Exploring the spatial and temporal dynamics of deforestation associated with oil palm expansion in the Central Peruvian Amazon

Oil palm is rapidly expanding at the expense of the Peruvian Amazon. Understanding the spatial and temporal relationships between deforestation and oil palm expansion is important to identify where forests are vulnerable to oil palm expansion and how oil palm promotes additional deforestation. Additionally, robust mapping of the distribution of oil palm using satellite imagery is critical to locate oil palm for deforestation analyses.

This thesis first explored the promotion of indirect land-use change (ILUC) deforestation by oil palm expansion using panel regressions between 2001 and 2015 in the Central Peruvian Amazon. The thesis then explored the spatial influence of human infrastructure and population density on deforestation for oil palm expansion between 2001 and 2018 using an XGBoost classifier. Finally, the thesis compared the utility of two satellite imagery programmes for mapping oil palm using Convolutional Neural Networks (CNNs). These were Planet (5 m) optical imagery freely available for analysis in tropical forest regions as part of the Norway International Climate and Forests Initiative (NICFI) programme, and freely available Sentinel (10 m) optical and radar imagery from the Copernicus European Space Agency programme.

This thesis found that oil palm expansion promoted 13% of ILUC deforestation from 2006 to 2015. It was theorised that oil palm expansion displaced smallholders to forested areas and that smallholders cultivated oil palm whilst clearing forests to diversify agricultural income.

Oil palm expansion into forests was influenced by increasing population density and was greater the closer to human infrastructure. The remaining fragmented foreststhat are closer to human infrastructure are more likely to be cleared for oil palm than remote forests.

The CNN trained on NICFI 2019 Planet imagery classified 446 km2 of oil palm with 96% overall accuracy, and the CNN trained on 2019 Sentinel imagery classified 429 km2 of oil palm with 97% overall accuracy. This result demonstrated that free monthly Planet imagery can be utilised for monitoring oil palm expansion, to explore the promotion of deforestation from oil palm expansion and the influence of human infrastructure in further areas of the Peruvian Amazon.