Artificial neural networks and geographic information systems for inland excess water classification

Due to its geographic position and climate, the Great Hungarian Plain is under continuous threat of droughts and floods. The year 2010 was one of the wettest years ever in Hungary. In the period October 2009 – December 2010, on the Great Hungarian Plain, 1149 mm precipitation fell, which corresponds...

Teljes leírás

Elmentve itt :
Bibliográfiai részletek
Szerző: van Leeuwen Boudewijn
További közreműködők: Szatmári József (Témavezető)
Mezősi Gábor (Témavezető)
Dokumentumtípus: Disszertáció
Megjelent: 2012-09-12
Tárgyszavak:
doi:10.14232/phd.1520

mtmt:2213420
Online Access:http://doktori.ek.szte.hu/1520
Leíró adatok
Tartalmi kivonat:Due to its geographic position and climate, the Great Hungarian Plain is under continuous threat of droughts and floods. The year 2010 was one of the wettest years ever in Hungary. In the period October 2009 – December 2010, on the Great Hungarian Plain, 1149 mm precipitation fell, which corresponds to a yearly precipitation of 919 mm, while the long term average yearly precipitation is 489 mm (in Szeged). The extreme precipitation caused exceptionally large areas to be flooded by inland excess water. The maximum total flooded area during this 15-months period was 355 000 ha on December 9, 2010 and the estimated financial damage to the agricultural sector alone exceeded 500 million Euros. Together with the consequential damage like soil degradation, inland excess water is one of the most severe natural hazards in the Carpathian basin. To be able to prevent or reduce damage due to inland excess water it is necessary to understand why and where it occurs. Most studies have tried to identify the factors that cause inland excess water and combined them using regression functions or other linear statistical methods. These methods have the disadvantage that they cannot deal with the nonlinear and complex functional relationships between those factors. Here, we present a different approach to identify and forecast inland excess water inundations using artificial neural networks (ANN) combined with geographic information systems (GIS). This approach has many advantages. First, it is independent of the statistical distribution of the data and there is no need to define the weight of the individual factors. Neural networks allow the target classes to be defined in relation to their distribution in the corresponding domain of each data source, and therefore the integration of remote sensing and GIS data is very convenient. Furthermore, ANNs are capable of incorporating uncertainty, incomplete data, incorrect sampling, multicollinearity between variables, spatial or temporal autocorrelation, and insignificance of single variables. These are common in geographic analysis, but especially in inland excess water research due to the fuzzy nature of the boundaries of the inundations and the complex interrelations between the factors. To facilitate the efficient application of classification of inland excess water occurrences by artificial neural networks, an integrated GIS ANN framework was created using a combination of ArcGIS, a geographic information system, Matlab, a mathematical modelling software and Python, an open source programming language. The framework was created to handle input data, intermediate results and output data in a flexible way in both ArcGIS and Matlab. In this way, it is possible to create the data files, test different network settings, perform training and simulation, and evaluate and visualize the training and simulation results efficiently. All steps are executed from within the GIS and no direct user interaction with the ANN software is needed. Many simulations have been executed to evaluate the quality of the method. The simulations were cross-referenced with the training data set and results of other traditional classification methods. The framework accurately classifies inland excess water based on the complex data sets. It gives better results than traditional classification, especially when more (GIS) data sets are incorporated.
Terjedelem/Fizikai jellemzők:111