2022 ESA Annual Meeting (August 14 - 19)

COS 242-4 GEDI LiDAR: Opportunities and challenges for studying understory forest structure in the South American Moist Tropical Forest

10:45 AM-11:00 AM
515A
Alyson M. East, Montana State University;
Background/Question/Methods

The recent launch of the GEDI satellite LiDAR system is an exciting new opportunity to look at stand-level vertical forest structure of some of the world’s most inaccessible and imperiled forests. While GEDI is optimized for collecting vegetation profiles, the validation and error of GEDI metrics vary across biomes and regions due to different biophysical challenges. Specifically, the moist tropical forests of the Amazon yield particular challenges given that they are characterized by their tall, dense forests and an abundance of cloud cover. Thus far, GEDI accuracy and error reduction studies have overwhelmingly focused on top of canopy measurements. However, the lower canopy data frequently plays an essential role in many applications, including biomass estimates, understory fire effects, and habitat for many species. If these data are to be utilized in such research applications, it is imperative that we understand how accurate those data are throughout the canopy, especially given the specific challenges of the biome. To assess GEDI accuracy, we compared GEDI relative height metrics through the canopy to simulated GEDI footprints derived from fine-scale ALS LiDAR. We also tested common error reduction methods to assess if some data were better suited to capturing the full forest profile.

Results/Conclusions

In the comparison between GEDI LiDAR and high-resolution ALS LiDAR, we found that accuracy in the GEDI data attenuates through the canopy. This relationship holds true under all tested error reduction methods, including geolocation correction, beam and acquisition time filtering, and other common methods of GEDI error reduction. While reasonable data accuracy was achieved in the top of canopy measurements, the mid to lower canopy had increasingly high percent Root Mean Square Error and Bias with low R-Squared results. Unexpectedly, geolocation error correction had little effect on error rates and, in some cases, introduced an increase in bias. The best case for error reduction came from extensive data exclusion, resulting in an overall 22% decrease in sample size. In addition, derived metrics (metrics calculated as a composite of several different values such as canopy ratio) showed even worse accuracy, with many having almost no correlation to the validation dataset. These findings also have broad implications for the application of GEDI LiDAR in tropical forest studies, especially where understory accuracy is important.