The assessment of forest aboveground biomass dynamics on a large scale is a complex task requiring intricate methods and reliable data sources. The combination of satellite Earth observation and reference measurements from forest inventory field plots has been instrumental in this field, although these reference data are often limited in availability and quality. To address this challenge, researchers have utilized data from the Geoscience Laser Altimeter System (GLAS) aboard the ICESat satellite, which provided valuable Light Detection and Ranging (LiDAR) measurements from 2003 to 2010. These measurements are crucial for understanding canopy height and other metrics that correlate with aboveground biomass.
In recent years, the Global Ecosystem Dynamics Investigation (GEDI) LiDAR instrument, operating from the International Space Station since 2019, has further enhanced the ability to map canopy height with its denser coverage and smaller footprint sizes. This data is invaluable for creating a comprehensive dataset for training and validation of forest canopy height models.
Leveraging machine learning techniques, researchers have employed L-band Synthetic Aperture Radar (SAR) backscatter image mosaics from ALOS PALSAR-1/PALSAR-2 along with optical multispectral Landsat-derived percent tree cover maps to expand the canopy height insights from GEDI across Africa. A model was developed using airborne LiDAR data from various African biomes, which estimates aboveground biomass density based on canopy height. The LiDAR-based estimates were crucial for training this model, which was then used to generate annual aboveground biomass density maps for Africa from 2007 to 2017. These maps were validated against a comprehensive dataset of field plot measurements.
Datasets
Spaceborne LiDAR from GEDI: The GEDI L2B Canopy Cover and Vertical Profile Metrics product, available from NASA, was collected between April and June 2019. These metrics are based on a 25-meter footprint but carry some geolocation errors, making it challenging to use with moderate spatial resolution imagery.
Airborne laser scanning (ALS): This study utilized LiDAR-derived biomass density maps from airborne LiDAR data acquired in Gabon and Kenya. These datasets were pivotal in developing and validating the canopy height to biomass density model.
Field plot data of aboveground biomass density: The research incorporated 10,837 field plot measurements from across Africa, collected mainly between 2000 and 2017, to validate the 2017 aboveground biomass density map. The data varied in plot design and accuracy, necessitating adjustments to align them with the biomass map estimates.
ALOS PALSAR/ALOS-2 PALSAR-2 radar image mosaics: These mosaics, provided by JAXA, span from 2007 to 2017 (with some years missing) and underwent normalization to ensure consistency across the time series.
Percent tree cover data: Using a 30-meter Landsat-derived map of percent tree canopy cover for 2000 and annual tree cover loss data, researchers generated annual tree cover maps for the study period, crucial for the prediction of canopy height and biomass.
Modelling Framework
GEDI footprint selection and clustering: The research utilized the maximum footprint height from GEDI’s L2B product as the reference canopy height metric. A rigorous filtering process ensured only high-quality data was used for training and validation.
Canopy height modelling: Employing a Random Forest regression algorithm, the study developed a canopy height model using radar backscatter and Landsat data as predictors. The model was trained using aggregated data from GEDI footprints, improving accuracy by mitigating small sample errors.
The conversion from canopy height to aboveground biomass density involved an empirical linear model, and significant changes in biomass were assessed through a detailed change analysis. The resulting data provides insights into biomass stock and change trends across Africa, contributing to improved carbon inventories and understanding of biomass dynamics.
Original Story at www.nature.com