Forest mapping with TanDEM-X: the global product and potentials for high-resolution classification

Abstract

Forests cover about 30% of the Earth’s landmasses, play an essential role in many dynamic processes of our planet, and are of extreme importance for all living species, allowing for the existence and preservation of biodiversity and healthy ecosystems. However, severe loss and degradation of forests is nowadays occurring at an alarming rate, hence putting this delicate balance in serious danger. For these reasons, an up-to-date assessment of the forest resource state becomes of crucial importance. In this context, spaceborne SAR represents a very attractive solution for regular mapping and monitoring of forested areas. Indeed, thanks to its weather and daylight independence and large coverage capabilities, spaceborne SAR is able to provide timely, high-resolution imaging at local, as well as regional and up to global scale. In this paper we present the new global Forest/Non-Forest classification map from TanDEM-X interferometric SAR data (InSAR) at X band, at a spatial resolution of 50 meters. The TanDEM-X mission comprises the two twin satellites TerraSAR-X and TanDEM-X, which fly in closed-controlled formation acting as a flexible single-pass radar interferometer. Among the different factors affecting the quality of InSAR data, the so-called volume correlation factor describes the coherence loss due to volume scattering effects, which typically occur in presence of vegetation. Therefore, the volume correlation factor is used as main indicator for the discrimination of forested from non-forested areas. The proposed classification method, the product validation with external reference data, and the comparison with existing land cover maps are presented. The global TanDEM-X classification mosaics will be released to the scientific community for free download and usage. In a second part of the paper the potentials of TanDEM-X data for high-resolution forest mapping are addressed. We present forest/non-forest classification mosaics of the State of Pennsylvania, USA, from TanDEM-X data processed at ground resolutions down to 6 m. For coherence estimation, both standard boxcar and nonlocal filtering methods have been considered, and a performance comparison verifies that nonlocal filters, thanks to their outstanding performance in terms of noise reduction capabilities and spatial features preservation, represent a promising approach to achieve a reliable classification at such fine resolutions.