Improved Assessments of Ecosystem Services through Active-Remote Sensing: Quantification of Timber and Snow Resources

Over the last century the natural environment has experienced an accelerated rate of climatic change in terms of both temperature and precipitation. The call for remote sensing methodologies that can provide reliable and accurate estimates of ecosystem services has never been stronger. This need is vital for assessing resources like above ground carbon stocks and water availability (from snow) as these resources have already been shown to be influenced by our changing climate. Although LiDAR (Light Detection and Ranging) is still a relatively young technology, only moving into mainstream use in the mid 1980's, it has since become the most promising remote sensing technique for high resolution elevation measurements in rugged and remote landscapes. Progressing our understanding of LiDAR error structure spatially, resulting from the interactions of vegetation and terrain, will allow us to generate more reliable assessments of carbon and water stocks in remote mountainous regions. The research within this dissertation is focused on improving the assessment of ecosystem services, by advancing our understanding of LiDAR-derived DEM vertical error and applying that knowledge to quantify the accuracy of associated value-added products. The work first evaluates the performance of two open-source and one commercial black-box LiDAR classification algorithms and identifies their sources of error in diverse landscapes. This showed that across algorithm there was no statistical difference, but that orders of magnitude differences could be found between different terrain and cover types. This demonstrated that LiDAR-derived DEM vertical error has little impact (<3% of timber volume) on derived forest metrics. These principals are then deployed in the development of LiDAR-aware allometrics for Abies grandis stem volume and biomass. Taking advantage of LiDAR's ability to measure tree height, it was shown that height based allometrics are as reliably accurate as traditional diameter-at-breast height equations. This project is intended as a proof of concept, meant to show how LiDAR data could be used operationally within forest management to directly estimate ecosystem services, including above ground carbon and stem volume. The same principals are then taken to the landscape scale and applied to spatially predict LiDAR-derived DEM vertical error through coupling Random Forest with intensive field surveying. The Random Forest produced model provided a representative, yet conservative, estimate of the DEM vertical error, on average predicted errors 10% greater than those seen in through field surveying. This methodology can be applied to any LiDAR-derived value-added product that is dependent on the creation of an accurate DEM (i.e. canopy height models, snow depth). The final object demonstrates the use of the predictive error methodology to a LiDAR-derived snow depth map. The model predicted errors deviated from the field measured errors with a RMSE of 0.09-0.34 m across the different cover types. The work highlighted the importance of snow drifting in semi-arid environments, with 30% of the catchment snow depositing in 10% of the area. This body of work; 1) improves the understanding of how vegetation and terrain influence the fine scale accuracy of LiDAR-derived DEMs; 2) demonstrates the minimal impact that LiDAR DEM errors have on LiDAR estimates of timber volume; 3) enhances the use of LiDAR in operational forestry by demonstrating the use of LiDAR-aware allometrics in the estimation of timber biomass and volume; 4) develops a method of predicting LiDAR-derived DEM vertical error at extended scales (e.g., catchments) through the use of Random Forest; and 5) provides a quantitative understanding of the spatial variability of error in a LiDAR-derived value-added product (snow depth), by apply the Random Forest methodology to a snow-on/off LiDAR acquisition.