Daniel L. Zvirzdin and Dr. Steven Petersen, Plant and Wildlife sciences

Introduction

Since European settlement piñon (Pinus) and juniper (Juniperus) (PJ) woodlands have expanded their range to more than 40 million hectares; this expansion constitutes one of the greatest afforestations of our time and is due to many factors including high intensity grazing, fire suppression, increased atmospheric CO2 concentrations, and climate change (Romme et al. 2009). As these woodlands expand into other ecosystems negative impacts to ecosystem services often occur as natural fire regimes, plant community structure, water and nutrient cycles, and biodiversity are altered. While the extent of these negative impacts are site dependent, PJ cover and density can be important indicators of the degree to which these woodlands have modified natural processes (Miller et al. 2008). Accordingly, methods that enable land managers to economically monitor PJ woodlands over large land areas are highly sought after.

With recent developments in the fields of remote sensing and geographic information systems (GIS), we hypothesized that both cover and density could be extracted remotely in these systems. The remote extraction of tree cover data from high resolution aerial photography has recently received significant attention in the literature ( Afinowicz et al. 2003); however, research has been lacking that explores remote methods for measuring density. To the best of our knowledge, no studies have been published that establish methodologies for remotely extracting cover or density from PJ woodlands.

The primary objective of this study was to develop an efficient and effective method for accurately quantifying PJ canopy cover and density directly from high resolution remotely sensed aerial photographs through the use of feature extraction (FE) software. In developing these methods, a secondary objective was to optimally simplify the extraction process to augment utility for land managers. Results of this research have important implications for monitoring PJ woodland encroachment, fuel loads, biomass energy potential and rangeland health.

Methods

Remote extraction was performed on 25 cm color aerial photography obtained from the Utah Automated Geographic Reference Center (AGRC). Cover was extracted from the imagery using the Feature Analyst software extension for ArcGIS® 9.3. This program takes user input, in the form of digitized polygon training sets representing the desired feature (i.e. individual piñon and juniper trees), and applies custom mathematical algorithms to model similar features across landscapes. In this study we experimented with multiple training techniques, algorithms, and post-processing tools to maximize accuracy and efficiency. We concluded that 1) a minimum of 20 training sets are needed to consistently produce an output extraction file with a minimum 90% on-screen accuracy (as determined using the accuracy assessment tool in ERDAS Imagine 9.1), 2) the optimal algorithm for this type of vegetation and resolution is a pre-defined foveal pattern of nine cells, and 3) that it is best to digitize training sets that slightly overestimate cover, following which the “remove clutter” post-processing tool can be employed to pare down the output extraction file to be visually accurate.

A direct extraction of density from the output cover file was not possible as cover polygons often represented multiple trees (i.e. two or more trees in close proximity often appeared as one polygon). To circumvent this: 1) cover polygons were ranked by area and two categories were created, one with low-area polygons representing single trees and the other with high-area polygons representing multiple trees, and 2) a negative buffer was applied to the high-area polygons, breaking them out into smaller polygons representing individual trees. Following the application of the negative buffer, if individual polygons still represented more than one tree these steps were repeated. The dividing point between the two categories was determined by visually selecting polygons in ascending order, based on area, until polygons representing more than one tree began to be selected.
Results

As a part of a larger study outside of the scope of this grant, these methods were compared to ground reference cover and density data at 6 Division of Wildlife Resources Range Trend Project (DWR-RTP) sites throughout the state of Utah. We found a strong relationship between ground reference and feature extracted cover and density data; cover evidenced a near 1:1 relationship and the strongest correlation (R= 0.99, P < 0.001), density was underestimated by the proposed negative buffer technique, but still evidenced a strong correlation (R= 0.96, P < 0.001).

Conclusions

These results verify that the proposed cover and density extraction techniques are a viable option for land mangers seeking to monitor juniper encroachment over large landscapes. Cover extraction from high resolution remotely sensed imagery is extremely efficient as compared to field-based methods, and requires no calibration to obtain results comparable to these field methods. Density extraction using the proposed negative buffer technique is also high efficient, but requires calibration with ground reference data maximize accuracy. Further research is needed to refine density extraction techniques such that no calibration is needed.

This research was presented at the annual International Association for Landscape Ecology meetings in April 2009, and will be published as a component of the larger paper previously mentioned early in 2010. Our work in outlining these cover and density techniques was featured in Visual Learning Systems (the maker of Feature Analyst) monthly newsletter (see: http://www.featureanalyst.com/newsletters/JanFebnews.htm). Further work will be done on this project in writing and publishing a more user-friendly guide to using the proposed techniques.

References

• Afinowicz, J.D., Munster, C.L., Wilcox, B.P. & Lacey, R.E. 2005. A process for assessing wooded plant cover by remote sensing. Rangeland Ecology and Management 58:184-190.
• Miller, R. F., Tausch, R.J., McArthur, E.D., Johnson, D.D. & Sanderson, S.C. 2008. Age structure and expansion of piñon-juniper woodlands: a regional perspective in the Intermountain West. Res. Pap. RMRS-RP-69. U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station. Fort Collins, CO.
• Romme, W.H., C.D. Allen, J.D. Bailey, W.L. Baker, B.T. Bestelmeyer, P.M. Brown, K.S. Eisenhart, M.L. Floyd, D.W. Huffman, B.F. Jacobs, R.F. Miller, E.H. Muldavin. T.W. Swetnam, R.J. Tausch, and P.J. Weisberg. 2009. Historical and Modern Disturbance Regimes, Stand Structures, and Landscape Dynamics in Pinon-Juniper Vegetation of the Western United States. Rangeland Ecol. Manage. 62:203-222.