Afforestation monitoring through automatic analysis of 36-years Landsat Best Available Composites

The study of afforestation is crucial to monitor land transformations and represents a central topic in sustainable development procedures, terms climate change, ecosystem services monitoring, planning policies activities. Although surveying important, the assessment growing forests difficult, since cover has different durations depending on species. In this context, remote sensing can be valid instrument evaluate process. Nevertheless, while vast literature forest disturbance exists, only few studies focus almost none directly exploits data. This aims automatically classify non-forest, afforestation, areas using To purpose, we constructed reference dataset 61 polygons that suffered change from non-forest period 1988-2020. data were with Land Use Inventory Italy through photointerpretation orthophotos (1988-2012, spatial resolution 50 × cm) very high-resolution images (2012-2020, 30 cm). Using Landsat Best Available Pixel composites time-series (1984-2020) calculated 52 temporal predictors: four metrics (median, standard deviation, Pearson’s correlation coefficient R, slope) for 13 bands (the six spectral bands, three Spectral Vegetation Indices, Tasseled Cap Indices). verify possibility distinguishing forest, given differences between them minimal, tested models aiming at classifying following categories: (i) non-forest/afforestation, (ii) afforestation/forest, (iii) non-forest/forest (iv) non-forest/afforestation/forest. Temporal predictors used random which was calibrated search, validated k-fold Cross-Validation Overall Accuracy (OAcv), further out-of-bag independent (OAoob). Results illustrate distinction afforestation/forest reaches largest OAcv (87%), followed by (83%), non-forest/afforestation (75%) non-forest/afforestation/forest (72%). OA values confirm difference photosynthetic activity analysed distinguish them. are currently not exploited our results suggest it may support country-level monitoring reporting.