Mapping Selective Logging in Tropical Forest with Spaceborne SAR Data
Rauste, Yrjö1; Antropov, Oleg1; Häme, Tuomas1; Ramminger, Gernot2; Gomez, Sharon2; Seifert, Frank Martin3
1VTT Technical Research Centre of Finland, FINLAND; 2GAF AG, GERMANY; 3ESA, ITALY

Selective logging is considered to cause forest degradation since it can reduce forest biomass for a longer period of time. Selective logging in the Republic of the Congo is conducted by following sustainable forest management. It is well controlled and documented and therefore offers an opportunity to test methods for mapping of forest degradation.

The objective of this study was to develop methods to effectively apply SAR data for the mapping of selective logging. The SAR data that were used in the project were acquired by ALOS PALSAR, Envisat ASAR and TerraSAR-X sensors. All elaborated techniques used are physically justified and based on microwave scattering phenomena, and utilize contextual, spectral, textural and statistical information of backscattered signal.

Mapping of selectively logged forest from dual-polarised (HH, HV) ALOS PALSAR data was based on multi-temporal change detection of cross-polarised signature from two years: 2007 and 2010. The algorithm also used additional land cover segmentation that was produced from 2010 dual-polarisation SAR data and contextual information from the detected network of newly constructed forest roads. The product obtained was a map of selectively logged forest area between the years 2007 and 2010 (Fig. 1). The ALOS PALSAR based map of selectively logged forest has spatial resolution of 12.5 m. Another map of selective logging was done with dual-polarised (HH, VV) TerraSAR-X spotlight data utilizing similar multitemporal approach.

Validation against optical VHR data showed high user's accuracy (95 %) for the ALOS/PALSAR-derived map of selectively logged forest. However, the area of selectively logged forest was under-estimated by 37.5 %. Overall accuracy was 70.4 %, but this figure was affected by the limited size of the validation area, which was chosen to include a large portion of selectively logged area. User’s accuracy was also very high (approaching 100 %) in the TerraSAR-X-based mapping of selectively logged forest, but the overall accuracy was 53.6 %.

Mapping of forest degradation due to selective logging with only Envisat ASAR data did not result in conclusive results. Newly constructed major forest roads could be visually detected in two scene pairs, but detection was not very clear. ASAR data availability in the service area did not allow using longer time-series before and after the logging operations. Possibility of use of multitemporal C-band SAR in mapping selective logging will be tested in spring-summer 2013.

ALOS-type L-band radar data could be applied in an operational system of selective logging detection over country-wide datasets, with proven potential to map newly constructed forest roads. If an under-estimation of area of selectively logged forest of about 40 % is not acceptable for selectively logged forest delineation, the L-band derived maps of selectively logged area and new roads can be used in a wider system to alert on areas of recent logging activity. The detected areas can then be covered by satellite or airborne optical data or ground surveys. The L-band map of areas with logging activity can also be used for stratification for additional surveys from ground reference or other data.

Figure 1. A sample of PALSAR-based selective logging in tropical forest: red = HH, green = HV, blue = binary layer of mapped selective logging. Image data copyright JAXA and METI 2010.