P. Behnia

A comparison of classification algorithms using Landsat-7 and Landsat-8 data for mapping lithology in Canada’s Arctic

To map Arctic lithology in central Victoria Island, Canada, the relative performance of advanced classifiers (Neural Network (NN), Support Vector Machine (SVM), and Random Forest (RF)) were compared to Maximum Likelihood Classifier (MLC) results using Landsat-7 and Landsat-8 imagery. A ten-repetition cross-validation classification approach was applied. Classification performance was evaluated visually and statistically using the global classification accuracy, producer’s and user’s accuracies for each individual lithological/spectral class, and cross-comparison agreement. The advanced classifiers outperformed MLC, especially when training data were not normally distributed. The Landsat-8 classification results were comparable to Landsat-7 using the advanced classifiers but differences were more pronounced when using MLC. Rescaling the Landsat-8 data from 16 bit to 8 bit substantially increased classification accuracy when MLC was applied but had little impact on results from the advanced classifiers.

Prospectivity mapping of Pb–Zn SEDEX mineralization using remote-sensing data in the Behabad area, Central Iran

Based on the characteristics of sedimentary exhalative (SEDEX) Pb–Zn deposits in the Central Iran Structural Zone, the following regional-scale geological criteria appear to be useful for prospectivity mapping of Pb–Zn mineralization in the study area: (1) presence and/or proximity to jarosite-, alunite-, and illite-bearing shales as the host rock; (2) presence of alteration iron oxide minerals; (3) proximity to – faults/lineaments; and (4) proximity to monzodioritic intrusive bodies with chlorite alteration minerals as the heat source. Spectral feature fitting (SFF) was applied to ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) data to map the alteration minerals using the reference spectra from the United States Geological Survey (USGS) library. X-ray diffraction analyses of samples taken from the shale and monzodioritic intrusive bodies were used to define the mineral content of these rocks. The geological map and field data were used to assess the alteration maps generated from the SFF method. Landsat ETM+ (Enhanced Thematic Mapper Plus) data were used to interpret faults/lineaments in the study area. Directional filters applied on the Landsat ETM+ 7-4-1 (red–green–blue (RGB)) colour composite image produced better images for visual interpretation of faults and lineaments. A fuzzy logic approach was used as a geographical information system-based geodata integration technique to predict the occurrence of SEDEX Pb–Zn deposits in the study area. The best predictive map generated from integration of input data defines 10% of the study area as having potential for Pb–Zn mineralization.

Remote predictive mapping of bedrock geology using image classification of Landsat and SPOT data, western Minto Inlier, Victoria Island, Northwest Territories, Canada

Supervised classification (robust classification method) of Landsat-7 and SPOT-5 data was used to analyse the bedrock geology of a part of the western Minto Inlier on Victoria Island, Canada. The robust classification method was used as it provides a series of uncertainty measures for evaluating the classification results. Six bedrock classes including gabbro, basalt, carbonate of the Wynniatt Formation, quartz-arenite of the Kuujjua Formation, evaporite of the Minto Inlet and Killian Formations and Paleozoic carbonate together with six surficial classes including vegetation were defined as the training data set. The resulting classified images derived from the Landsat and SPOT data were very similar in terms of the regional distribution of lithological classes, as reflected by fairly high classification accuracies for both image types. Gabbro and basalt, despite having a similar mineralogical composition, are spectrally distinct throughout most of the study area. Complicating spectral signatures of overlying glacial sediments and/or other overburden materials and spectral similarities between some of the lithologies caused poorer classification in some areas. Generally, the Landsat imagery provided better spectral separability between most of the lithological units than the SPOT imagery. However, in certain areas where the spectral separation between different lithologies is not dependant on the shortwave infrared-2 (SWIR-2; band 7 on Landsat) and/or blue bands (band 1 on Landsat), the SPOT imagery provided a better classification because of higher spatial resolution.

Landslide susceptibility modelling using the quantitative random forest method along the northern portion of the Yukon Alaska Highway Corridor, Canada

The random forest method was used to generate susceptibility maps for debris flows, rock slides, and active layer detachment slides in the Donjek River area within the Yukon Alaska Highway Corridor, based on an inventory of landslides compiled by the Geological Survey of Canada in collaboration with the Yukon Geological Survey. The aim of this study is to develop data-driven landslide susceptibility models which can provide information on risk assessment to existing and planned infrastructure. The factors contributing to slope failure used in the models include slope angle, slope aspect, plan and profile curvatures, bedrock geology, surficial geology, proximity to faults, permafrost distribution, vegetation distribution, wetness index, and proximity to drainage system. A total of 83 debris flow deposits, 181 active layer detachment slides, and 104 rock slides were compiled in the landslide inventory. The samples representing the landslide free zones were randomly selected. The ratio of landslide/landslide free zones was set to 1:1 and 1:2 to examine the results of different sample ratios on the classification. Two-thirds of the samples for each landslide type were used in the classification, and the remaining 1/3 were used to evaluate the results. In addition to the classification maps, probability maps were also created, which served as the susceptibility maps for debris flows, rock slides, and active layer detachment slides. Success and prediction rate curves created to evaluate the performance of the resulting models indicate a high performance of the random forest in landslide susceptibility modelling.

Prospectivity mapping for gold deposits hosted by iron formation, in a portion of Western Churchill Province that includes Melville Peninsula, Nunavut, Canada

Data-driven models employing weights of evidence (WofE) and logistic regression (LR) have been used in a series of experiments to characterize the spatial association of iron formation-hosted gold with geological and geophysical features in the Meadowbank to Melville Peninsula corridor of the northern Rae Domain. A number of geological features were extracted from a 1:550,000 bedrock compilation map and used as evidence. A total of 52 BIF-hosted gold occurrences were available for use in the training sets. For each experiment, weights were determined for individual evidence layers based upon the selected training sites. The evidence maps were reclassified into binary or ternary (2- or 3-class) maps guided by contrast values and associated studentized contrast values calculated on the ordered data. The evidence layers were combined using WofE and LR models to create posterior probability target maps for BIF-hosted gold deposits. To evaluate the validity of the mineral potential maps, training and testing sets were used to assess efficiency of classification and prediction of potential maps respectively.