Large-Scale Canopy Height Mapping Using C-band InSAR Coherence and GEDI LiDAR Data

Dry forests and savannas have highly heterogeneous horizontal and vertical structure of woody vegetation and high temporal variability in moisture and phenology. This presents distinct challenges to SAR-based approaches to mapping canopy-height compared to tropical and temperate forests. Dense time-series of C-band (Sentinel-1A/1B) and, in future, L-band (NASA/ISRO NISAR) provide a pathway to reduce the impact of signal-noise and environmental conditions and to extract additional information more directly related to height, using repeat-pass Interferometric SAR (InSAR) techniques.  This work explores the potential information in Sentinel-1(S1) C-band InSAR temporal decorrelation and Global Ecosystem Dynamics Investigation (GEDI)-derived relative height metrics for canopy height estimation. The 12-day InSAR correlation (𝛾) and seasonal median were analyzed over a wide range of canopy heights and woody ecosystems, with a focus on dry forests and savannas in Australia, India, and South Africa as part of a NASA Carbon Monitoring System (CMS) investigation. To scale up the geographical extent, the algorithm was deployed on NASA’s Multi-Mission Algorithm and Analysis Platform (MAAP). To further enhance the computational efficiency, a tile-based approach is employed using the Military Grid Reference System. The canopy height maps were produced for  ~12.3 million km²  by processing nearly 1500 1⁰x1⁰ Sentinel-1 C-band Coherence tiles and approximately 130 million GEDI samples. In summary, this study provides new insights into the applicability and limitations of using InSAR data for canopy-height estimation in dry forest and savanna studies. Implementation of the proposed canopy-height estimation algorithm on the NASA-MAAP enables global scalability with current and future InSAR time series. The limitations with the C-band could be overcome partly, if not completely, with longer wavelengths like the L-band, such as those proposed for the upcoming NISAR, ALOS-3/PALSAR-4, and ROSE-L missions.

Figure 1. Maps of estimated canopy height for three countries using Sentinel-1 InSAR coherence and GEDI RH98. 

Test site	No. of1ᵒ x 1ᵒtiles	No. of GEDI samples x106	100m		500m		1000m
			r	RMSE	r	RMSE	r	RMSE
Australia	910	68.3	0.75	3.74	0.85	2.56	0.89	2.05
India	408	52.1	0.66	5.18	0.83	3.49	0.88	2.71
South Africa	165	9.2	0.44	3.35	0.56	2.38	0.59	2.02

Narayanarao Bhogapurapu¹ Paul Siqueira¹, John Armston², Mikhail Urbazaev²,  Xiaoxuan Li³, Konrad Wessels³, Laura Duncanson²

¹ University of Massachusetts Amherst,² University of Maryland, ³ George Mason University

Related Publications: 

• N. Bhogapurapu, P. Siqueira, J. Armston, Xiaoxuan Li, M. Urbazaev, K. Wessels, and L. Duncanson 2023 “Largescale forest stand height estimation using C-band InSAR correlation,” Geoscience and Remote Sensing Symposium(IGARSS), IEEE International.
• N. Bhogapurapu, P. Siqueira, J. Armston, Xiaoxuan Li, M. Urbazaev, K. Wessels, and L. Duncanson 2023 “Large-scale Canopy Height Estimation using C-band InSAR Correlation,” PolInSAR 2023: 11th International Workshop on Science and Applications of SAR Polarimetry and Polarimetric Interferometry and BIOMASS Workshop
• N. Bhogapurapu, P. Siqueira, J. Armston, Xiaoxuan Li, K. Wessels, L. Duncanson 2022 “Temporal analysis of C-band InSAR decorrelation for canopy height mapping over dry forests and tropical savannas “, AGU Fall Meeting 2022.