Fine Resolution Tree Height Estimation from Lidar Data and Its Application in SRTM DEM Correction across Forests of Sierra Nevada, California, USA

Abstract:

Sierra Nevada (SN) is a mountain range located in the northeastern California, USA, covering an area of 63,100 km². As one of the most diverse temperate conifer forests on the Earth, forests of SN serve a series of ecosystem functions and are valuable natural heritages for the region and even the country. The still existed gap of accurate fine-resolution tree height estimation has lagged ecological, hydrological and forestry studies within the region. Moreover, the Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM), as one of the most frequently used land surface elevation product in the region, has been proved systematically higher than actual land surface in vegetated mountain areas due to the absorption and reflection effects of canopy on the SRTM radar signal. An accurate fine resolution tree height product across the region is urgently needed for developing models to correct SRTM DEM.

In this study, we firstly developed a method to estimate SN tree height distribution (defined by Lorey’s height) through the combination of airborne lidar data, spaceborne lidar data, optical imagery, climate surfaces, and field measurements. Over 5 470 km²airborne lidar data and 1 000 plot measurements were collected across the SN to address this mission. Our method involved three main steps: 1) estimate tree heights within airborne lidar footprints using step-wise regression; 2) link the airborne lidar derived tree height to spaceborne lidar data and compute tree heights at spaceborne lidar footprints; 3) extrapolate tree height estimation from spaceborne lidar footprints to the whole region using Random Forest. The obtained SN tree height product showed good correspondence with independent field plot measurements. The coefficient of determination is higher than 0.65, and the root-mean-square error is around 5 m.

With the obtained tree height product, we further explored the possibility of correcting SRTM DEM. The results showed that the obtained tree height product can be used to effectively reduce the systematic bias of SRTM DEM using a regression model. After correction, the mean difference between SRTM DEM and airborne lidar derived DEM decreased from 12.15 m to -0.82 m, and standard deviation of the difference dropped by 2 m.