Creating a forest disturbance dataset for continental Portugal

Quality data is of the utmost importance to effectively evaluate any classification technique. In the realm of remote sensing, more precisely forest disturbance, different studies use different custom datasets. This short paper shows the methodology used to create a forest disturbance dataset for continental Portugal. The dataset contains 664 forest points generated from a stratified random sampling based on the tree species and climate zone that cover continental Portugal. Every point contains the auxiliary data used and a thorough list of all years where a disturbance occurs and the respective reason for the disturbance; the years span from 1986 until 2019. The analysis was done with the Google Earth Engine platform for a fast and flexible solution, tailored to the datasets needs. To complement the satellite time series, four auxiliary datasets were used to understand the cause of the disturbances and increase confidence. The resulting dataset shows known difference among various parts of the county. The south has a higher concentration of rejected data points due to its low-density forests. While the north area of the study contains a bigger number of points with multiple disturbances, the south is exactly the opposite with and abundance of undisturbed points. These asymmetries are also reflected on the species present in these different regions. This dataset may be of interest for studies that need to evaluate forest disturbance techniques, or even changepoint detection techniques, finally it may also be useful in studying that focus on differentiating types of forest recovery. The dataset is available in GitHub at https://github.com/EduardoFAFernandes/portuguese-forest-disturbance-dataset/ This work has been partially supported by project Floresta Limpa (PCIF/MOG/0161/2019).