FIRE-MAPS: SAR-derived tree canopy cover - Sedgwick

This dataset provides per-pixel percent tree canopy cover (TCC) derived from polarimetric radar backscatter and airborne lidar. TCC is defined as the proportional, vertically projected area of vegetation (including leaves, stems, branches, etc.) of woody plants above a given height (Sexton et al. 2013) and measured as the horizontal percentage of the ground surface that is covered by tree crowns, considering 1-arcsec (roughly 30x30m) pixels. The study site is Sedgwick Reserve, a 5,896-acre research, conservation and education facility located in Santa Barbara County’s Santa Ynez Valley in California. It consists largely of oak woodland, savanna, grassland, sage scrub, and riparian areas. The study area was surveyed during the NASA Fire Sense project and imaged with UAVSAR between 2023 and 2025. Airborne lidar observations acquired as part of the National Center for Airborne Laser Mapping (NCALM) in November 2020 were used to generate a canopy height model (CHM) at 1.0-m spatial resolution with an accuracy of 5.4 cm (Brande 2021). The lidar data covers approximately 16,309 acres, encompassing the entire Sedgwick Reserve. After making a reference 1-arcsec posting map, we computed, for each 1-arcsec pixel, the number of lidar pixels that were part of the canopy. The minimum height associated with tree canopies was determined by applying a range of height thresholds (Montesano et al. 2023). Results were compared against high-resolution optical images, and the threshold used in the present dataset was 1.37 meters, which is the standard height for measuring Diameter at Breast Height (DBH) of trees. Synthetic Aperture Radar analyses utilized UAVSAR observations collected on September 27, 2023 (Granule ID: SanAnd_26503_23019_000_230927_L090_CX_01). The analyses focused on backscatter values at polarizations HH, HV, and VV. We trained a 2-D convolutional neural network (CNN) using UAVSAR backscatter imagery (HH/HV/VV) as input features and NCALM lidar TCC as training data. Ten percent of lidar TCC data were set aside as a validation set, and a good correspondence between UAVSAR TCC estimates and lidar TCC was achieved (R² = 0.62, bias = -1.49 %, RMSE = 9.47 %).