Fusion of lidar and multispectral data to quantify salt marsh carbon stocks
Academic Article
Overview
Research
Identity
Additional Document Info
Other
View All
Overview
abstract
2014. Herbaceous salt marshes are among the most productive ecosystems on earth. Unfortunately, quantification of the above-ground portion of biomass using passive optical remote sensing is constrained by the complexities of mixed spectral appearance in the water-land environment. Lidar remote sensing, on the other hand, has been extensively used to estimate forest biomass, and a few studies have reported their use in characterizing short or herbaceous plants. However, no empirical studies have demonstrated the combined use of lidar and spectral data to quantify above-ground biomass in herbaceous environments, including salt marshes. Thus, the findings of this study will contribute substantially to the understanding of potentials and limitations of using lidar and multi-spectral data for vegetation characterization and biomass estimates in salt marshes and other similar herbaceous environments. In this study, we evaluate the increased capability of a data fusion approach using small footprint discrete return lidar and multispectral data to quantify above-ground biomass and thus, carbon stocks in salt marshes. The specific objectives of our study were the following: 1) to understand the interaction between discrete-return airborne lidar and marsh vegetation; 2) to determine the appropriate grid size/s of lidar-derived datasets for characterizing marsh terrain and vegetation; 3) to investigate the applicability of a number of lidar metrics to predict salt marsh vegetation height and above-ground biomass; and 4) to evaluate the utility of integrating multispectral imagery with lidar to improve the predictability of regression models for quantifying above-ground biomass and carbon. Our results showed that salt marsh Digital Terrain Models (DTMs) derived using local minima in a grid spacing of 5. m. . 5. m provided the best accuracy in terrain elevation estimates with an RMSE of less than 10. cm. Lidar-derived maximum vegetation heights (Lmax) provided the best agreement with field height measurements, but explained only 41% of the variance in vegetation height measurements (RMSE. =. 5.85. cm). Regardless of the metrics used, lidar-measured heights underestimated the field vegetation height, which is consistent with the findings of previous studies in short or herbaceous vegetation. The fusion of lidar with multispectral data improved model predictions of live, dead, and total biomass quantities. The improvement provided by the fusion over the use of lidar or multispectral data alone was marginal; the combination explained 47% of the variance, whereas the best models using lidar and multispectral data separately explained 37% and 28% of variances in live biomass measurements, respectively. However, the best biomass prediction models reported considerably low RMSEs and % root square errors (%RSEs). For example, %RSE for the biomass prediction model using lidar-derived maximum vegetation height (Lmax) was closer to 20%, which is the recommended error threshold for remote sensing based forest biomass prediction models that can be repeatedly applicable for estimating forest carbon stock change. Our findings also demonstrate that lidar as compared to spectral data can provide better estimates of above-ground biomass and carbon, even in the herbaceous and low-relief context of a salt marsh.
