Tandem-X interferometry in the prediction of forest inventory attributes in managed boreal forests

- Forest variables were estimated at plot level using Tandem-X InSAR data and a DTM.
- Height and coherence data were used in the estimation.
- Stem volume, basal area and Lorey's height were estimated.
- The corresponding relative RMSE values were 32%, 29% and 20%.
- Tandem-X InSAR data are a viable option in the mapping of forest resources.

Forest inventory attributes can be estimated using forest height data derived from remote sensing datasets. In this study, we estimated forest inventory attributes; i.e. stem volume, basal area, and Lorey's height using Tandem-X (TDX) interferometric synthetic aperture radar (InSAR) elevation models and coherence data assisted by ancillary terrain models. It is well known that SAR interferometry is sensitive to the structural changes taking place in the target between image acquisitions; e.g. changes caused by wind in vegetated areas. As regards non-simultaneous InSAR imaging, temporal changes in the target lead to loss of coherence and consequently to digital elevation models of low quality. Therefore, repeat-pass InSAR data have limited use in forest resource mapping. The problem of low coherence in forested areas can be partially avoided by using simultaneous InSAR data acquisition, e.g. the bistatic TDX SAR satellite system, which was launched in June 2010. We processed five interferometric TDX pairs from a test site located in Southern Finland, and collected data from 335 field plots, to study the accuracy of TDX InSAR elevation models and coherence data in forest resource mapping at a resolution of 314 m2. The SAR-derived elevation models (16 m2 pixel size) were converted to heights above the ground level using a digital terrain model based on airborne laser scanning data. The random forest (RF) method was used to create a model for estimating the forest attributes using remote-sensing-derived metrics as the predictors. The results for accuracy of prediction were the following: the relative RMSE of 32% was obtained for stem volume, 20% for Lorey's height, and 29% for basal area. The forest inventory attributes were derived from the TDX data with an accuracy equivalent with the accuracy of other remote sensing techniques. This result shows the potential of TDX data in robust and cost-effective forest inventory covering large areas.