Large-scale forest stand height mapping in the northeastern U.S. and China using L-band spaceborne repeat-pass InSAR and GEDI LiDAR data

Abstract. This paper presents a global-to-local fusion approach combining spaceborne Synthetic Aperture Radar (SAR) Interferometry (InSAR) and LiDAR to create large-scale mosaics of forest stand height. The forest height estimates are derived based on a semi-empirical InSAR scattering model, which links the forest height to repeat-pass InSAR coherence magnitudes. The sparsely yet extensively distributed LiDAR samples provided by Global Ecosystem Dynamics Investigation (GEDI) mission enable the parametrization of signal model at a finer spatial scale. The proposed global-to-local fitting strategy allows for efficient use of LiDAR samples to determine adaptive model at reginal scale, leading to improved forest height estimates by integrating InSAR-LiDAR under nearly concurrent acquisition condition. This is supported by fusing the ALOS-2 and GEDI data at several representative forest sites. This approach is further applied to the open-access ALOS InSAR data to evaluate its large-scale mapping capabilities. To address temporal mismatch between the GEDI and ALOS acquisitions, the forest disturbances or deforestation areas are first identified by integrating ALOS-2 backscatter products and GEDI data. Further, a modified signal model is developed and analysed accounting for natural forest growth over temperate forest regions where the intact forest landscape along with forest height remain quite stable and only change slightly as trees grow. In the absence of detailed statistical data on forest growth, the modified signal model can be well approximated using the original model at regional scale via local fitting. To validate this, two forest height mosaic maps based on ALOS-1 data were generated for the entire northeastern regions of United States and China with total area of 18 million and 152 million hectares, respectively. The validation of the forest height estimates demonstrates improved accuracy achieved by the proposed approach compared to the previous efforts i.e., reducing from a 4 m RMSE on the order of 3–6-ha aggregated pixel size to 3.8 m RMSE at 0.8-ha pixel size. This updated fusion approach not only fills in the sparse spatial sampling of individual GEDI footprints, but also improves the accuracy of forest height estimates by 20 % compared to the interpolated GEDI maps. Extensive evaluation of forest height inversion against LVIS LiDAR data indicates an accuracy 3–4 m over flat areas and 4–5 m over hilly areas in the New England region, whereas the forest height estimates over northeastern China are best compared with small footprint LiDAR validation data even at an accuracy of below 3.5 m and with a coefficient of determination, R², mostly above 0.6. Given the achieved accuracy for forest height estimates, this fusion prototype offers as a cost-effective solution for public users to obtain wall-to-wall forest height maps at large scale using freely accessible spaceborne repeat-pass L-band InSAR (e.g. forthcoming NISAR) and LiDAR (e.g. GEDI) data.