The following is a summary of key findings following application of the methodology to eight case study sites:
Comparing Initial Classification to Revised Classification
Revised classification was necessary to correct over and underestimation by the initial classification in the large majority (92%) of cases.
Across the eight case study sites, the methodology measured disturbance at 48 time steps between 1984 and 2018. Disturbance estimates ranged from 0.1 km2 for a relatively small advanced exploration site to 26.1 km2 for a relatively large operating open pit mine.
Among the 48 time steps in which in which disturbance was measured, initial classification was more likely to overestimate disturbance than underestimate. In 62% of cases disturbance was initially overestimated, disturbance was underestimated in 30% of cases, while 8% showed no difference in disturbance between initial and revised classification.
For the large majority of time steps (79%), results of initial classification and those of revised classification differed by more than 10%, most of which were overestimates. In absolute terms, initial and revised classification results differed by less than 1 km2 in 46% of cases, by 1 – 5 km2 in 40% of cases, and by more than 5 km2 in 15% of cases, all of which were overestimates.
Effectiveness of the Methodology
The results of the classification, including initial and revised classification, when applied to the case study sites, indicate that the methodology, using medium-resolution Landsat images is effective at identifying and quantifying physical disturbance associated with mineral exploration and mining projects. Classified disturbance includes open pits, surface production facilities, on-site buildings, waste rock storage and tailings facilities, roads, and airstrips. Initial classification is effective at estimating physical disturbance when cloud and snow-free images are available and disturbance features can easily be distinguished from natural features, water, and disturbances associated with other human activities, particularly forestry. In other cases, e.g., when cloud or snow-free images are not available, and in areas with a high level of forestry activity, revised classification, supported by site-specific information and information provided by subject matter experts is necessary to prevent over or underestimates of disturbance size.
Limitations of the Methodology
Notwithstanding the foregoing, the classification process is often unable to distinguish between natural water bodies and on-site water features associated with a mine, such as settling ponds and flooded pits, which can, in initial classification, result in underestimation of disturbance. In two of the case studies, subject matter experts identified the underestimations, which amounted to more than 20% of the total disturbed area.
The classification process, using medium-resolution imagery, is effective at identifying and measuring disturbances associated with advanced-stage exploration projects, including buildings, roads, and airstrips. However, using medium-resolution imagery, the classification is less effective at identifying smaller disturbances associated with earlier stages of the exploration process. Drill pads, cut lines, and small exploration trails, for example, are not easily discerned by the classification using medium-resolution imagery. Many of these features share a spectral signature with natural features, such as wetlands or exposed rock, and forestry disturbances, which can result in overestimations of disturbance following initial training. In addition to using high-resolution imagery, a manualized process consisting of drawing polygons around known exploration-related disturbances, supported by site-specific information, including the location of trails, cut lines, and other disturbance features, would be required to reliably measure early and intermediate-stage exploration disturbances.