Digitization of Trail Network Using Remotely-Sensed Data in the CFB Suffield National Wildlife Area

Transcription

1

2

3

4 Digitization of Trail Network Using Remotely-Sensed Data in the CFB Suffield National Wildlife Area Brent Smith DLE 5-5 and Mike Tulis G3 GIS Technician Department of National Defence 27 March 2007

5 Introduction The Canadian Forces Base Suffield National Wildlife Area (NWA) contains an extensive trail network created naturally by wildlife, grazing cattle, and artificially by oil and gas exploitation. The trail network has never been monitored in the past, and experience shows that increased human activity in NWA has a direct impact on expansion of the trail network. This is a concern from a wildlife conservation perspective because such trails cause habitat fragmentation. Objectives The objectives of this project were twofold: 1. to create a baseline digital database of trails using remotely-sensed imagery in a GIS system, to allow for year-to-year monitoring; and 2. to determine whether trails could be manually classified into distinct categories, based on pixel brightness and trail width characteristics identified from remotely-sensed imagery. Methods GIS Digitizing The NWA trail network database was created by heads-up digitizing 2005 SPOT ortho-rectified panchromatic imagery (2.5 meter resolution) in ESRI ArcGIS 8.3. In addition to digitizing the spatial extent of the network, trails were manually classified into 5 categories in an unsupervised approach; no pre-existing knowledge of ground-based trail characteristics was used to classify them. Classification was carried out using pixel brightness values (based on unprocessed 8-bit digital number/dn) and width of visible trail--the combination of the trail s width and brightness resulted in respective trail classification. After the trails were digitized and classified, ground-truthing was conducted in the Spring of 2006 to verify the classification accuracy. GIS Analysis In order to identify trends between trail categories and SPOT imagery, zonal statistics were created for each trail type using the ArcGIS Spatial Analyst extension. Trail network polylines were buffered by 2.5 m. Resulting buffered polygons were analyzed against 2005 imagery to identify and correlate SPOT DN values with each trail category. Ground-truthing Analysis In order to determine whether heads-up digitizing is a suitable method for digitizing and categorizing trails, a ground-truthing exercise was carried out in May Ground-based assessment included the field identification and characterization of digitized trail types, and an accuracy assessment. Initially, a field reconnaissance was employed to identify general characteristics of each of the categories identified by heads-up digitizing.

6 Several days of fieldwork were then required to collect information from 150 randomly-selected trail sites, stratified across all trail types. To minimize potential bias, GPS locations for each site did not include the predicted trail category; this was confirmed only after all of the field data had been collected. Field data included the collection of digital photographs, trail measurements (including rut widths, trail topography--the depth of the rut relative to surrounding areas--percentage of rut vegetative cover, and rut soil exposure. Results GIS Digitizing Based on the completed database, the following was digitized for each trail category: GIS Analysis a. category 1: km; b. category 2: km; c. category 3: km; d. category 4: km; and e. category 5: 49.0 km. Based on zonal statistics of raw DN values of 2005 SPOT panchromatic imagery underlying each trail type, there are insignificant differences between pixel brightness of each trail type (Table 1). Table 1 Trail Categories and DN attributes CATEGORY MIN DN MAX DN RANGE MEAN DN STANDARD DEVIATION MEDIAN Ground-truthing Based on pixel brightness and trail width characteristics, five major categories were identified and described as follows: a. Category #1 represents barely visible trails, minor continuous differences in pixel brightness in comparison to surrounding area, dark pixels with low pixel value (for example animal/cattle trails; b. Category #2 represents minor trails including heavily vegetated trails or pipeline in final stages of recovery, or multiple animal trails;

7 c. Category #3 represents moderate size trails with brighter pixels, interspersed with a few very bright pixels (visible, distinct trails); d. Category #4 represents wider trails with high pixel brightness values, including heavily-used trails with high soil exposure, poorly vegetated pipelines, or pipeline and trail combinations; and e. Category #5 represents trails with the highest pixel brightness values, and are not distinguishable from major roads except that trail widths are narrower. Accuracy Assessment Based on the accuracy assessment (Table 2), overall classification accuracy is 71%, with user accuracy ranging from 57 to 80%. Table 2 Accuracy Assessment of Digitized Trails Producer's Accuracy > Ground Condition (Field Observation) CATEGORY Total User's Accuracy Digitized V Total Producer's Accuracy User's Accuracy Overall Accuracy = 71% Categ 1 = 24 / 35 = 69% Categ 1 = 24 / 30 = 80% Categ 2 = 18 / 24 = 75% Categ 2 = 18 / 30 = 60% Statistic Value Categ 3 = 25 / 35 = 71% Categ 3 = 25 / 30 = 83% N = 150 Categ 4 = 23 / 37 = 62% Categ 4 = 23 / 30 = 77% Part A = 107 Categ 5 = 17 / 19 = 89% Categ 5 = 17 / 30 = 57% Part B = 4500 Khat = 64.17% OMISSION > Ground Condition (Field Observation) CATEGORY Total COMMISSION Digitized V Total Omission Commission Overall Accuracy = 71% Categ 1 = / 30 = 20% Categ 1 = 7+2+2/30 = 37% Categ 2 = / 30 = 40% Categ 2 = 4+2/30 = 20% Statistic Value Categ 3 = 2+2+1/30 = 17% Categ 3 = 2+5+3/30 = 34% N = 150 Categ 4 = 2+3+2/30 = 23% Categ 4 = 1+13/30 = 47% Part A = 107 Categ 5 = 13 / 30 = 43% Categ 5 = 2 / 30 = 7% Part B = 4500 Khat = 64.17%

8 Discussion Based on the combined attributes of field observations and GIS analysis, each trail category is described below, using trail width, soil micro-topography, vegetation cover, soil exposure of a single rut, and raw DN characteristics, including mean, and standard deviation (Std). Category 1: Width cm Topography 0-5 cm Vegetation Cover % (with exception of animal trails 5-15) Soil Exposure 0-20 % (with exception of animal trails 85+ %) Mean DN +/- Std /- 7.9 Note: Cattle trails have significantly greater soil exposure than access trails, but significantly narrower disturbance. Nonetheless, pixel brightness values are indistinguishable between animal trails and access trails using SPOT imagery. Category 2: Width cm Topography 2-20 cm Vegetation Cover % Soil Exposure 0-20 % Mean DN +/- Std /- 8.5 Note: Multiple parallel cattle trails could be digitized as category 2 trail, because of the increased soil exposure and higher pixel brightness values. Category 3: Width cm Topography 5-30 cm Vegetation Cover % Soil Exposure % Mean DN+/- Std / Note: Average size and the most common trail. Category 4: Width cm Topography 5-35 cm Vegetation Cover 0-50 % Soil Exposure % Mean DN+/- Std / Note: Wide, eroded trail with little vegetation.

9 Category 5: Width cm Topography 0-60 cm Vegetation Cover 0-15 % Soil Exposure % Mean DN +/- Std / Note: Typical category 5 trail has very wide ruts with no vegetation. Conclusions The field component and accuracy assessment of this study positively supported the objectives of this project. We conclude that manually digitizing and classifying trails based on remote-sensing imagery is effective and sufficiently accurate for creating and monitoring the trail network within the NWA. Because of the subtle and insignificant differences in pixel brightness between each of the trail categories, we conclude computer-based classification is not possible. The separation between category 4 and category 5, and separation between category 1 and category 2 trails during heads up digitizing is difficult to the inexperienced eye. Because of subtle differences in pixel brightness values, sufficient training is required to ensure that these categories can be accurately classified during heads-up digitizing.

10 Field Examples The following provide examples of field-based descriptions of each category. Category 1: Cattle Trail Grid Category 1 Pictures No to 47 Width (cm) Topography (cm) 0-5 Vegetation (%) 0-15 Soil Exp. (%) Heavily vegetated trail Grid Category 1 Pictures No to 55 Width (cm) Topography (cm) 0-5 Vegetation (%) Soil Exp. (%) Completely recovered trail Grid Category 1 Pictures No Width (cm) Topography (cm) 0-5 Vegetation (%) 95+ Soil Exp. (%) 5

11 Category 2: Heavily vegetated trail Grid Category 2 Pictures No to 43 Width (cm) Topography 0-10 (cm) Vegetation (%) Soil Exp. (%) 0-20 Heavily vegetated trail Grid Category 2 Pictures No to 45 Width (cm) Topography (cm) 0-5 Vegetation (%) Soil Exp. (%) 0-10 Heavily vegetated trail Grid Category 2 Pictures No Width (cm) Topography (cm) 5-20 Vegetation (%) Soil Exp. (%) 10-20

12 Category 3: Moderately-eroded trail Grid Category 3 Pictures No to 25 Width (cm) 60+ Topography (cm) Vegetation (%) Soil Exp. (%) Moderately-eroded trail Grid Category 3 Pictures No to 52 Width (cm) Topography (cm) 5-20 Vegetation (%) Soil Exp. (%) Eroded trail and pipeline Grid Category 3 Pictures No Width (cm) Topography (cm) Vegetation (%) Soil Exp. (%) 10-35

13 Category 3 to 4 transition: Heavily eroded trail on side slope Grid Category bottom 3 / top 4 Pictures No Width (cm) 35 / 90 Topography (cm) 5-10 / Vegetation (%) / 0-20 Soil Exp. (%) / Category 4: Heavily eroded trail Grid Category 4 Pictures No to 35 Width (cm) Topography (cm) Vegetation (%) 0-30 Soil Exp. (%) Eroded pipeline right-away Grid Category 4 Pictures No Width (cm) Topography (cm) 5-25 Vegetation (%) 0-50 Soil Exp. (%)

14 Category 5: High bare ground exposure Grid Category 5 Pictures No Width (cm) 80+ Topography (cm) Vegetation (%) 0-15 Soil Exp. (%) High bare ground exposure Grid Category 5 Pictures No Width (cm) 125 Topography (cm) 5 Vegetation (%) 0 Soil Exp. (%) 100