spatial data mining implementations. vi Consequently, a new approach is suggested for spatial prediction by using map algebra and dealing with the maps area as a matrix. This spatial approach was applied taking into consideration the surrounding four neighbors for each location point to ensure the inclusion of the effect of these surrounding areas, fields, properties and effects. All the methods were applied to the training data and tested by cross validation. The adjusted residuals were also calculated and compared. In conclusion, The results of each implementation were examined. The best results were achieved using the neural network method with 0.0718 mean absolute error ,0.1664 root mean squared error and 0.3714% relative absolute error. Residuals of testing data and cross validation were minimized and compared. Satisfying results were achieved and presented. A clear improvement was achieved to here, The mean absolute error is reduced from 1.3837 to 0.0718, by the suggested spatial model in three levels when spatial neighbor's variables were considered. Matrix map algebra was done and furthermore the target values themselves of the neighbor areas were considered using iterations. The results of the iterations have been tested by calculating the maximum difference between the current and previous iteration. Consequently an impressive improvement was achieved as the K maximum error is reduced from 14.377 to 1.251, for the number of iterations. This, in turn, encourages more research in this field and suggests future work that includes advanced analysis and modeling.
Description: Master`s degree in Computer Science
URI: http://repository.aaup.edu/jspui/handle/123456789/2899
Appears in Collections:Master Theses and Ph.D. Dissertations

Files in This Item:
File Description SizeFormat 
محمد عليات.pdf3.2 MBAdobe PDFThumbnail
View/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

Admin Tools