Understanding the drivers affecting land use change in Ecuador: an application of the Land Change Modeler software
Dissertation Paper.pdf (18.56Mb)
Item statusRestricted Access
Krishna Rajan, Dhruva
MetadataShow full item record
Deforestation modelling is a relatively new field of study. The importance of this science has been advanced with the emergence of deforestation as one of the leading causes of global climate change. The advent of REDD (reduced emissions from deforestation and degradation) and related policy mechanisms has accelerated the need for modelling deforestation. This project looks at developing a methodology for modelling deforestation using the Land Change Modeler software. To generate the model of change in forest cover, satellite images (years 1996 and 2001) were used to produce land cover maps which were then used with the software to estimate probabilities of pixels changing from forests to other land use types. Various drivers of deforestation input into the model were proximity to roads, proximity to towns and slope. The model was developed by analysing change in forest cover between 1996 and 2001 and computing the probability of each cell undergoing change by deforestation. The success was measured by validating the different models (generated using various drivers with the LCM) for the year 2006 with a classified image of actual deforestation in 2006. The predicted rate of deforestation (based on the model generated) was 1.5% per year. This compares to an ‘actual’ rate calculated by a differencing of forest pixels in two satellite images from 2001 and 2006 of 1.8%. The results show that the software was able to accurately model 61% of the pixels that underwent deforestation when using all of the three drivers mentioned above. Testing the various drivers revealed that proximity to towns was the single driver most strongly influencing the predicted pattern and that predicted deforestation using a finer 20m resolution slope data was more similar to the actual pattern of deforestation observed in the 2006 imager data than the prediction based on a coarser 90m slope.