Modeling land cover changes in the cloudy region of Chocó (South America), one of the biodiversity hotspots of the world

Background/Question/Methods

The lowland rainforests in north-west South America are known as the Chocó biogeographic region. Due to a high degree of threat from deforestation as well as its high biodiversity and endemicity, the Chocó is one of the 25 biodiversity hotspots of the world. Analysis of Land Cover Changes (LCCh) in Chocó is a challenge in terms of remote sensors because this area is considered one of the rainiest places on the planet. Furthermore, the availability of high-resolution remote sensor data is low for undeveloped country areas. Consequently, the few satellite images available usually have a high percentage of clouds which poses a problem for the construction of land cover maps.  We sought to determine the LCCh in the Chocó region by building annual maps of land covers from 2002 to 2014. These maps were constructed by relating MODIS data after Fourier analysis (the correction of cloud effects) as predictor variables (EVI, NDVI, MIR, NIR, RED and BLUE) to visual interpretations of covers based on reference data of high spatial resolution (WorldView, Ikonos, QuickBird) using learning algorithms (Random Forest). 

Results/Conclusions

We classified 10576 MODIS pixel areas in nine land cover classes for the training and accuracy assessment. The land cover maps have a general accuracy of 79% (Kappa = 79%, SD=4.7) and show a reduction of the rainforest cover; however, it differed by the country that shares the Choco region: Panama, Colombia and Ecuador. NIR and EVI were the most important predictor variables for the classification.