Impact of Spatial and Temporal Variability of Tropical Forest Structure on LIDAR Estimation of Aboveground Biomass

Victoria Meyer, CALTECH/JPL, vic.meyer@gmail.com (Presenter)
Sassan Saatchi, CALTECH/JPL, sassan.saatchi@jpl.nasa.gov (Presenter)
Geoffrey Andrew Fricker, UCLA, africker@ucla.edu
Jerome Chave, CNRS, chave@cict.fr
Maxim Neumann, CALTECH/JPL, maxim.neumann@jpl.nasa.gov

Estimation of tropical forests biomass and their carbon stocks is an important factor in research on carbon cycle and environmental science. This study focuses on the estimation of carbon stocks in Barro Colorado Island (BCI), Panama by combining ground data and remote sensing data. Special attention is paid to the errors related to our method and to the limitations of the data. The spatial component of the analysis consists in determining what scale/resolution (0.04 ha, 0.25 ha, and 1.0 ha) works best in terms of biomass estimations, whereas the temporal component of the analysis focuses on comparing data from different dates. The ground data was collected every five years since 1995 over a 50 ha plot and was first analyzed independently. Forest biomass was estimated at each scale using allometric equations based on the tree diameter at breast height (DBH) and wood density. The Lidar data includes a small footprint discrete return lidar (DRL) dataset acquired in 2009 and a Laser Vegetation Imaging Sensor (LVIS) large footprint lidar dataset acquired in 1998. The percentiles of energy, or relative height, were extracted from both datasets.

LVIS metrics were corrected using DRL data by comparing their ground elevation. Each dataset was then used separately to estimate biomass using a nonlinear model based on all height metrics calculated at 0.04 ha, 0.25 ha, and 1.0 ha scales within the 50Ha. The results show that tropical forests have high spatial variability at smaller scales (0.04 ha). It gradually becomes less variable and with normal distribution when reaching larger scales (1.0 ha). Similarly, forest structure and biomass are highly dynamic at smaller scales due to gap formations and disturbances compared to their stable structure at coarser scale. Consequently, uncertainty in Lidar estimation of aboveground biomass decreases rapidly by moving up the scale and reaches a stable error at 100 m plot size (pixels). At this scale, the estimation model is also able to predict the changes of biomass to assess if any 1 ha forest area is gaining or losing biomass. However, the forest biomass change estimation between two dates and two sensors is not accurate enough to estimate gain and loss quantitatively. We used the models developed at the BCI 50 ha plots to predict the biomass over the entire island for 1998 (LVIS) and 2009 (DRL). These two maps were used to create a change detection map of BCI. This study is particularly interesting in terms of error estimation and propagation.

Presentation Type:  Poster

Session:  Coupled Processes at Land-Atmosphere-Ocean Interfaces   (Mon 4:00 PM)

Associated Project(s): 

• Saatchi, Sassan: Requirements for Spaceborne Fusion of Lidar and Radar Measurements of Forest 3-D Structure and Above-ground Biomass ...details

Poster Location ID: 56