Wetlands play a key role in regional and global environments, and are critically linked to major issues such as climate change, wildlife habitat, biodiversity, water quality protection, and global carbon and methane cycles. Remotely-sensed imagery provides a means to detect and monitor wetlands on large scales and with regular frequency. In this project, methodologies were developed to classify wetlands (Open Bog, Treed Bog, Open Fen, Treed Fen, and Swamps) from multi-source remotely sensed data using advanced classification algorithms. The data utilized included multispectral optical and thermal data (Landsat-5) and Radar imagery from RADARSAT-2 and Sentinel-1. The goals were to determine the best way to combine the aforementioned imagery to classify wetlands, and determine the most significant image features. Classification algorithms investigated in this study were Naive Bayes, K-Nearest Neighbor (K-NN), Support Vector Machine (SVM), and Random Forest (RF). Based on the test results in the study area in Northern Ontario, Canada (49°31′.34N, 80°43′37.04W), a RF based classification methodology produced the most accurate classification result (87.51%). SVM, in some cases, produced results of comparable or better accuracy than RF. Our work also showed that the use of surface temperature (an untraditional feature choice) could aid in the classification process if the image is from an abnormally warm spring. This study found that wetlands were best classified using the NDVI (Normalized Difference Vegetative Index) calculated from optical imagery obtained in the spring months, radar backscatter coefficients, surface temperature, and ancillary data such as surface slope, computed through either an RF or SVM classifier. It was also found that preselection of features using Log-normal or RF variable importance analysis was an effective way of identifying low quality features and to a lesser extent features which were of higher quality.

A critical step in object-oriented geospatial analysis (OBIA) is image segmentation. Segments determined from a lower-spatial resolution image can be used as the context to analyse a corresponding image at a higher-spatial resolution. Due to inherent differences in perceptions of a scene at different spatial resolutions and co-registration, segment boundaries from the low spatial resolution image need to be adjusted before being applied to the high-spatial resolution image. This is a non-trivial task due to considerations such as noise, image complexity, and determining appropriate boundaries, etc. An innovative method was developed in the study to solve this. Adjustments were executed for each boundary pixel based on the minimization of an energy function characterizing local homogeneity. It executed adjustments based on a structure which rewarded movement towards edges, and superior changes towards homogeneity. The developed method was tested on a set of Quickbird, ASTER and a lower resolution, resampled, Quickbird image, over a study area in Ontario, Canada. Results showed that the adjusted-segment boundaries obtained from the lower resolution imagery aligned well with the features in the Quickbird imagery.

Methodologies were developed to classify wetlands (Open Bog, Treed Bog, Open Fen, Treed Fen, and Swamps) from remotely sensed data using advanced classification algorithms through two hierarchical approaches. The data utilized included multispectral optical and thermal data (Landsat-5, and Landsat-8), radar imagery (Sentinel-1), and a digital elevation model. Goals were to determine the best way to combine imagery to classify wetlands through hierarchically based classification approaches to produce more accurate and efficient maps compared to standard classification. Algorithms used were Random Forest (RF), and Naïve Bayes. A hierarchically based RF classification methodology produced the most accurate classification result (91.94%). The hierarchically based approaches also improved classification accuracies for low-quality data, as defined through feature analysis, when compared to a nonhierarchical classifier. The hierarchical approaches also produced a significant increase in classification accuracy for the Naïve Bayes classifier versus the standard approach (∼12% increase) while not significantly increasing computation time – comparable in accuracy to the RF tests for around 20% the computational effort. Preselection of spectral bands, polarizations and other input parameters (Normalized Difference Vegetation Index, Normalized Difference Water Index, albedo, slope, etc.) using log-normal or RF variable importance analysis was very effective at identifying low-quality features and features which were of higher quality.