GEOSCAN, résultats de la recherche

Menu GEOSCAN


TitreRandom forest modelling and classification of Landsat ETM+ and DEM data for surficial material mapping
AuteurParkinson, W; Richardson, M; Russell, H
Source33rd Canadian symposium on remote sensing, abstracts; par Canadian Remote Sensing Society; 2012 p. 40
Année2012
Séries alt.Secteur des sciences de la Terre, Contribution externe 20120280
Réunion33rd Canadian symposium on remote sensing; Ottawa; CA; juin 11-14, 2012
Documentlivre
Lang.anglais
Mediaen ligne; numérique
Formatspdf
ProvinceNunavut
SNRC75M
Lat/Long OENS-112.0000 -110.0000 64.0000 63.0000
Sujetstélédétection; établissement de modèles; divers
ProgrammeBases de données couvrant les trois territoires (la télécartographie prédictive), GEM : La géocartographie de l'énergie et des minéraux
LiensAbstracts (PDF, 1.22 MB)
LiensOnline - En ligne
Résumé(disponible en anglais seulement)
There is a need at the Geological Survey of Canada to apply improved accuracy assessments of satellite image classification and support remote predictive mapping techniques for geological map production and field operations. Most existing classification algorithms, however, lack any robust capabilities for assessing image classification accuracy and its variability throughout the landscape. In this study, random forest classification was used to improve overall image classification accuracy and to better describe its spatial variability across a heterogeneous landscape in Northern Canada.

Random forest models are stochastic implementations of classification and regression trees, which are computationally efficient and effectively handle outlier bias in predictor variables. This algorithm can also be used on non-parametric data sources, which is practical for surficial materials when similar materials types can manifest on a wide variety of spectral scales. The importance of individual variables is quantitatively assessed with respect to its ability to explain variance within the response variable. Prediction assessment metrics can be generated in order to identify areas of uncertainty in the classification. Random forest provides an enhanced classification compared to the standard maximum likelihood algorithms and greatly improves predictive capacity of satellite imagery.

Case study results from NTS map sheet 75M (northeast of Great Slave Lake) is presented for a synoptic geological material interpretation. The study area was classified with accuracy assessment diagnostics included as part of the classification product. Adoption of this methodology could improve field mapping planning, focus validation, and identify specific areas of interest in the landscape.
GEOSCAN ID291979