Cellular automata modelling of desertification using artificial neural networks: In south eastern Spain
Author
Petersen, Thomas Hallundbæk
Term
4. semester
Publication year
2014
Submitted on
2014-06-12
Pages
86
Abstract
Desertifikation bliver i stigende grad anerkendt som et problem i det sydøstlige Spanien. For at handle effektivt har politikere og planlæggere brug for klar viden om, hvor alvorlig den er, og hvor den breder sig. Dette speciale bygger en computersimulation for at give sådan viden. Modellen er en kontinuerlig cellulær automata, hvor et kunstigt neuralt netværk fungerer som overgangsregel (det vil sige den regel, der bestemmer, hvordan hver lokalitet ændrer sig over tid). Den kombinerer satellitbilleder, klimaprognoser og topografiske data for at simulere desertifikation som en geografisk proces. Et desertifikationsindeks konstrueres ud fra NDVI-satellitbilleder—et mål for planters grønhed—og jordfugtighed, og erosion medtages ved hjælp af RUSLE (Revised Universal Soil Loss Equation), en standardmetode til at estimere jorderosion. Med data fra 2001–2012 forudsiger modellen desertifikationsniveauer i 2013 med en gennemsnitlig unøjagtighed på 9,84 %. Resultaterne viser, at tilgangen kan fange komplekse rumlige processer som desertifikation, men fejlen er stadig for stor til direkte brug i politiske beslutninger. Der er behov for yderligere forskning for at forbedre modellens nøjagtighed.
Desertification is increasingly recognized as a problem in south-eastern Spain. To respond effectively, policymakers and planners need clear information about how severe it is and where it is spreading. This thesis builds a computer simulation to provide that knowledge. The model is a continuous cellular automata, where an artificial neural network serves as the transition rule (the rule that determines how each location changes over time). It combines satellite imagery, climate forecasts, and topographic data to simulate desertification as a geographic process. A desertification index is created from NDVI satellite imagery—a measure of vegetation greenness—and soil moisture, and erosion is considered using RUSLE (Revised Universal Soil Loss Equation), a standard method for estimating soil loss. Using data from 2001–2012, the model predicts desertification levels in 2013 with an average inaccuracy of 9.84%. The results show that this approach can capture complex spatial processes like desertification, but the error is still too high for direct use in policy decisions. Further research is needed to improve the model’s accuracy.
[This abstract was generated with the help of AI]
Keywords
Documents