Résumé : En raison de sa grande étendue, le Nord canadien présente plusieurs défis logistiques pour une exploitation rentable de ses ressources minérales. La TéléCartographie Prédictive (TCP) vise à faciliter la localisation de gisements en produisant des cartes du potentiel géologique. Des données altimétriques sont nécessaires pour générer ces cartes. Or, celles actuellement disponibles au nord du 60e parallèle ne sont pas optimales principalement parce qu’elles sont dérivés de courbes à équidistance variable et avec une valeur au mètre. Parallèlement, il est essentiel de connaître l'exactitude verticale des données altimétriques pour être en mesure de les utiliser adéquatement, en considérant les contraintes liées à son exactitude. Le projet présenté vise à aborder ces deux problématiques afin d'améliorer la qualité des données altimétriques et contribuer à raffiner la cartographie prédictive réalisée par TCP dans le Nord canadien, pour une zone d’étude située au Territoire du Nord-Ouest. Le premier objectif était de produire des points de contrôles permettant une évaluation précise de l'exactitude verticale des données altimétriques. Le second objectif était de produire un modèle altimétrique amélioré pour la zone d'étude. Le mémoire présente d'abord une méthode de filtrage pour des données Global Land and Surface Altimetry Data (GLA14) de la mission ICESat (Ice, Cloud and land Elevation SATellite). Le filtrage est basé sur l'application d'une série d'indicateurs calculés à partir d’informations disponibles dans les données GLA14 et des conditions du terrain. Ces indicateurs permettent d'éliminer les points d'élévation potentiellement contaminés. Les points sont donc filtrés en fonction de la qualité de l’attitude calculée, de la saturation du signal, du bruit d'équipement, des conditions atmosphériques, de la pente et du nombre d'échos. Ensuite, le document décrit une méthode de production de Modèles Numériques de Surfaces (MNS) améliorés, par stéréoradargrammétrie (SRG) avec Radarsat-2 (RS-2). La première partie de la méthodologie adoptée consiste à faire la stéréorestitution des MNS à partir de paires d'images RS-2, sans point de contrôle. L'exactitude des MNS préliminaires ainsi produits est calculée à partir des points de contrôles issus du filtrage des données GLA14 et analysée en fonction des combinaisons d’angles d'incidences utilisées pour la stéréorestitution. Ensuite, des sélections de MNS préliminaires sont assemblées afin de produire 5 MNS couvrant chacun la zone d'étude en totalité. Ces MNS sont analysés afin d'identifier la sélection optimale pour la zone d'intérêt. Les indicateurs sélectionnés pour la méthode de filtrage ont pu être validés comme performant et complémentaires, à l’exception de l’indicateur basé sur le ratio signal/bruit puisqu’il était redondant avec l’indicateur basé sur le gain. Autrement, chaque indicateur a permis de filtrer des points de manière exclusive. La méthode de filtrage a permis de réduire de 19% l'erreur quadratique moyenne sur l'élévation, lorsque que comparée aux Données d'Élévation Numérique du Canada (DNEC). Malgré un taux de rejet de 69% suite au filtrage, la densité initiale des données GLA14 a permis de conserver une distribution spatiale homogène. À partir des 136 MNS préliminaires analysés, aucune combinaison d’angles d’incidences des images RS-2 acquises n’a pu être identifiée comme étant idéale pour la SRG, en raison de la grande variabilité des exactitudes verticales. Par contre, l'analyse a indiqué que les images devraient idéalement être acquises à des températures en dessous de 0°C, pour minimiser les disparités radiométriques entre les scènes. Les résultats ont aussi confirmé que la pente est le principal facteur d’influence sur l’exactitude de MNS produits par SRG. La meilleure exactitude verticale, soit 4 m, a été atteinte par l’assemblage de configurations de même direction de visées. Par contre, les configurations de visées opposées, en plus de produire une exactitude du même ordre (5 m), ont permis de réduire le nombre d’images utilisées de 30%, par rapport au nombre d'images acquises initialement. Par conséquent, l'utilisation d'images de visées opposées pourrait permettre d’augmenter l’efficacité de réalisation de projets de SRG en diminuant la période d’acquisition. Les données altimétriques produites pourraient à leur tour contribuer à améliorer les résultats de la TCP, et augmenter la performance de l’industrie minière canadienne et finalement, améliorer la qualité de vie des citoyens du Nord du Canada. / Abstract : Due to its vast extent, Northern Canada faces several logistical challenges for a profitable exploitation of its mineral resources. Remote Predictive Mapping (RPM) aims to help in targeting mineral deposits through the production of geological potential maps. Elevation data is necessary for the generation of these maps. However, the currently available elevation data north of the 60th parallel are not optimal primarily because it has been derived from contours with values at a metric precision. Additionally, exact knowledge of the vertical accuracy of elevation data is essential to insure a suitable use, within its accuracy constraints. This project aimed to improve the quality of elevation data and to contribute to the refinement of RPM products for a study site located in the Northwest Territories. The first objective was to generate control points to evaluate vertical accuracy with precision. The second objective was to generate an improved elevation model for the study site. First, a filtering method for Global Land and Surface Altimetry Data (GLA14) from the ICESat (Ice, Cloud and land Elevation SATellite) mission is presented. This filtering is based on indicators, derived from information available in GLA14 data and terrain conditions, which are then applied successively to remove potentially contaminated elevation points. The points are filtered based on the attitude calculation, signal saturation, equipment noise, atmospheric conditions, slope and number of peaks. Next, a method to generate an improved Digital Surface Models (DSM) using StereoRadarGrammetry (SRG) with Radarsat-2 (RS-2) images is described. In the first part of the adopted methodology, DSM are stereorestituted from RS-2 image pairs, without control point. Then, the vertical accuracy of the DSM is calculated using the control points resulting from the filtering of GLA14 data, and analysed according to the incidence angles combination used for the stereorestitution. Next, selections from the preliminary DSM are assembled to generate 5 DSM, each covering entirely the study site. Finally, the DSM are analysed to identify the optimal selection for the area of interest. The selected indicators were found to be efficient and complementary, except for the indicator based on the noise/signal ratio. Otherwise, all indicators allowed to filter out points exclusively. A 19% reduction of the elevation mean square error was achieved with the filtering method, when compared to Canadian Digital Elevation Data (CDED). The initial density of the GLA14 allowed maintaining a spatially homogeneous distribution of the post-filtering elevation points despite a 69% rejection rate. From the analysis of the 136 preliminary DSM, no specific combination of the acquired RS-2 images incidence angles stood out as being ideal with SRG due to high variability in vertical accuracy. Nonetheless, the analysis showed that images should be ideally acquired at sub-zero temperatures to minimize radiometric discrepancies between scenes. Results also showed that the slope is the main factor influencing the accuracy of DSM generated with SRG. The best vertical accuracy (4 m) was achieved with same-side view configurations. Opposite-side view configurations, despite achieving a vertical accuracy of 5 m, allowed a 30% reduction in the amount of images initially acquired. Therefore, the use of opposite-side view configurations could help to improve the efficiency of SRG projects by reducing considerably the acquisition period. Elevation data generated using the proposed method could help to improve results from RPM and increase the efficiency of the mining industry in Northern Canada and finally contribute to the betterment of the lives of Northern Canada’s citizens.

La télédétection possède plusieurs applications potentielles pour le suivi de l’humidité de surface du sol (0 à 5 cm de profondeur). Un suivi de l’humidité du sol à période régulière permettrait de nombreuses applications en hydrologie, climatologie, suivi d’événements météorologiques et agriculture de précision. Le signal radar à synthèse d’ouverture (RSO) en bande C tel que celui de RADARSAT-2 est sensible aux variations des paramètres du sol et de la végétation selon certaines conditions. L’inversion de modèles de rétrodiffusion linéaire a permis l’estimation de l’humidité du sol en zone agricole, mais pour des domaines de validité très restreints. Diverses missions satellitaires en cours ou futures permettent l’acquisition d’images radars polarimétriques. Les variables cohérentes déduites de ces images permettent de mieux décrire les cibles observées et elles ont permis l’estimation de l’humidité du sol pour un sol nu. Toutefois, le potentiel d’utilisation de la polarimétrie pour des cibles couvertes de végétation est encore mal connu. L’objectif de ce projet est d’évaluer le potentiel de la polarimétrie pour l’inversion de l’humidité du sol en zone agricole à partir d’images RSO en bande C. La campagne SMAPVEX12 menée à l’été 2012 au Manitoba a permis l’acquisition simultanée d’images polarimétriques RADARSAT-2, ainsi que des conditions du sol et de la végétation pour des champs de blé. La rétrodiffusion radar, en polarisation linéaire ou circulaire, est très sensible à l’humidité du sol avant l’épiaison du blé. Après, la végétation domine le signal. La calibration du modèle semi-empirique des canaux linéaires de rétrodiffusion, développé par Gherboudj et al. (2011) n’a pu correctement représenter les relations de la rétrodiffusion avec les caractéristiques agricoles observées. L’information de phase conservée par le capteur de RADARSAT-2 permet l’extraction de variables polarimétriques telles que la différence de phase HH-VV et la hauteur de socle, l’anisotropie A et l’entropie H issues de la décomposition de Cloude-Pottier dont la sensibilité à l’humidité du sol sera étudiée. Des modèles empiriques simples, calibrés par régression linéaire multiple de termes utilisant de 2 à 6 variables polarimétriques, ont été développés et ont permis d’estimer l’humidité du sol sur 5 champs de blé pour toute leur période de croissance avec une erreur RMSD de 0,074 m³/m³ en expliquant plus de 53.5% (R2) de la variance des valeurs d’humidité du sol observées, contre une erreur de 0.098 m³/m³ et une variance expliquée de 19.0% pour un modèle empirique basé que sur les variables incohérentes. / Abstract: Remote sensing has been widely researched toward estimation of soil conditions over agricultural fields. Monitoring of surface soil moisture mv would benefit many applications in hydrology, climatology, precision agriculture and risk reduction applied to meteorological events. C-band synthetic aperture radar (SAR) signal’s, such as that of RADARSAT-2, is sensitive to soil and vegetation characteristics. Backscattering coefficients obtained from those sensors allowed the estimation of mv by inverting empirical or semi-empirical models, under very strict conditions that limit their applicability. Many on-going or future missions provides polarimetric SAR images. However, the potential of polarimetric SAR sensors operated in c-band is not yet fully understood for soil moisture estimation over vegetated fields. This paper study the effects of soil and vegetation characteristics on polarimetric RADARSAT-2 images and proposes a simple empirical model based on polarimetric parameters extracted from RADARSAT-2 imagery to retrieve surface soil moisture (0-5 cm) over agricultural fields. The data used in this study was obtained during the SMAPVEX12 campaign, which occurred on the summer of 2012 between june 6th and july 17th in Manitoba, Canada. Fully polarimetric RADARSAT-2 images were acquired over 13 wheat fields over their whole growth cycle while their soil and vegetation conditions were monitored. Linear backscattering showed significant correlations for all polarizations before crops flowering. Sensitivity analysis of the extracted polarimetric variables to soil moisture demonstrated distinct correlations before and after the beginning of the crops flowering stage. The calibrated semi-empirical model proposed by Gherboudj et al. (2011) showed poor representation of the observed relationships between linear backscattering channels and crop conditions. The phase information, obtained by the RADARSAT-2 sensor, allowed extraction of polarimetric variables. Among those, phase difference HH-VV, the pedestal height and both the anisotropy H and entropy H obtained from Cloude-Pottier decomposition, showed significant correlations to soil moisture. A simple empirical model, calibrated with multiple linear regression from 2 to 6 polarimetric variables, allowed to retrieve soil moisture with a RMSD of 0,074 m³/m³ while it explained more than de 53.5% (R2) of observed soil moisture variability, while a simple linear model based only on incoherent variables could only estimate soil moisture with a RMSD of 0.098 m³/m³ and a R2 value of 19.0%.

Wildlife habitat mapping strongly supports applications in natural resource management, environmental conservation, impacts of anthropogenic activity, perturbed ecosystem restoration, species-at-risk recovery and species inventory. Remote sensing has long been identified as a feasible and effective technology for large-area wildlife habitat mapping. However, existing and future uncertainties in remote sensing will definitely have a significant effect on relevant scientific research, such as the limitation of Landsat-series data; the negative impact of cloud and cloud shadows (CCS) in optical imagery; and landscape pattern analysis using remote sensing classification products. This thesis adopted a manuscript-style format; it addresses these challenges (or uncertainties) and opportunities through exploring the state-of-the-art optical and radar remotely sensed data for large-area wildlife habitat mapping, and investigating their feasibility and applicability primarily by comparison either on the level of direct remote sensing products (e.g. classification accuracy) or indirect ecological model (e.g. presence/absence and frequency of use model based on landscape pattern analysis). A framework designed to identify and investigate the potential remotely sensed data, including Disaster Monitoring Constellation (DMC), Landsat Thematic Mapper (TM), Indian Remote Sensing (IRS), and RADARSAT-2, has been developed. The chosen DMC and RADARSAT-2 imagery have acceptable capability of addressing the existing and potential challenges (or uncertainties) in remote sensing of large-area habitat mapping, in order to produce cloud-free thematic maps for the study of wildlife habitat. A quantitative comparison between Landsat-based and IRS-based analyses showed that the characteristics of remote sensing products play an important role in landscape pattern analysis to build grizzly bear presence/absence and frequency of use models.

Wildlife habitat mapping strongly supports applications in natural resource management, environmental conservation, impacts of anthropogenic activity, perturbed ecosystem restoration, species-at-risk recovery and species inventory. Remote sensing has long been identified as a feasible and effective technology for large-area wildlife habitat mapping. However, existing and future uncertainties in remote sensing will definitely have a significant effect on relevant scientific research, such as the limitation of Landsat-series data; the negative impact of cloud and cloud shadows (CCS) in optical imagery; and landscape pattern analysis using remote sensing classification products. This thesis adopted a manuscript-style format; it addresses these challenges (or uncertainties) and opportunities through exploring the state-of-the-art optical and radar remotely sensed data for large-area wildlife habitat mapping, and investigating their feasibility and applicability primarily by comparison either on the level of direct remote sensing products (e.g. classification accuracy) or indirect ecological model (e.g. presence/absence and frequency of use model based on landscape pattern analysis). A framework designed to identify and investigate the potential remotely sensed data, including Disaster Monitoring Constellation (DMC), Landsat Thematic Mapper (TM), Indian Remote Sensing (IRS), and RADARSAT-2, has been developed. The chosen DMC and RADARSAT-2 imagery have acceptable capability of addressing the existing and potential challenges (or uncertainties) in remote sensing of large-area habitat mapping, in order to produce cloud-free thematic maps for the study of wildlife habitat. A quantitative comparison between Landsat-based and IRS-based analyses showed that the characteristics of remote sensing products play an important role in landscape pattern analysis to build grizzly bear presence/absence and frequency of use models.

Soil freezing and thawing processes are of particular importance for agricultural areas. For example, frozen soils can increase the runoff during snowmelt in the spring. Freezing and thawing also have a direct influence on the sowing and harvesting dates, as well as on the crop yield. A better understanding of those phenomena is therefore important, and several researchers focused on this topic in the past. Due to its sensitivity to changes in the state of water, microwave remote sensing is an appropriate tool for that purpose. The main objective of this study is to monitor soil freezing and thawing processes using AMSR-E and RADARSAT-2 polarimetric data acquired over an agricultural area located near Saskatoon (Saskatchewan). With AMSR-E data, the goals are to compare different combinations of frequencies for the spectral gradient's algorithm regarding their capacity for detecting frozen soils, and to analyze the temporal dynamics of the brightness temperature in order to find a new indicator of soil freezing. As for RADARSAT-2 data, several polarimetric parameters and techniques are tested in order to identify soil freezing. For the first part concerning AMSR-E data, a global precision for the discrimination of frozen and thawed soils higher than 90% was obtained with the spectral gradient's algorithm, for the combinations including high (18.7 and 36.5 GHz) and low (6.9 and 10.7 GHz) frequencies as well as for the one using only high frequencies. It is shown that, for the combination based on the 18.7 and 36.5 GHz frequencies, results are improved when a negative threshold is used for the spectral gradient. When high and low AMSR-E frequencies are combined, a null threshold is on the contrary appropriate, which constitutes an operational advantage. A new algorithm for detecting frozen soils, based on a thresholding approach applied to the spectral gradient of polarization difference and the brightness temperature at 36.5 GHz, was also proposed. The performances of the new algorithm to discriminate frozen and thawed soils are very similar to those obtained using the spectral gradient of brightness temperature (global precision around 90% and probability of detecting frozen soils between 70% and 85%). The performances are also slightly higher for the combinations including the lower AMSR-E frequencies. However, annual statistics for the spectral gradient of polarization difference are required to calculate the thresholds. The results obtained with AMSR-E data highlight the relevance of including SMOS L-band brightness temperatures for the calculation of brightness temperature and polarization difference spectral gradients. The qualitative analysis of the results obtained using RADARSAT-2 data shows that surface scattering dominates volume scattering for frozen soils, which can be explained by the rough fields in the study area, as compared to the signal's wavelength (C-band). Nevertheless, several polarimetric parameters indicate a slight increase of the volume scattering in frozen soils, which is theoretically expected. This was observed for the linear and circular depolarization ratios, the amplitude of the HHVV, RLLL and RLRR correlation coefficients, as well as for the pedestal height. Also, the entropy and [alpha overline]-angle of the Cloude-Pottier target decomposition increase slightly in frozen soils ; the same is true for the volume scattering component of the Freeman-Durden and Yamaguchi target decompositions, with an equivalent decrease of the surface scattering component. Despite these interesting observations, a quantitative analysis of the results is necessary in order to evaluate the usefulness of polarimetry regarding the detection of frozen soils. This would allow the validation of the behavior, possibly caused by soil freezing, of the mean value and the standard deviation of the HHVV phase difference and the standard deviation of the RLLL and RLRR phase differences.

The mapping of sea ice is an important task for understanding global climate and for safe shipping. Currently, sea ice maps are created by human analysts with the help of remote sensing imagery, including synthetic aperture radar (SAR) imagery. While the maps are generally correct, they can be somewhat subjective and do not have pixel-level resolution due to the time consuming nature of manual segmentation. Therefore, automated sea ice mapping algorithms such as the multivariate iterative region growing with semantics (MIRGS) sea ice image segmentation algorithm are needed. MIRGS was designed to work with one-channel single-polarization SAR imagery from the RADARSAT-1 satellite. The launch of RADARSAT-2 has made available two-channel dual-polarization SAR imagery for the purposes of sea ice mapping. Dual-polarization imagery provides more information for distinguishing ice types, and one of the channels is less sensitive to changes in the backscatter caused by the SAR incidence angle parameter. In the past, this change in backscatter due to the incidence angle was a key limitation that prevented automatic segmentation of full SAR scenes. This thesis investigates techniques to make use of the dual-polarization data in MIRGS. An evaluation of MIRGS with RADARSAT-2 data was performed and showed that some detail was lost and that the incidence angle caused errors in segmentation. Several data fusion schemes were investigated to determine if they can improve performance. Gradient generation methods designed to take advantage of dual-polarization data, feature space fusion using linear and non-linear transforms as well as image fusion methods based on wavelet combination rules were implemented and tested. Tuning of the MIRGS parameters was performed to find the best set of parameters for segmentation of dual-polarization data. Results show that the standard MIRGS algorithm with default parameters provides the highest accuracy, so no changes are necessary for dual-polarization data. A hierarchical segmentation scheme that segments the dual-polarization channels separately was implemented to overcome the incidence angle errors. The technique is effective but requires more user input than the standard MIRGS algorithm.

The mapping of sea ice is an important task for understanding global climate and for safe shipping. Currently, sea ice maps are created by human analysts with the help of remote sensing imagery, including synthetic aperture radar (SAR) imagery. While the maps are generally correct, they can be somewhat subjective and do not have pixel-level resolution due to the time consuming nature of manual segmentation. Therefore, automated sea ice mapping algorithms such as the multivariate iterative region growing with semantics (MIRGS) sea ice image segmentation algorithm are needed. MIRGS was designed to work with one-channel single-polarization SAR imagery from the RADARSAT-1 satellite. The launch of RADARSAT-2 has made available two-channel dual-polarization SAR imagery for the purposes of sea ice mapping. Dual-polarization imagery provides more information for distinguishing ice types, and one of the channels is less sensitive to changes in the backscatter caused by the SAR incidence angle parameter. In the past, this change in backscatter due to the incidence angle was a key limitation that prevented automatic segmentation of full SAR scenes. This thesis investigates techniques to make use of the dual-polarization data in MIRGS. An evaluation of MIRGS with RADARSAT-2 data was performed and showed that some detail was lost and that the incidence angle caused errors in segmentation. Several data fusion schemes were investigated to determine if they can improve performance. Gradient generation methods designed to take advantage of dual-polarization data, feature space fusion using linear and non-linear transforms as well as image fusion methods based on wavelet combination rules were implemented and tested. Tuning of the MIRGS parameters was performed to find the best set of parameters for segmentation of dual-polarization data. Results show that the standard MIRGS algorithm with default parameters provides the highest accuracy, so no changes are necessary for dual-polarization data. A hierarchical segmentation scheme that segments the dual-polarization channels separately was implemented to overcome the incidence angle errors. The technique is effective but requires more user input than the standard MIRGS algorithm.

This thesis is focused on the utilization of RADARSAT-2 polarimetric SAR data for mapping two surficial aspects of the Umiujalik Lake area, Nunavut, Canada: i) materials, such as bedrock, boulders, organic material, sand and gravel, thick and thin till; and ii) lineaments. To achieve these tasks, RADARSAT-2 polarimetric SAR images with three west-looking, increasing incidence angles (FQ1, FQ12, and FQ20, respectively) were used alone and in combination with LANDSAT-7 ETM+ and Digital Elevation Model (DEM) image data. The surficial materials mapping study tested: i) the effects of incidence angles on mapping accuracy; and ii) non-polarimetric and polarimetric classifiers. For non-polarimetric analysis, a Maximum Likelihood Classification (MLC) algorithm was applied to different combinations of RADARSAT-2, LANDSAT-7 ETM+, and DEM images, achieving a maximum overall classification accuracy of 85%. Polarimetric analyses first included computation of polarimetric signatures to understand the scattering mechanisms of the considered surficial materials, i.e., surface, volume, and multiple scatterings. It also tested three polarimetric classifiers: supervised Wishart (overall accuracy of 48.7% from FQ12 image), and unsupervised Freeman-Wishart, and Wishart-H/ /A. Three main conclusions were reached: i) high incidence angle greatly decreases classification accuracy for the HH polarized image when used alone, but incidence angle has little effect when the HV polarization is added; ii) combining images with three incidence angles (FQ1, FQ12, and FQ20) gives higher accuracy with the maximum likelihood classifier; and iii) the medium incidence angle image (FQ12) produces the best classification accuracy using polarimetric classifiers. In the second part of the study, surface lineaments were mapped using RADARSAT-2 SAR single-polarized images, RGB HH, HV, VV composites, polarimetric total power images, and LANDSAT-7 ETM+ principal component images. Polarization effect analysis showed that regardless of beam mode, more lineaments were identified on the HH image than on the HV image, and the maximum number of lineaments was identified on the multi-polarized RGB composite. Incidence angle effects results showed that regardless of polarization modes, the FQ12 image yielded more lineaments than the FQ1 or FQ20 images. The majority of lineaments are oriented in NW and NNW directions, which correspond to the ice flow direction during the last glaciation.

This project presents regional velocity measurements for the eastern portion of the St. Elias Mountains, including the entire glaciated area of Kluane National Park, derived from speckle tracking of Radarsat-2 imagery acquired in winter 2011 and 2012. This technique uses a cross-correlation approach to determine the displacement of the ‘speckle’ pattern of radar phase returns between two repeat-pass images. Further reconstruction of past velocities is performed on a selection of key glaciers using feature tracking of Landsat-5 imagery, allowing for the investigation of variability in glacier motion on interannual and decadal time scales. The results of the analysis showed that there is a strong velocity gradient across the region reflecting high accumulation rates on the Pacific-facing slope of the mountain range. These glaciers may have velocities an order of magnitude greater than glaciers of a similar size on the landward slope. Interannual variability was high, both in relation to surge events, of which a number were identified, and variation of other unknown controls on glacier motion. A long-term trend of velocity decrease was observed on the Kaskawulsh Glacier when comparing the results of this analysis to work carried out in the 1960s, the pattern of which is broadly congruent to measurements of surface elevation change over a similar period.

Thesis (MA)--Stellenbosch University, 2015. / ENGLISH ABSTRACT: Image classification has long been used in earth observation and is driven by the need for accurate maps to develop conceptual and predictive models of Earth system processes. Synthetic aperture radar (SAR) imagery is used ever more frequently in land cover classification due to its complementary nature with optical data. There is therefore a growing need for reliable, accurate methods for using SAR and optical data together in land use and land cover classifications. However, combining data sets inevitably increases data dimensionality and these large, complex data sets are difficult to handle. It is therefore important to assess the benefits and limitations of using multi-temporal, dual-sensor data for applications such as land cover classification. This thesis undertakes this assessment through four main experiments based on combined RADARSAT-2 and SPOT-5 imagery of the southern part of Reunion Island. In Experiment 1, the use of feature selection for dimensionality reduction was considered. The rankings of important features for both single-sensor and dual-sensor data were assessed for four dates spanning a 6-month period, which coincided with both the wet and dry season. The mean textural features produced from the optical bands were consistently ranked highly across all dates. In the two later dates (29 May and 9 August 2014), the SAR features were more prevalent, showing that SAR and optical data have complementary natures. SAR data can be used to separate classes when optical imagery is insufficient. Experiment 2 compared the accuracy of six supervised and machine learning classification algorithms to determine which performed best with this complex data set. The Random Forest classification algorithm produced the highest accuracies and was therefore used in Experiments 3 and 4. Experiment 3 assessed the benefits of using combined SAR-optical imagery over single-sensor imagery for land cover classifications on four separate dates. The fused imagery produced consistently higher overall accuracies. The 29 May 2014 fused data produced the best accuracy of 69.8%. The fused classifications had more consistent results over the four dates than the single-sensor imagery, which suffered lower accuracies, especially for imagery acquired later in the season. In Experiment 4, the use of multi-temporal, dual-sensor data for classification was evaluated. Feature selection was used to reduce the data set from 638 potential training features to 50, which produced the best accuracy of 74.1% in comparison to 71.9% using all of the features. This result validated the use of multi-temporal data over single-date data for land cover classifications. It also validated the use of feature selection to successfully inform data reduction without compromising the accuracy of the final product. Multi-temporal and dual-sensor data shows potential for mapping land cover in a tropical, mountainous region that would otherwise be challenging to map using single-sensor data. However, accuracies Stellenbosch University https://scholar.sun.ac.za iv generally remained lower than would allow for transferability and replication of the current methodology. Classification algorithm optimisation, supervised segmentation and improved training data should be considered to improve these results. / AFRIKAANSE OPSOMMING: Beeld-klassifikasie word al ‘n geruime tyd in aardwaarneming gebruik en word gedryf deur die behoefte aan akkurate kaarte om konseptuele en voorspellende modelle van aard-stelsel prosesse te ontwikkel. Sintetiese apertuur radar (SAR) beelde word ook meer dikwels in landdekking klassifikasie gebruik as gevolg van die aanvullende waarde daarvan met optiese data. Daar is dus 'n groeiende behoefte aan betroubare, akkurate metodes vir die gesamentlike gebruik van SAR en optiese data in landdekking klassifikasies. Die kombinasie van datastelle bring egter ‘n onvermydelike verhoging in data dimensionaliteit mee, en hierdie groot, komplekse datastelle is moeilik om te hanteer. Dus is dit belangrik om die voordele en beperkings van die gebruik van multi-temporale, dubbel-sensor data vir toepassings soos landdekking-klassifikasie te evalueer. Die waarde van gekombineerde (versmelte) RADARSAT-2 en SPOT-5 beelde word in hierdie tesis deur middel van vier eksperimente geevalueer. In Eksperiment 1 is die gebruik van kenmerk seleksie vir dimensionaliteit-vermindering toegepas. Die ranglys van belangrike kenmerke vir beide enkel-sensor en 'n dubbel-sensor data is beoordeel vir vier datums wat oor 'n tydperk van 6 maande strek. Die gemiddelde tekstuur kenmerke uit die optiese lae is konsekwent hoog oor alle datums geplaas. In die twee later datums (29 Mei en 9 Augustus 2014) was die SAR kenmerke meer algemeen, wat dui op die aanvullende aard van SAR en optiese data. SAR data dus gebruik kan word om klasse te onderskei wanneer optiese beelde onvoldoende daarvoor is. Eksperiment 2 het die akkuraatheid van ses gerigte en masjien-leer klassifikasie algoritmes vergelyk om te bepaal watter die beste met hierdie komplekse datastel presteer. Die random gorest klassifikasie algoritme het die hoogste akkuraatheid bereik en is dus in Eksperimente 3 en 4 gebruik. Eksperiment 3 het die voordele van gekombineerde SAR-optiese beelde oor enkel-sensor beelde vir landdekking klassifikasies op vier afsonderlike datums beoordeel. Die versmelte beelde het konsekwent hoër algehele akkuraathede as enkel-sensor beelde gelewer. Die 29 Mei 2014 data het die hoogste akkuraatheid van 69,8% bereik. Die versmelte klassifikasies het ook meer konsekwente resultate oor die vier datums gelewer en die enkel-sensor beelde het tot laer akkuraathede gelei, veral vir die later datums. In Eksperiment 4 is die gebruik van multi-temporale, dubbel-sensor data vir klassifikasie ge-evalueer. Kenmerkseleksie is gebruik om die data stel van 638 potensiële kenmerke na 50 te verminder, wat die beste akkuraatheid van 74,1% gelewer het. Hierdie resultaat bevestig die belangrikheid van multi-temporale data vir grond dekking klassifikasies. Dit bekragtig ook die gebruik van kenmerkseleksie om data vermindering suksesvol te rig sonder om die akkuraatheid van die finale produk te belemmer. Stellenbosch University https://scholar.sun.ac.za vi Multi-temporale en dubbel-sensor data toon potensiaal vir die kartering van landdekking in 'n tropiese, bergagtige streek wat andersins uitdagend sou wees om te karteer met behulp van enkel-sensor data. Oor die algemeen het akkuraathede egter te laag gebly om vir oordraagbaarheid en herhaling van die huidige metode toe te laat. Klassifikasie algoritme optimalisering, gerigte segmentering en verbeterde opleiding data moet oorweeg word om hierdie resultate te verbeter.