|Title||Exploring human behaviour using agent-based modelling, neural networks and land use/land cover (LU/LC): a case study of flooding in the Limpopo River Basin, Xai-Xai, Mozambique|
|Author||Fligg, R A; Barros, J|
|Source||International Geographical Union (IGU) 2013 - Applied GIS and Spatial Modelling; 2013 p. 1-25|
|Alt Series||Earth Sciences Sector, Contribution Series 20130401|
|Meeting||International Geographical Union (IGU) 2013 - Applied GIS and Spatial Modelling; Leeds; UK; 2013|
|Program||Canada's Survey Program (9850) Canada's Survey Program|
Decision-making in GIS is often based on static mapping lacking the temporal dimension and emergent behaviour. Agent-based models offer a dynamic innovative way to solve
spatially related problems that can augment information for natural disaster mitigation and emergency management.
The paper presents an agent-based model for a flooding area of the Limpopo River Basin, Xai-Xai, Mozambique, using classified and
categorized spectral signatures of land features from GeoEye1 satellite imagery. The model explores the use of six types of neural networks; neuro-fuzzy, feed-forward, modular, recurrent, stochastic and learning to rank as a means for decision making
within this environment.
A hybrid design of neural networks was used to simulate the agent's cognitive ability to sense, learn and adapt when travelling over a landscape during a flooding episode. In the model, agents look for a safer route away
from the flooding, making decisions based on their assessment of land cover. The simulated human thinking process incorporates an optimal operating range of weighted information that increases the agent's ability to adapt to change.
was done using spatial statistics, regression analysis, and also by comparing the route selection to the original 50cm imagery. There were four significant results:
- Agents made intelligent decisions about the terrain that suggested simulated
human cognitive senses for survival.
- Limited awareness of the overall environment often resulted in poor decisions and agents getting stuck and drowning, in a behaviour which resembles 'panic'.
- Multiple agent interaction in the same
geographic area contributed to the neural network producing more efficient individual results.
- Multiple agent interaction in different geographic areas contributed to the neural network producing inefficient individual results.
suggest human behaviour was exhibited in the model; however a higher level of cognition and decision making by the agents would be expected with further development.
|Summary||(Plain Language Summary, not published)|
Decision-making in GIS is often based on static mapping lacking the temporal dimension and emergent behaviour. Agent-based models offer a dynamic
innovative way to solve spatially related problems that can augment information for natural disaster mitigation and emergency management.