Search Results (7)

Land-use and land-cover (LULC) change detection (CD) is a pivotal research area in remote sensing applications, posing a significant challenge due to variations in illumination, radiation, and image noise between bi-temporal images. Currently, deep learning solutions, particularly convolutional neural networks (CNNs), represent the state of the art (SOTA) for CD. However, CNN-based models require substantial amounts of annotated data, which can be both expensive and time-consuming. Conversely, acquiring a large volume of unannotated images is relatively easy. Recently, self-supervised contrastive learning has emerged as a promising method for learning from unannotated images, thereby reducing the need for annotation. However, most existing methods employ random values or ImageNet pre-trained models to initialize their encoders and lack prior knowledge tailored to the demands of CD tasks, thus constraining the performance of CD models. To address these challenges, we introduce a novel feature-differencing-based framework called Barlow Twins for self-supervised pre-training and fine-tuning in CD (BTCD). The proposed approach employs absolute feature differences to directly learn unique representations associated with regions that have changed from unlabeled bi-temporal remote sensing images in a self-supervised manner. Moreover, we introduce invariant prediction loss and change consistency regularization loss to enhance image alignment between bi-temporal images in both the decision and feature space during network training, thereby mitigating the impact of variation in radiation conditions, noise, and imaging viewpoints. We select the improved UNet++ model for fine-tuning self-supervised pre-training models and conduct experiments using two publicly available LULC CD datasets. The experimental results demonstrate that our proposed approach outperforms existing SOTA methods in terms of competitive quantitative and qualitative performance metrics. Full article

Historical land cover (LC) maps are an essential instrument for studying long-term spatio-temporal changes of the landscape. However, manual labelling on low-quality monochromatic historical orthophotos for semantic segmentation (pixel-level classification) is particularly challenging and time consuming. Therefore, this paper proposes a methodology for the automated extraction of very-high-resolution (VHR) multi-class LC maps from historical orthophotos under the absence of target-specific ground truth annotations. The methodology builds on recent evolutions in deep learning, leveraging domain adaptation and transfer learning. First, an unpaired image-to-image (I2I) translation between a source domain (recent RGB image of high quality, annotations available) and the target domain (historical monochromatic image of low quality, no annotations available) is learned using a conditional generative adversarial network (GAN). Second, a state-of-the-art fully convolutional network (FCN) for semantic segmentation is pre-trained on a large annotated RGB earth observation (EO) dataset that is converted to the target domain using the I2I function. Third, the FCN is fine-tuned using self-annotated data on a recent RGB orthophoto of the study area under consideration, after conversion using again the I2I function. The methodology is tested on a new custom dataset: the ‘Sagalassos historical land cover dataset’, which consists of three historical monochromatic orthophotos (1971, 1981, 1992) and one recent RGB orthophoto (2015) of VHR (0.3–0.84 m GSD) all capturing the same greater area around Sagalassos archaeological site (Turkey), and corresponding manually created annotations (2.7 km² per orthophoto) distinguishing 14 different LC classes. Furthermore, a comprehensive overview of open-source annotated EO datasets for multiclass semantic segmentation is provided, based on which an appropriate pretraining dataset can be selected. Results indicate that the proposed methodology is effective, increasing the mean intersection over union by 27.2% when using domain adaptation, and by 13.0% when using domain pretraining, and that transferring weights from a model pretrained on a dataset closer to the target domain is preferred. Full article

Meteorological and climate prediction models at the urban scale increasingly require more accurate and high-resolution data. The Local Climate Zone (LCZ) system is an initiative to standardize a classification scheme of the urban landscape, based mainly on the properties of surface structure (e.g., building, tree height, density) and surface cover (pervious vs. impervious). This approach is especially useful for studying the influence of urban morphology and fabric on the surface urban heat island (SUHI) effect and to evaluate how changes in land use and structures affect thermal regulation in the city. This article will demonstrate three different methodologies of creating LCZs: first, the World Urban Database and Access Portal Tools (WUDAPT); second, using Copernicus Urban Atlas (UA) data via a geographic information system (GIS) client directly; and third via Google Earth Engine (GEE) using Oslo, Norway as the case study. The WUDAPT and GEE methods incorporate a machine learning (random forest) procedure using Landsat 8 imagery, and offer the most precision while requiring the most time and familiarity with GIS usage and satellite imagery processing. The WUDAPT method is performed principally using multiple GIS clients and image processing tools. The GEE method is somewhat quicker to perform, with work performed entirely on Google’s sites. The UA or GIS method is performed solely via a GIS client and is a conversion of pre-existing vector data to LCZ classes via scripting. This is the quickest method of the three; however, the reclassification of the vector data determines the accuracy of the LCZs produced. Finally, as an illustration of a practical use of LCZs and to further compare the results of the three methods, we map the distribution of the temperature according to the LCZs of each method, correlating to the land surface temperature (LST) from a Landsat 8 image pertaining to a heat wave episode that occurred in Oslo in 2018. These results show, in addition to a clear LCZ-LST correspondence, that the three methods produce accurate and similar results and are all viable options. Full article

Achieving urban water security requires sustaining the trade-offs between the exploitation of water/environmental resources and ecosystem services. This achievement not only reduces the pollution and contamination in the environment, level of water stress, but also secures good ambient water quality and future for people’s well-being and livelihoods. Changes in land use and land cover and growth of impervious structures can immediately generate severe ecological and social issues and increase the level of natural or manmade risks, affecting the condition of ecosystem services within and in the vicinity of an urban region. As a result of these transformations and further exploitation, due to the growing anthropogenic pressure, surface water and groundwater quality can be deteriorated compared to ambient water quality standards (for both chemical and biological pollutants). Based on land use and land cover (LULC) data retrieved from remote sensing interpretation, we computed the changes of the ecosystem service values (ESV) associated with the LULC dynamics, water quality and, finally, urban water security during the pre- and post-monsoon periods of 2009, 2014 and 2019 in Kolkata, an Asian deltaic megacity, and its peri-urban wetlands named East Kolkata Wetlands (EKW). The area under wetlands reduced comprehensively in 2009–2019 due to the conversion of wetlands into various other classes such as urban settlement, etc. The quality of surface water bodies (such as rivers, lakes, canals and inland wetlands) deteriorated. The groundwater quality is still under control, but the presence of arsenic, manganese and other metals are a clear indication of urban expansion and related activities in the area. As a result, there was a change in the ESV during this timeframe. In the pre-monsoon period, there was an increase in total ESV from US$53.14 million in 2009 to US$53.36 million and US$59.01 million in 2014 and 2019, respectively. In the post-monsoon period, the ESV decreased from US$67.42 million in 2009 to US$64.13 and US$61.89 million in 2014 and 2019, respectively. These changes can be attributed to the peri-urban wetlands and the benefits or services arising out of them that contribute more than 50% of the total ESV. This study found that the area under wetlands has reduced comprehensively in the past 10 years due to the conversion of wetlands for various other uses such as urban expansion of the Kolkata City, but still, this peri-urban wetland supports the urban water security by providing sufficient ecosystem services. In conclusion, the transformation in extent of the water-related ecosystem is a crucial indicator of urban water security, which also measures the quantity of water contained in various water-related ecosystems. Quantitative analysis of the LULC change, hence, is important for studying the corresponding impact on the ecosystem service value (ESV) and water quality that helps in decision-making in securing urban water future and ecosystem conservation. Full article

The expansion of hydroelectric dams that is planned, and under construction, in the Amazon basin is a proposal to generate “clean” energy, with the purposes of meeting the regional energy demand, and the insertion of Brazil into the international economic market. However, this type of megaproject can change the dynamics of natural ecosystems. In the present article, the spatiotemporal patterns of deforestation according to distance from the reservoir in the vicinity of the lake of Tucuruí, and within a radius of 30 km from it, are analyzed. A linear spectral mixture model of segmented Landsat-thematic mapper (TM), enhanced thematic mapper plus (ETM+), and operational land imager (OLI) images, and proximity analysis were used for the mapping of the land-cover classes in the vicinity of the artificial lake of Tucuruí. Likewise, landscape metrics were determined with the purpose of quantifying the reduction of primary forest, as a mechanism of loss of ecosystem services in the region. These methods were also used for the evaluation of the influence of the distance from the reservoir on the expansion of anthropogenic activities. This methodology was used for the scenarios of pre-inauguration, completion of phase I, beginning of construction phase II, full completion of the Tucuruí hydroelectric project, and the current scenario of the region. The results showed that the highest deforestation rate occurred in the first period of the analysis, due to the areas submerged by the reservoir and due to the anthropogenic disturbances, such as timber extraction, road construction, and the conversion of forests into large areas of agribusiness. Full article

The Lake Victoria Basin (LVB) is a significant resource for five states within East Africa, which faces major land use land cover changes that threaten ecosystem integrity and ecosystem services derived from the basin’s resources. To assess land use land cover changes between 1985 and 2014, and subsequently determine the trends and drivers of these changes, we used a series of Landsat images and field data obtained from the LVB. Landsat image pre-processing and band combinations were done in ENVI 5.1. A supervised classification was applied on 118 Landsat scenes using the maximum likelihood classifier in ENVI 5.1. The overall accuracy of classified images was computed for the 2014 images using 124 reference data points collected through stratified random sampling. Computations of area under various land cover classes were calculated between the 1985 and 2014 images. We also correlated the area from natural vegetation classes to farmlands and settlements (urban areas) to explore relationships between land use land cover conversions among these classes. Based on our land cover classifications, we obtained overall accuracy of 71% and a moderate Kappa statistic of 0.56. Our results indicate that the LVB has undergone drastic changes in land use land cover, mainly driven by human activities that led to the conversion of forests, woodlands, grasslands, and wetlands to either farmlands or settlements. We conclude that information from this work is useful not only for basin-scale assessments and monitoring of land cover changes but also for targeting, prioritizing, and monitoring of small scale, community led efforts to restore degraded and fragmented areas in the basin. Such efforts could mitigate the loss of ecosystem services previously derived from large contiguous land covers which are no longer tenable to restore. We recommend adoption of a basin scale, operational, Earth observation-based, land use change monitoring framework. Such a framework can facilitate rapid and frequent assessments of gains and losses in specific land cover classes and thus focus strategic interventions in areas experiencing major losses, through mitigation and compensatory approaches. Full article

Knowing the stand age of rubber tree (Hevea brasiliensis) plantations is vitally important for best management practices, estimations of rubber latex yields, and carbon cycle studies (e.g., biomass, carbon pools, and fluxes). However, the stand age (as estimated from the establishment year of rubber plantation) is not available across large regions. In this study, we analyzed Landsat time series images from 1987–2015 and developed algorithms to identify (1) the establishment year of rubber plantations; and (2) the pre-conversion land cover types, such as old rubber plantations, evergreen forests, and cropland. Exposed soil during plantation establishment and linear increases in canopy closure during non-production periods (rubber seedling to mature plantation) were used to identify the establishment year of rubber plantations. Based on the rubber plantation map for 2015 (overall accuracy = 97%), and 1981 Landsat images since 1987, we mapped the establishment year of rubber plantations on Hainan Island (R² = 0.85/0.99, and RMSE = 2.34/0.54 years at pixel/plantation scale). The results show that: (1) significant conversion of croplands and old rubber plantations to new rubber plantations has occurred substantially in the northwest and northern regions of Hainan Island since 2000, while old rubber plantations were mainly distributed in the southeastern inland strip; (2) the pattern of rubber plantation expansion since 1987 consisted of fragmented plantations from smallholders, and there was no tendency to expand towards a higher altitude and steep slope regions; (3) the largest land source for new rubber plantations since 1988 was old rubber plantations (1.26 × 10⁵ ha), followed by cropland (0.95 × 10⁵ ha), and evergreen forests (0.68 × 10⁵ ha). The resultant algorithms and maps of establishment year and pre-conversion land cover types are likely to be useful in plantation management, and ecological assessments of rubber plantation expansion in China. Full article