Reusing Remote Sensing-Based Validation Data: Comparing Direct and Indirect Approaches for Afforestation Monitoring

Afforestation is one of the most effective processes for removing carbon dioxide from atmosphere and combating global warming. Landsat data machine learning approaches can be used to map afforestation (i) indirectly, by constructing two maps same area over different periods then predicting changes, or (ii) directly, a single analyzing observations change in both response remotely sensed variables. Of crucial importance, no comprehensive comparisons direct indirect monitoring are known have been conducted date. estimated through analysis may serve as intermediate products guiding selection samples production statistics. In this similar studies, huge effort dedicated collecting validation data. turn, those datasets varying sampling intensities areas, which complicates their use assessing accuracies new maps. As result, work done collect often not sufficiently exploited, with some being just once. study, we addressed main aims. First, implemented methodology reuse acquired via stratified strata constructed remote sensing Second, method acquiring comparing accuracy estimates precision country-wide mapping that occurred Italy between 1985 2019. To facilitate these comparisons, imagery, random forest classification, Google Earth Engine. The herein-presented produced 95% confidence interval classes. ranged 53 ± 5.9% class inside buffer—defined stratum within 120 m forest/non-forest mask boundaries—and 26 3.4% outside buffer. non-afforestation classes was much greater, ranging 87 1.9% buffer 99 1.3% Additionally, overall (with CI) were large (87 89 1.6%, respectively), confirming effectiveness introduced reusing relevance afforestation.