Cloud cover and satellite images of deforestation

Most studies of deforestation rely on land-cover data interpreted from satellite images. However, it is often difficult, particularly in moist forest areas, to obtain cloud-free images. Dropping or imputing missing values from satellite data relies on the assumption of independence of irrelevant alternatives (IIA), which implies that clouds are independent of the occurrence of forests. We test whether IIA holds for the case of Madagascar. We reject IIA for the amount of forest lost, demonstrating that estimates of deforestation and studies that use deforestation data from satellite images could potentially be biased without the proper corrections for cloud cover.