Quantifying forest growing stock volume in the boreal zone using spaceborne Envisat ASAR observations

Abstract

Bulleted list of abstract highlights:

• Quantification of forest growing stock volume (GSV) in the boreal zone with spaceborne synthetic aperture radar (SAR) observations of the backscattered intensity acquired by Envisat ASAR.
• Multi-temporal combination of estimates of GSV from single backscatter SAR measurements. GSV is retrieved by inverting a semi-empirical Water Cloud-like model of the forest backscatter.
• Forest GSV estimates for the year 2010 across entire boreal zone and for the year 2005 at pilot study regions. Comparison against forest field inventory measurements and Earth Observation raster datasets of GSV indicate a mean difference on the order of 40% at full resolution (1 km) and 25% when aggregating at resolution of at least 10 km.
• Consistency of estimates across different forest landscapes and composition. Relevance for updating obsolete or unavailable inventory records and in support of ecosystem modeling approaches.

Extended Abstract:

The use of observations of the backscattered radar intensity to quantify forest resources has been of considerable interest. Yet large-scale applications are few and mostly rely on data acquired during the 1990s. Since 2002, the Envisat Advanced Synthetic Aperture Radar (ASAR) has acquired images in the ScanSAR mode in a repeated manner. The large swath furthermore implies significant overlap of adjacent swaths, which in turn increases the number of observations of the radar backscatter for a given location. Exploitation of multi-temporal backscatter observations has been proven to increase the accuracy of forest growing stock volume (GSV) estimates compared to retrieval based on backscatter obtained from a single date. The ability to process multi-temporal backscatter has been implemented in an automated retrieval algorithm, referred to as BIOMASAR algorithm. The forest backscatter is modeled using a Water Cloud-like model; model parameters are determined pixel-wise with a self-calibration approach that exploits statistics of the radar backscatter for unvegetated and dense forest areas around the pixel. Such areas are identified by means of the MODIS Vegetation Continuous Fields product. The model is then inverted for each backscatter measurement available for the given pixel to obtain an estimate of the GSV. The estimates of GSV are finally combined to form a multi-temporal estimate using a weighted approach; the weights correspond to the difference of the backscatter between unvegetated and dense forests, which were estimated during the modeling phase.

In this paper, we provide an assessment of forest GSV for the entire boreal zone for the year 2010 using spaceborne radar data acquired by Envisat ASAR. In addition, we assess GSV changes with respect to 2005 for several regions located in Eurasia and north America. Some of these (Sweden, Central Siberia and Quèbec) served as pilot areas when assessing the capability of the BIOMASAR algorithm to map GSV over large regions. The resolution of the ASAR data is 1 km and estimates have been obtained for the 0.01 degree pixel size. Aggregated estimates at 0.1 and 0.5 degree are also considered because of interest to the ecosystem modeling community.

For the pilot study areas (in total 5.3 106 km2), the mean squared difference (MSD) between the ASAR-based GSV and several datasets of GSV based on field inventory measurements and Earth Observation data was between 36% and 55% for the pixel size of 0.01 degree; stronger agreement was obtained when the size of the samples used in the reference dataset was closer to the 1 km resolution of the ASAR data. The agreement increased with aggregation factor; at 0.5 degree the MSD was between 22% and 31%. ASAR-based estimates of GSV revealed (i) 10-15% overestimation of GSV in mature forest for Sweden, (ii) a 3% net increase of forest resources at selected locations of the Irkutsk Oblast, Central Siberia, with respect to 1998, (iii) extended patches of disturbances and underestimation not accounted for by the inventory records in Central Siberia, (iv) similar performance compared to established monitoring system based on optical imagery in Sweden and Québec and (v) superior quality with respect to information in global carbon stocks datasets.

The processing to generate the pan-boreal map of GSV for the year 2010 has been concluded at the time of submitting this abstract. The assessment of the GSV estimates is ongoing. Initial results were highly consistent to those generated from 2005, including areas of change resulting from harvesting (e.g., Russian Far East) and fire events (e.g., northeast China).

The GSV datasets derived from the Envisat ASAR imagery provide a significant contribution to the assessment of the current state of forests in remote areas of polar and boreal domains occurring in both Eurasian and North American continents, particularly where forest inventory is either obsolete or not available.

Conference

ForestSAT 2012

SLU Authors

• Fransson, Johan

  • Department of Forest Resource Management, Swedish University of Agricultural Sciences
    

UKÄ Subject classification

Remote Sensing
Forest Science


Permanent link to this page (URI)

https://res.slu.se/id/publ/43751