1.

Introduction

The cumulative area of British Columbia (BC) that was affected to some degree by the mountain pine beetle (MPB) is estimated at 18.1 millions ha, and the Ministry of Forests, Lands and Natural Resources Operations (MFLNRO) predicts that 57% of the provincial lodgepole pine (Pinus contorta) inventory could be killed by 2021—these trees make-up over half the BC Interior’s annual harvest.¹ To lessen the impacts of the forest health crisis, the BC MPB Action Plan included a provision to recover the greatest value from dead timber before it burns or decays, while conserving the long-term forest values identified in land-use plans.²

The resulting accelerated timber harvest in the MPB epidemic area could compromise long-term forest productivity as wet soil conditions have expanded.³ With less interception and evapotranspiration, dead and dying lodgepole pine stands experience changes in water balance resulting in more soil moisture, a quicker hydrological response to precipitation events, and a greater frequency of increased water.⁴ Where soils “water up,” there is a higher risk of compaction and rutting during salvage operations.⁵

To evaluate if current forest practices conserve and protect soils in BC at the local scale, a monitoring process was put in place that measures soil-based indicators developed to assess the effectiveness of forest and range legislation in achieving stewardship objectives.^6,7 Indicators provide a protocol to assess changes in soil conditions in line with international standards and criteria for sustainable forest management set out by the Montreal Process.⁸9.10.^–11 Indicators are assessed through soil disturbance observable on the ground as proxy for longer-term effects, and therefore, provide a qualitative assessment of soil productivity and hydrologic function within the last two years after harvesting at the cutblock level.⁶ Measuring indicators are done mostly using high-resolution air photography in 10-cm pixel size to provide a rapid evaluation of soil attributes. A ground-based survey can complement the air photo review when site conditions are not favorable, i.e., certain disturbances are not visible on brush-rich grounds from scarifying the mineral soil to remove surface organic materials.

High-resolution air photography provides crisp-quality image for the assessment of soil disturbance at the local scale, but must be acquired specifically for a project. Their utility has been shown for measuring small-scale disturbance in forestry.^12,13 However, they are not readily available as satellite imagery. Additionally,, high-resolution air photography is costly considering quality satellite imagery are becoming increasingly available (i.e., IKONOS), and some are being provided for free or at reasonable cost over the Internet.¹⁴15.^–16 Within the context of the information needed, the immediate goal of the current project was to investigate whether there are sources of data other than highly pixelated aerial photography (e.g., $<10\mathrm{cm}$ resolution) that can be used to identify and map soil disturbance indicators following salvage harvesting using the imagery analysis method developed for Forest and Range Evaluation Program (FREP).

In this research, we examined available remote-sensed images of different spatial resolution with respect to (1) level of recognition and (2) accuracy assessment of indicators of soil conservation in association with harvesting activities within MPB areas.

2.

Methodology

2.2.

Study Area

The study area is mostly located in the central part of the interior plateau, which lies in the rain shadow of the BC Coast Mountains. Between 800 and 1400 m above sea level, the landform is rolled with few high peaks, and parent materials are mostly consolidated and of glacial origin. Precipitation can be as little as 250 mm annually in the valleys, but can reach 750 mm or more on western slopes. Soils are frozen in the winter, but in the summer soils “water-up” due to increased delivery of precipitation to the ground as the MPB affected trees die.¹⁹ Additional salvaged areas were added to the sample population just to the east of the Interior Plateau in the southern Rocky Mountains, where digital imagery was also available and site disturbance field data existed to confirm the photo interpretation. During the summer months, in each cutblock, clearcut harvests were implemented, and trees were felled mechanically and yarded to the roadside areas for processing. Ground survey results presented in this study are based on the BC Soil Conservation Survey Guidebook.²⁰ The study focused in forest districts, where the MPB outbreak was important within the Vanderhoof, Mackenzie, Rocky Mountain, Prince George, 100 Mile House, Central Cariboo, and Chilcotin Forest Districts (Fig. 1).

Fig. 1

Map of northern, central, and eastern British Columbia (BC), Canada. Locations of study sites are shown with a symbol “$\mathbf{K}$.”

3.

Results

3.1.

Identifiable Features

In central BC, we were able to find randomly 23 sites that had been salvage harvested and where digital imagery was acquired within 1 year of harvest. The amount of information available in each remote-sensed image is a function of its lowest object detail. The spatial resolution of some images limits the number of indicators that can be evaluated (Table 3). Access structures, particularly roads, and green tree retention in patches are evident on all remote-sensed images, but easier to discern on aerial photographs (Fig. 2). On Landsat 5 Thematic Mapper (TM) color images (30 m), roads are generally represented by dark pink pixels, whereas tree retention stands out in green. Exposed soils in landings and roads yield a brighter spot than the surrounding areas on black and white SPOT 5 (Satellite Pour l’Observation de la Terre) images. Surface wetness from drainage diversion displays as a darker color due to poor reflection response and, unfortunately, often blends in with natural ingress and brush making interpretation difficult.

Fig. 2

Examples of access structure and green tree retention on Mackenzie cutblock #239: (a) Landsat 5 Thematic Mapper (TM) (30 m), (b) BC orthophoto (1 m), and (c) aerial photo (0.1 m).

Small areas ($<0.2\mathrm{ha}$) where soil has been disturbed, such as landings and inordinate disturbance areas, often appeared fuzzy and blended in with other ground features on low-resolution images with a pixel size greater than 2.5 m (Table 3). In comparison, wheel ruts and skid trails are clearly visible on higher-quality images less than 1-m pixel (Fig. 3). They were mostly found near or within roadside work areas and often contributed to impeded natural drainage patterns. High-resolution data greater than 1-m pixel size allow discernment of single-tree retention. All other indicators of soil conservation, except for dead wood, escape notice on satellite images until pixel size equaled 2.5 m. Higher quality images facilitate detection of other disturbances such as areas of altered drainage and roadside work areas (Fig. 3). A spatial resolution of ${10\times 10\mathrm{cm}}^2$ is best suited for assessing the amount of dead wood (Fig. 4).

Fig. 3

Lynx Lake cutblock in Prince George Forest District showing smaller areas of soil disturbance such as altered drainage and roadside work areas: (a) Landsat 5 TM (30 m), (b) SPOT 5 (2.5 m), and (c) aerial photo (0.1 m).

Fig. 4

Coarse woody debris across a gully on Mackenzie block #239: (a) BC orthophoto (1 m) and (b) aerial photo (0.1 m).

4.

Discussion

Aerial photos ($<50\mathrm{cm}$ pixel size or approximately $1:16,000$) show the more obvious types of soil disturbance quite clearly, whereas the more subtle kinds of disturbances, i.e., when forest floor layers are still present, make it difficult to interpret compaction. Knowing the textural composition of the soil is important to assess compaction. However, soil texture cannot be extracted from an image (i.e., without a remote spectral signature). Rutting and other machine traffic were then used as proxy indicators of compaction.

Ground surveys for soil disturbance were conducted on 5 of the 23 sampled cutblocks to validate our results. The method using aerial photos succeeded to identify blocks that had significant amounts of soil disturbance (e.g., ruts and gouges) ranging from 8.1% to 49.1% of the net area to be reforested. Except for Landsat TM, satellite imagery, although less pixelated, was adequate to pick up roads and patches of tree retention. We were confident that both satellite imagery of 15 m resolution or less and high-resolution aerial photography were able to show actually what was there on the ground.

To get the most from remote-sensed images used for detecting soil disturbance, one needs to understand that spatial resolution limits the amount of information (i.e., the level of details that can be detected from a selected image format) (Fig. 5). Satellite imagery trades off quality against ground coverage in one swath (e.g., Landsat TM 185-km width versus small area coverage low elevation aerial photography), although pixel size as well as positional accuracy—not examined here—are rapidly increasing with advanced sensor technology. As a result, smaller size features, such as landings, were sometimes identified incorrectly as a result of fuzzy boundaries in the satellite image, something not observed on aerial photographs. These “hits and misses” can impact the estimates, but not significantly in the case of access structure considering that landings receive a smaller weight compared with roads in the computation of access structure due to their small size.

Fig. 5

(a) Landsat 5 TM image (30 m) provides limited information. Even permanent access structures are very difficult to identify. (b) The same block mapped at 10-cm pixel resolution from an aerial photo.

Overall, satellite images provided acceptable results for tracking productivity losses on access structures (Fig. 3). Although the amount of green tree retention patches were not assessed between image formats, our work suggests its accuracy would be similar to that of access structures. This review indicates that aerial photographs are more suited to map indicators other than loss of productivity due to access structures (e.g., areas affected by landslides, drainage diversion, or erosion), because there is less difficulty in distinguishing soil features from slash, brush, and natural regeneration or microrelief.²¹ High-resolution satellite imagery like SPOT 5 with pixel size ranging from 2.5 to 5 m could be useful, with an understanding that some information can be missed. However, newer satellite sensors such as GeoEye-1 (0.41 m) and IKONOS (0.82 m) may detect finer detailed features, and therefore, potentially be useful for estimating indicators that require a higher level of discrimination.²²

Good visibility of soil features in satellite images also depends on other factors. Local variation in terrain slope will influence how objects are reflected and captured by the sensors. In smaller blocks, various elements such as tree edge, landings, roads, or topography are present in one pixel and as a result, influence signal return producing a complex response making interpretation difficult (Fig. 6). With spatial resolution data less than or equal to ${5\times 5\mathrm{m}}^2$, roads running along cutblock boundaries are sometime difficult to map because they are obscured by the tree crowns. The appearance of the ground on satellite images also depends on weather conditions as satellites follow predetermined orbital passes and work on a fixed schedule. Figure 7 illustrates that detection can be difficult on a cloudy or hazy day. In contrast, aerial photograph can be taken whenever the sky is clear but ground coverage is limited, and the cost of acquiring them is generally higher than that for satellite photos. Whether or not one image format is more appropriate than another one should be decided by the user’s needs because information that can be derived from each image format is limited by its spatial resolution.

Fig. 6

(a) A Landsat 5 TM (30 m) image of Cranbrook block showing how block size affects the amount of identifiable features, in comparison with (b) high-resolution aerial photo (0.1 m).

Fig. 7

Weather conditions affected feature identification: (a) A cloud on a Landsat 5 TM image and its associated shadow, and (b) a cloud patch (outlined in red) on a SPOT 5 (2.5 m).

5.

Management Implications

The extreme level of soil disturbance found in at least one block suggests that there may be a very good reason to be concerned about the level of soil disturbance associated with salvage harvest operations. This level of soil disturbance would never be considered acceptable for normal timber harvesting, and it is important to determine what harvest operations were used on this cutblock and if it is indicative of increased soil disturbance caused by the changes in hydrologic function of the site. We were able to identify and measure soil disturbance using both satellite imagery and low-level aerial photography. Therefore, digital imagery can be used to assess the overall state of soil management in relation to salvage harvesting.

Information needs should guide the choice of an appropriate data source for monitoring soil conservation following salvage operations within MPB-affected areas. Moderate resolution satellite imagery between 15 and 30 m (e.g., Landsat 5 TM) would be likely sufficient for detecting and measuring the extent of roads and landings. Finer images provide higher accuracy measurements if necessary. Areas occupied by landslides, erosion, drainage diversion, inordinate disturbance, or roadside work areas can be captured on remote-sensed images with spatial resolution greater than 2.5 m, such as SPOT 5, and 1-m pixel size image, such as BC government orthophotos.

Based on this research and previous work, aerial photographs are best suited to describe and accurately assess the less-evident harvesting-related soil disturbance. Higher-resolution data deliver the best information for end-users, but costs and availability will also need to be considered before making a final decision.

The information presented in this study was obtained from manual digital images interpretation, which is a process that has its limitations.²² As digital imagery evolves to provide finer details, techniques such as classification (either pixel- or object-based) or digital elevation model (e.g., ConMap) can complement manual interpretation to achieve accuracy gain in soil mapping.²³ Our method will contribute information to the Montreal Process-Member Countries in evaluating the protective and productive functions of forests in relation to soil and water resources (Criterion 2—maintenance of productive capacity of forest ecosystems and Criterion 4—conservation and maintenance of soil and water resources).¹¹