Extended Data Fig. 9: Modelling of deforestation risk based on canopy heights and distance to forest edge.
From: Human degradation of tropical moist forests is greater than previously estimated
a) Marginal model plot of the binary logistic regression model of RH98 predicting whether deforestation was reported. The plot shows the expected influence of degraded forest structure (canopy height, expressed as a percentage of intact forest RH98) on the probability of deforestation. The grey shaded areas indicate the 95% credibility intervals of the predicted values. OR is the odds ratio. Stacked dots represent the GEDI samples deforested/not deforested. The marginal effect at the mean (MEM) quantifies for a one-point increase in RH98 (i.e. the x-axis) the associated percentage point variation in the probability of deforestation. AUC is the area under the ROC curve101. Marginal effects are partial derivatives of the regression equation for each variable in the model for each unit in the data. b) Plot of the marginal effects showing the probability of deforestation based on degraded forest structure (canopy height, expressed as a percentage of intact forest height) and the distance to the forest edge.