Object-based Random Forest Classification Research Articles

Comparing Pixel- and Object-Based Approaches for Classifying Multispectral Drone Imagery of a Salt Marsh Restoration and Reference Site

Monitoring salt marshes with remote sensing is necessary to evaluate their state and restoration. Determining appropriate techniques for this can be overwhelming. Our study provides insight into whether a pixel- or object-based Random Forest classification approach is best for mapping vegetation in north temperate salt marshes. We used input variables from drone images (raw reflectances, vegetation indices, and textural features) acquired in June, July, and August 2021 of a salt marsh restoration and reference site in Aulac, New Brunswick, Canada. We also investigated the importance of input variables and whether using landcover classes representing areas of change was a practical way to evaluate variation in the monthly images. Our results indicated that (1) the classifiers achieved overall validation accuracies of 91.1–95.2%; (2) pixel-based classifiers outperformed object-based classifiers by 1.3–2.0%; (3) input variables extracted from the August images were more important than those extracted from the June and July images; (4) certain raw reflectances, vegetation indices, and textural features were among the most important variables; and (5) classes that changed temporally were mapped with user’s and producer’s validation accuracies of 86.7–100.0%. Knowledge gained during this study will inform assessments of salt marsh restoration trajectories spanning multiple years.

Accuracy Improvements to Pixel-Based and Object-Based LULC Classification with Auxiliary Datasets from Google Earth Engine

The monitoring and assessment of land use/land cover (LULC) change over large areas are significantly important in numerous research areas, such as natural resource protection, sustainable development, and climate change. However, accurately extracting LULC only using the spectral features of satellite images is difficult owing to landscape heterogeneities over large areas. To improve the accuracy of LULC classification, numerous studies have introduced other auxiliary features to the classification model. The Google Earth Engine (GEE) not only provides powerful computing capabilities, but also provides a large amount of remote sensing data and various auxiliary datasets. However, the different effects of various auxiliary datasets in the GEE on the improvement of the LULC classification accuracy need to be elucidated along with methods that can optimize combinations of auxiliary datasets for pixel- and object-based classification. Herein, we comprehensively analyze the performance of different auxiliary features in improving the accuracy of pixel- and object-based LULC classification models with medium resolution. We select the Yangtze River Delta in China as the study area and Landsat-8 OLI data as the main dataset. Six types of features, including spectral features, remote sensing multi-indices, topographic features, soil features, distance to the water source, and phenological features, are derived from auxiliary open-source datasets in GEE. We then examine the effect of auxiliary datasets on the improvement of the accuracy of seven pixels-based and seven object-based random forest classification models. The results show that regardless of the types of auxiliary features, the overall accuracy of the classification can be improved. The results further show that the object-based classification achieves higher overall accuracy compared to that obtained by the pixel-based classification. The best overall accuracy from the pixel-based (object-based) classification model is 94.20% (96.01%). The topographic features play the most important role in improving the overall accuracy of classification in the pixel- and object-based models comprising all features. Although a higher accuracy is achieved when the object-based method is used with only spectral data, small objects on the ground cannot be monitored. However, combined with many types of auxiliary features, the object-based method can identify small objects while also achieving greater accuracy. Thus, when applying object-based classification models to mid-resolution remote sensing images, different types of auxiliary features are required. Our research results improve the accuracy of LULC classification in the Yangtze River Delta and further provide a benchmark for other regions with large landscape heterogeneity.

Monitoring Invasion Process of Spartina alterniflora by Seasonal Sentinel-2 Imagery and an Object-Based Random Forest Classification

In the late 1990s, the exotic plant Spartina alterniflora (S. alterniflora), was introduced to the Zhangjiang Estuary of China for tidal zone reclamation and protection. However, it invaded rapidly and has caused serious ecological problems. Accurate information on the seasonal invasion of S. alterniflora is vital to understand invasion pattern and mechanism, especially at a high temporal resolution. This study aimed to explore the S. alterniflora invasion process at a seasonal scale from 2016 to 2018. However, due to the uncertainties caused by periodic inundation of local tides, accurately monitoring the spatial extent of S. alterniflora is challenging. Thus, to achieve the goal and address the challenge, we firstly built a high-quality seasonal Sentinel-2 image collection by developing a new submerged S. alterniflora index (SAI) to reduce the errors caused by high tide fluctuations. Then, an object-based random forest (RF) classification method was applied to the image collection. Finally, seasonal extents of S. alterniflora were captured. Results showed that (1) the red edge bands (bands 5, 6, and 7) of Sentinel-2 imagery played critical roles in delineating submerged S. alterniflora; (2) during March 2016 to November 2018, the extent of S. alterniflora increased from 151.7 to 270.3 ha, with an annual invasion rate of 39.5 ha; (3) S. alterniflora invaded with a rate of 31.5 ha/season during growing season and 12.1 ha/season during dormant season. To our knowledge, this is the first study monitoring S. alterniflora invasion process at a seasonal scale during continuous years, discovering that S. alterniflora also expands during dormant seasons. This discovery is of great significance for understanding the invasion pattern and mechanism of S. alterniflora and will facilitate coastal biodiversity conservation efforts.

Object-based random forest classification for informal settlements identification in the Middle East: Jeddah a case study

ABSTRACTThe identification of informal settlements in urban areas is an important step in developing and implementing pro-poor urban policies. Understanding when, where and who lives inside informal settlements is critical in efforts to improve their resilience. This study aims to analyse the capability of machine-learning (ML) methods to map informal settlement areas in Jeddah, Saudi Arabia, using very-high-resolution (VHR) imagery and terrain data. Fourteen indicators of settlement characteristics were derived and mapped using an object-based ML approach and VHR imagery. These indicators were categorized according to three different spatial levels: environ, settlement and object. The most useful indicators for prediction were found to be density and texture measures, (with random forest (RF) relative importance measures of over 25% and 23% respectively). The success of this approach was evaluated using a small, fully independent validation dataset. Informal areas were mapped with an overall accuracy of 91%. Object-based ML as a processing chain approach performed better (8%) than object-based image analysis alone due to its ability to encompass all available geospatial levels.

Big Data for a Big Country: The First Generation of Canadian Wetland Inventory Map at a Spatial Resolution of 10-m Using Sentinel-1 and Sentinel-2 Data on the Google Earth Engine Cloud Computing Platform

Detailed information on the spatial distribution of wetlands is crucial for sustainable management and resource assessment. Furthermore, regularly updated wetland inventories are of particular importance given that wetlands comprise a dynamic, rather than permanent, land condition. Accordingly, satellite-derived wetland maps are greatly beneficial, as they capture a synoptic and multi-temporal view of landscapes. Leveraging state-of-the-art remote sensing data and tools, this study produces a high-resolution 10-m wetland inventory map of Canada, covering an approximate area of one billion hectares, using multi-year, multi-source (Sentinel-1 and Sentinel-2) Earth Observation (EO) data on the Google Earth Engine™ cloud computing platform. The whole country is mapped using a large volume of reference samples using an object-based random forest classification scheme with an overall accuracy approaching 80% and individual accuracies varying from 74% to 84% in different provinces. This nationwide wetland inventory map illustrates that 19% of Canada’s land area is covered by wetlands, most of which are peatlands dominate in the northern ecozones. Importantly, the resulting ever-demanding wetland inventory map of Canada provides unprecedented details on the extent, status, and spatial distribution of wetlands and thus, is useful for many stakeholders, including federal and provincial governments, municipalities, NGOs, and environmental consultants.

Land Cover Classification Based on Cloud Computing Platform

Land cover is a critical factor for climate change and hydrological models. The extraction of land cover data from remote sensing images has been carried out by specialized commercial software. However, the limitations of computer hardware and algorithms of the commercial software are costly and make it take a lot of time, patience, and skills to do the classification. The cloud computing platform Google Earth Engine brought a breakthrough in 2010 for analyzing and processing spatial data. This study applied Object-based Random Forest classification in the Google Earth Engine platform to produce land cover data in 2010 in the Vu Gia - Thu Bon river basin. The classification results showed 7 categories of land cover consisting of plantation forest, natural forest, paddy field, urban residence, rural residence, bare land, and water surface, with an overall accuracy of 73.9% and kappa of 0.70.

The First Wetland Inventory Map of Newfoundland at a Spatial Resolution of 10 m Using Sentinel-1 and Sentinel-2 Data on the Google Earth Engine Cloud Computing Platform

Wetlands are one of the most important ecosystems that provide a desirable habitat for a great variety of flora and fauna. Wetland mapping and modeling using Earth Observation (EO) data are essential for natural resource management at both regional and national levels. However, accurate wetland mapping is challenging, especially on a large scale, given their heterogeneous and fragmented landscape, as well as the spectral similarity of differing wetland classes. Currently, precise, consistent, and comprehensive wetland inventories on a national- or provincial-scale are lacking globally, with most studies focused on the generation of local-scale maps from limited remote sensing data. Leveraging the Google Earth Engine (GEE) computational power and the availability of high spatial resolution remote sensing data collected by Copernicus Sentinels, this study introduces the first detailed, provincial-scale wetland inventory map of one of the richest Canadian provinces in terms of wetland extent. In particular, multi-year summer Synthetic Aperture Radar (SAR) Sentinel-1 and optical Sentinel-2 data composites were used to identify the spatial distribution of five wetland and three non-wetland classes on the Island of Newfoundland, covering an approximate area of 106,000 km2. The classification results were evaluated using both pixel-based and object-based random forest (RF) classifications implemented on the GEE platform. The results revealed the superiority of the object-based approach relative to the pixel-based classification for wetland mapping. Although the classification using multi-year optical data was more accurate compared to that of SAR, the inclusion of both types of data significantly improved the classification accuracies of wetland classes. In particular, an overall accuracy of 88.37% and a Kappa coefficient of 0.85 were achieved with the multi-year summer SAR/optical composite using an object-based RF classification, wherein all wetland and non-wetland classes were correctly identified with accuracies beyond 70% and 90%, respectively. The results suggest a paradigm-shift from standard static products and approaches toward generating more dynamic, on-demand, large-scale wetland coverage maps through advanced cloud computing resources that simplify access to and processing of the “Geo Big Data.” In addition, the resulting ever-demanding inventory map of Newfoundland is of great interest to and can be used by many stakeholders, including federal and provincial governments, municipalities, NGOs, and environmental consultants to name a few.

The effect of seasonal variation on automated land cover mapping from multispectral airborne laser scanning data

ABSTRACTMultispectral airborne laser scanning (MS-ALS) sensors are a new promising source of data for automated mapping methods. Finding an optimal time for data acquisition is important in all mapping applications based on remotely sensed datasets. In this study, three MS-ALS datasets acquired at different times of the growing season were compared for automated land cover mapping and road detection in a suburban area. In addition, changes in the intensity were studied. An object-based random forest classification was carried out using reference points. The overall accuracy of the land cover classification was 93.9% (May dataset), 96.4% (June) and 95.9% (August). The use of the May dataset acquired under leafless conditions resulted in more complete roads than the other datasets acquired when trees were in leaf. It was concluded that all datasets used in the study are applicable for suburban land cover mapping, however small differences in accuracies between land cover classes exist.

Monitoring surface changes in discontinuous permafrost terrain using small baseline SAR interferometry, object-based classification, and geological features: a case study from Mayo, Yukon Territory, Canada

Permafrost-induced deformation of ground features is threating infrastructure in northern communities. An understanding of permafrost distribution is therefore critical for sustainable adaptation planning and infrastructure maintenance. Considering the large area underlain by permafrost in the Yukon Territory, there is a need for baseline information to characterize the permafrost in this region. In this study, the Differential Interferometric Synthetic Aperture Radar (DInSAR) technique was used to identify areas of ground movement likely caused by changes in permafrost. The DInSAR technique was applied to a series of repeat-pass C-band RADARSAT-2 observations collected in 2015 over the Village of Mayo, in central Yukon Territory, Canada. The conventional DInSAR technique demonstrated that ground deformation could be detected in this area, but the resulting deformation maps contained errors due to a loss of coherence from changes in vegetation and atmospheric phase delay. To address these limitations, the Small BAseline Subset (SBAS) InSAR technique was applied to reduce phase error, thus improving the deformation maps. To understand the relationship between the deformation maps and land cover types, an object-based Random Forest classification was developed to classify the study area into different land cover types. Integration of the InSAR results and the classification map revealed that the built-up class (e.g., airport) was affected by subsidence on the order of −2 to −4 cm. The spatial extent of the surface displacement map obtained using the SBAS InSAR technique was then correlated with the surficial geology map. This revealed that much of the main infrastructure in the Village of Mayo is underlain by interbedded glaciofluvial and glaciolacustrine sediments, the latter of which caused the most damage to human made structures. This study provides a method for permafrost monitoring that builds upon the synergistic use of the SBAS InSAR technique, object-based image analysis, and surficial geology data.

Individual Tree Crown Segmentation and Classification of 13 Tree Species Using Airborne Hyperspectral Data

Knowledge of the distribution of tree species within a forest is key for multiple economic and ecological applications. This information is traditionally acquired through time-consuming and thereby expensive field work. Our study evaluates the suitability of a visible to near-infrared (VNIR) hyperspectral dataset with a spatial resolution of 0.4 m for the classification of 13 tree species (8 broadleaf, 5 coniferous) on an individual tree crown level in the UNESCO Biosphere Reserve ‘Wienerwald’, a temperate Austrian forest. The study also assesses the automation potential for the delineation of tree crowns using a mean shift segmentation algorithm in order to permit model application over large areas. Object-based Random Forest classification was carried out on variables that were derived from 699 manually delineated as well as automatically segmented reference trees. The models were trained separately for two strata: small and/or conifer stands and high broadleaf forests. The two strata were delineated beforehand using CHM-based tree height and NDVI. The predictor variables encompassed spectral reflectance, vegetation indices, textural metrics and principal components. After feature selection, the overall classification accuracy (OA) of the classification based on manual delineations of the 13 tree species was 91.7% (Cohen’s kappa (κ) = 0.909). The highest user’s and producer’s accuracies were most frequently obtained for Weymouth pine and Scots Pine, while European ash was most often associated with the lowest accuracies. The classification that was based on mean shift segmentation yielded similarly good results (OA = 89.4% κ = 0.883). Based on the automatically segmented trees, the Random Forest models were also applied to the whole study site (1050 ha). The resulting tree map of the study area confirmed a high abundance of European beech (58%) with smaller amounts of oak (6%) and Scots pine (5%). We conclude that highly accurate tree species classifications can be obtained from hyperspectral data covering the visible and near-infrared parts of the electromagnetic spectrum. Our results also indicate a high automation potential of the method, as the results from the automatically segmented tree crowns were similar to those that were obtained for the manually delineated tree crowns.

Multitemporal Classification of River Floodplain Vegetation Using Time Series of UAV Images

The functions of river floodplains often conflict spatially, for example, water conveyance during peak discharge and diverse riparian ecology. Such functions are often associated with floodplain vegetation. Frequent monitoring of floodplain land cover is necessary to capture the dynamics of this vegetation. However, low classification accuracies are found with existing methods, especially for relatively similar vegetation types, such as grassland and herbaceous vegetation. Unmanned aerial vehicle (UAV) imagery has great potential to improve the classification of these vegetation types owing to its high spatial resolution and flexibility in image acquisition timing. This study aimed to evaluate the increase in classification accuracy obtained using multitemporal UAV images versus single time step data on floodplain land cover classification and to assess the effect of varying the number and timing of imagery acquisition moments. We obtained a dataset of multitemporal UAV imagery and field reference observations and applied object-based Random Forest classification (RF) to data of different time step combinations. High overall accuracies (OA) exceeding 90% were found for the RF of floodplain land cover, with six vegetation classes and four non-vegetation classes. Using two or more time steps compared with a single time step increased the OA from 96.9% to 99.3%. The user’s accuracies of the classes with large similarity, such as natural grassland and herbaceous vegetation, also exceeded 90%. The combination of imagery from June and September resulted in the highest OA (98%) for two time steps. Our method is a practical and highly accurate solution for monitoring areas of a few square kilometres. For large-scale monitoring of floodplains, the same method can be used, but with data from airborne platforms covering larger extents.

Multi-temporal, multi-frequency, and multi-polarization coherence and SAR backscatter analysis of wetlands

Despite recent research into the Interferometric Synthetic Aperture Radar (InSAR) technique for wetland mapping worldwide, its capability has not yet been thoroughly investigated for Canadian wetland ecosystems. Accordingly, this study statistically analysed interferometric coherence and SAR backscatter variation in a study area located on the Avalon Peninsula, Newfoundland and Labrador, Canada, consisting of various wetland classes, including bog, fen, marsh, swamp, and shallow-water. Specifically, multi-temporal L-band ALOS PALSAR-1, C-band RADARSAT-2, and X-band TerraSAR-X data were used to investigate the effect of SAR frequency and polarization, as well as temporal baselines on the coherence degree in the various wetland classes. SAR backscatter and coherence maps were also used as input features into an object-based Random Forest classification scheme to examine the contribution of these features to the overall classification accuracy. Our findings suggested that the temporal baseline was the most influential factor for coherence maintenance in herbaceous wetlands, especially for shorter wavelengths. In general, coherence was the highest in L-band and intermediate/low for both X- and C-band, depending on the wetland classes and temporal baseline. The Wilcoxon rank sum test at the 5% significance level found significant difference (P-value < 0.05) between the mean values of HH/HV coherence at the peak of growing season. The test also suggested that L-band intensity and X-band coherence observations were advantageous to discriminate complex wetland classes. Notably, an overall classification accuracy of 74.33% was attained for land cover classification by synergistic use of both SAR backscatter and interferometric coherence. Thus, the results of this study confirmed the potential of incorporating SAR and InSAR features for mapping Canadian wetlands and those elsewhere in the world with similar ecological characteristics.

Object-based random forest classification of Landsat ETM+ and WorldView-2 satellite imagery for mapping lowland native grassland communities in Tasmania, Australia

This paper presents a random forest classification approach for identifying and mapping three types of lowland native grassland communities found in the Tasmanian Midlands region. Due to the high conservation priority assigned to these communities, there has been an increasing need to identify appropriate datasets that can be used to derive accurate and frequently updateable maps of community extent. Therefore, this paper proposes a method employing repeat classification and statistical significance testing as a means of identifying the most appropriate dataset for mapping these communities. Two datasets were acquired and analysed; a Landsat ETM+ scene, and a WorldView-2 scene, both from 2010. Training and validation data were randomly subset using a k-fold (k=50) approach from a pre-existing field dataset. Poa labillardierei, Themeda triandra and lowland native grassland complex communities were identified in addition to dry woodland and agriculture. For each subset of randomly allocated points, a random forest model was trained based on each dataset, and then used to classify the corresponding imagery. Validation was performed using the reciprocal points from the independent subset that had not been used to train the model. Final training and classification accuracies were reported as per class means for each satellite dataset. Analysis of Variance (ANOVA) was undertaken to determine whether classification accuracy differed between the two datasets, as well as between classifications. Results showed mean class accuracies between 54% and 87%. Class accuracy only differed significantly between datasets for the dry woodland and Themeda grassland classes, with the WorldView-2 dataset showing higher mean classification accuracies. The results of this study indicate that remote sensing is a viable method for the identification of lowland native grassland communities in the Tasmanian Midlands, and that repeat classification and statistical significant testing can be used to identify optimal datasets for vegetation community mapping.

An Assessment of Simulated Compact Polarimetric SAR Data for Wetland Classification Using Random Forest Algorithm

ABSTRACTSynthetic aperture radar (SAR) compact polarimetry (CP) systems are of great interest for large area monitoring because of their ability to acquire data in a wider swath compared to full polarimetry (FP) systems and a significant improvement in information content compared to single or dual polarimetry (DP) sensors. In this study, we compared the potential of DP, FP, and CP SAR data for wetland classification in a case study located in Newfoundland, Canada. The DP and CP data were simulated using full polarimetric RADARSAT-2 data. We compared the classification results for different input features using an object-based random forest classification. The results demonstrated the superiority of FP imagery relative to both DP and CP data. However, CP indicated significant improvements in classification accuracy compared to DP data. An overall classification accuracy of approximately 76% and 84% was achieved with the inclusion of all polarimetric features extracted from CP and FP data, respectively. In summary, although full polarimetric SAR data provide the best classification accuracy, the results demonstrate the potential of RADARSAT Constellation Mission for mapping wetlands in a large landscape.