Comparison of SAR and Optical derived Data used in Forest Cover Detection; PALSAR-FNF vs. ESRI LAND-COVER over North Central Türkiye

Abstract

Swiftly and reliably establishing a spatially and geometrically correct land-cover map of any region is rather important in natural resource planning for conservation and utilization. JAXA’s PALSAR2/PALSAR/JERS-1 Mosaic and Forest / Non-forest maps, which as the name suggested, have specifically focused on global forest cover since 2007, benefiting from L-band SAR imagery. ESRI Land-cover, on the other hand, owing to exceptional Sentinel-2 imagery, has produced rather detailed land-cover maps including a distinct forest class. In this particular study, coverages of 2017–2020 readied by both institutions, utilizing the aforementioned imageries, were questioned on yearly basis against a rather detailed geodatabase which is still-in-effective use by two of the current regional directorates of forestry, Kastamonu and Sinop in Türkiye, utilizing long adopted accuracy metrics (user, producer and overall accuracies). When all year coverages were concerned, the best overall accuracies were held with 82% in 2017 ESRI land-cover and 83% in 2017 PALSAR-FNF. Both datasets yielded relatively good results in the forest class when user accuracies were investigated. ESRI land-covers managed more than 87% across all four years, while PALSAR-FNFs produced 84.33% in 2020 as the highest scoring year. As for producer accuracies, PALSAR-FNFs produced over 89% across all year coverages, while ESRI produced 84% in 2017 as the highest scoring year. It is worth noting that the ESRI land-covers had better compliance with the compartment boundaries of the reference geodatabase.