Cole Waters – Analyzing Structure in Motion Technology to Measure Above Ground Biomass – Cole Anwyl Walters

Introduction

The growing concern for the increase in carbon dioxide and industrial development has called for a paradigm shift to understanding the effects of fossil fuel combustion, land use change and forest land use. This is in turn has called for a fundamental understanding of global carbon sinks and the related biomass of densely vegetated environments being effected by infrastructure development and human fuel needs. Not only is this agenda evident and actively pursued in many major governments across the globe, a communal interest has matured to develop research about climate change and the associated factors. In the most recent decades, spatial analysts and environmental specialists have adopted remote sensing techniques to estimate the biomass of major areas of interest. However, as diverse as environments exist, the varied methods for measuring the biomass of respected locations have followed.

New remote sensing techniques have developed in the past few years to actively scan and develop models of canopy structure, stand height, vegetation biomass and carbon stocks in micro level areas through forms of LIDAR and a modified version of Photosynth (Dandois et. al 2010). Recent algorithms have been put into place to develop 3D models of local areas. Dandois used this concept and mounted a digital camera to a balloon and/or hexacopter to render canopy structure from a modified version of Photosynth called Ecosynth to begin gauging the abilities of 3D rendered images in an environmental setting. Similar to Photosynth, Photoscan is a software developed by Agisoft to render 3D models in a systematic manner. This system can then be georeferenced to measure area and volume of a structure. Through advanced processing and simple calculations, relating volume to biomass is possible from a basic estimation perspective.

This study will attempt to and pilot Photoscan’s ability to measure biomass from the perspective of volume and actively address how Photoscan can be modified to produce realistic estimations of biomass in a micro area upon further development and knowledge. The area under study takes place in the Ka’a’awa valley located on Oahu, Hawaii. The micro plot was chosen due to its bicultural arrangement of Monkey pod trees and Hau trees, thus presenting two very different types of canopy structure.

Methodology

(1)   Imagery and the associated geographic data was collected using a 16 megapixel Pentax WG-III with GPS capabilities mounted on a DJI Phantom Quadcopter with a GPS supported hovering accuracy of ± 2.5 horizontal meters and ± 0.8 vertical meters.

(2)   Georeference points were collected using a Trimble Yuma installed with TerraServer. Points were collected for approximately 270 seconds and exported off the field. Points were manually corrected via Google Earth due to technical difficulties with the Yuma Trimble and GPS Office.

(3)   Imagery was then rendered in a 3D format using Agisoft’s Photoscan. The 3D model and TIN was georeferenced from the previously collected points in the field. The point accuracy cloud was developed on “High.” Geometry and Texture was developed on “Medium.”

(4)   The 3D model was clipped to focus on the study area and prepare for volume calculation. In addition, the model was closed using a feature from Photoscan. Volume was calculated based on formation of the canopy. The volume was then exported to prepare for biomass calculation.

(5)   Biomass was calculated from a simplified version of a study developed looking at Above Ground Biomass in the Amazon (Brown 1997). From the perspective of this study, Monkey Pod Tree and Hau were separated for wood density calculations. Volume was calculated for each of the two species. Equations from the Brown study are represented in the context to this study. 

However, Wood Density (WD) for this purpose of this study needs to be weighted towards the more populous species in the proximity. An alternate form of the first Wood Density equation that is more suited for this study is provided directly below. As you can see, Monkey Pod (mp) and Hau (hau) are the only species of trees represented in this plot.

The Biomass Expansion factor is completely dependent on the size of the volume plot. In this case, the Biomass Expansion Factor can be derived by merely knowing the volume itself which is directly derived within Photoscan. This parameter is as follows:

Results & Discussion

The closed portion of the study area shows a biomass of approximately 22,607.3 tons. While the plot of study is heavily vegetated and dense in the portion of the Hau, this approximation is much too high. However, while this estimated amount shows the fallacies of the equation used, it shows the strengths, weaknesses and future of Photoscan’s ability to actively estimate biomass of a microenvironment.

The criticism of this equation’s methodology begins with its coupling with Photoscan. As we can see from the basic TIN of the studied area, Photoscan begins to “melt” the canopy when point cloud elevation drops. This is fairly extreme near the two locations of the monkey pod trees. While this is obviously unrealistic, Photoscan lacks the ability to recognize a geometric change from a canopy crown to the stock/trunk. This is probably mainly due to the lack of ground photos taken from an arbitrary angle. This is a fairly expected and accounted for problem when using Photoscan in a research setting, but it turns to be problematic when calculating the volume of a given location.

In response, Photoscan lacks the ability to measure volume let alone AG biomass of a fully developed crown-based canopy. However, Photoscan excels in being able to recognize and render very dense vegetation (ie. the Hau) that has fully developed shape and structure. Considering the heavy density of a proposed environment, estimated volume from a Photoscan naturally approaches the true volume due to absence of space between limbs. With the current Photoscan algorithms, Photoscan can be applied to understanding the formation and biomass of closed woodland, shrubland or certain types of agriculture.

Secondly, this academic community lacks to ability to justly measure minute values of above ground biomass from canopy measurements without the use of LIDAR. On a species basis, the ability to recognize the amount of space not occupied by a specimen is essential. Not only is the space unoccupied under the point cloud the largest creator for error, it negates Photoscan’s usefulness when directly compared to waveform LIDAR. Further research would ideally be focused towards creating a volume expansion index accounting for the unoccupied space per volume area of a unit. This might be created from knowledge about the species, average volume per species, surface area per unit and general understanding of growth patterns.

Lastly, Photoscan fully negates the ability to understand the terrain under the canopy. While this statement seems obvious due to Photoscan dependence on visible light to render a useable point cloud, it leaves a major flaw in the usefulness of Photoscan in a natural terrain where hills and depressions are likely. This issue might be quickly addressed in accepting that Photoscan will never have the ability to penetrate a fully developed brush canopy. In essence, Photoscan’s strengths in measuring AG biomass notes that it is not a costly version of LIDAR which certainly has the capabilities to penetrate a fully developed crowned ecosystem. Instead, accompanied with developed knowledge on the tool, Photoscan can be used on a micro level environment where elevation is relatively static (i.e. agriculture).

Conclusion

            Further development and research needs to be focused on creating a volume expansion index that can estimate the average amount of space not occupied by a species in a given area unit. As the paradigm moves away from individual based ASEs, it is perhaps Photoscan’s and other 3D modeling tools best interest to be centered around stand-level harvests and be directly related to the collected remotely sensed data. This way, Photoscan will able to adapt from previously predicted models and correct itself until the desired levels of accuracy and precision are achieved (Clark et al. 2012).

 Works Cited

Brown, S. 1997. Estimating biomass and biomass change of tropical forests. FAO Forestry Paper 124:0259-2800.

Clark D.B. and Kellner J.M. 2012. Tropical forest biomass estimation and the fallacy of misplaced concreteness. Journal of Vegetation Science 2012. Doi: 10.1111/j.1654-1103

Dandois, J. P. and E. C. Ellis. 2010. Remote sensing of vegetation structure using computer vision. Remote Sensing 2(4):1157-1176