Table 1.
Details of the input data used to calibrate the model for the transition period 2001–2002 and 2009–2010(data name, description, source, reference year and type).
Figure 1.
Modelling procedure flowchart.
The flowchart illustrates the construction and running of the deforestation model. i is the model iteration, t is the year, ROC refers to the Receiver Operating Characteristic and AUC is the area under the ROC curve.
Figure 2.
Figure shows both training and testing maximum likelihoods achieved by each of the 106 models used to explain deforestation rates in the Brazilian Amazon in the (a) pre-PPCDAM and (b) post-PPCDAM scenarios.
Table 2.
Mean and 95% confidence intervals of the single variable models, for each scenario (pre- and post-PPCDAM).
Table 3.
Mean and 95% confidence intervals of the final set of parameter inputs used in the deforestation simulations, for each scenario (pre- and post-PPCDAM).
Figure 3.
Pre-PPCDAM model validation results comparing pixel by pixel predicted and observed deforestation between 2002 and 2010.
Three validation statistics are presented: (a) mean percent of perfect match; (b)errors of omission; and (c)errors of commission. Validations were conducted in two ways. The ‘annual’ validations compare predictions from a single year with observations for that same year, whereas the‘cumulative’ validations compare all deforestation predictions up to and including that year with observations of cumulative deforestation over the same time period. Variation in these values arises from the 100 model iterations.
Figure 4.
Pre-PPCDAM model validation showing the spatial dependence of model accuracy.
Values represent the proportion of (a) annual and (b) cumulative observed deforestation from 2002 through 2010 that fell within a threshold distance from predicted deforestation.
Figure 5.
Predicted deforestation rate in the Brazilian Amazon between 2002 and 2050.
Deforestation rates emerged from the local deforestation probabilities in the spatial model, for both the pre- and post-PPCDAM scenarios, and variation in these values arises from the 100 model iterations. Thick lines represent the median, boxes the inter-quartile range and whiskers the maximum and minimum simulated deforestation rates.
Figure 6.
Deforestation predictions for the Brazilian Amazon under the pre- and post-PPCDAM scenarios.
Cumulative deforestation probability in the year2050 (a) under the pre-PPCDAM and (b) the post-PPCDAM scenario; the wave of deforestation, represented as the median year in which each pixel was deforested (c) under the pre-PPCDAM and (d) the post-PPCDAM scenario; and uncertainty in the model predictions, quantified as the inter-quantile range of the year in which each pixel was deforested (e) under the pre-PPCDAM and (f) the post-PPCDAM scenario. In panels (c,d) and (e,f), measures of central tendency and variation were obtained by comparing model outputs from the 100 model iterations.