Application of Future TerraSAR Data for Improvement of Forest Resource Assessments

In the context of sustainable forest management new requirements related to the content, scale, timeliness and cost efficiency of information are being imposed on timber suppliers and environmental conservationists. Traditionally, forest inventories have been based on cost-intensive and time consuming field inventories. Thus new requirements and financial constraints in public and private sector have caused a significant need for new cost-effective methods for data collection respectively the development of efficient inventory designs. For extensive areas forest surveys cannot be realized as full tallies but have to utilize sampling approaches. The cost-efficiency of sampling surveys may be increased by the integration of field assessments and wall-to-wall remote sensing imagery. In this context, the integration of SAR (Synthetic Aperture Radar) as remote sensing component into combined sampling methods is of great interest for forest inventory and monitoring due to the sensitivity of radar to structural characteristics of forests and its capability to cover large areas at low cost with high temporal resolution. The current study presents an approach for an efficient integration of X-band dualpol and L-band polarimetric SAR in preparation for the future TerraSAR system as a remote sensing component under boreal conditions. Two different sampling approaches were examined in test areas located in the boreal forest of Eastern Finland: (1) stratified sampling and (2) sampling with regression estimators. Two alternative stratification rules were tested: stratification with and without terrestrial apriori information. Simple and multiple linear regression estimators were applied. Using a terrestrial survey with systematically distributed samples provided a detailed reference data source for studying the relation of the backscatter signal as auxiliary variable and the terrestrial information as variable of interest. Segmentation was performed to estimate radar cross section as a basis to apply regression or stratification. This results either in a segment-wise estimate of the variable of interest or broad strata. As a central problem of the integration of remote sensing data and field assessments appeared the so called Small Area Problem (SAP); as terrestrial sample plots cover the segment area only partly, they do not reflect the entire variation of forest patches, resulting in differences between the information obtained on terrestrial sample plots and segments as application objects in relation to size, position and sensitivity of the backscatter means to bias caused by radar specific image features.