Detailed Land Cover Map Research Articles

Unsupervised Domain Adaptation with Contrastive Learning-Based Discriminative Feature Augmentation for RS Image Classification

High- and very high-resolution (HR, VHR) remote sensing (RS) images can provide comprehensive and intricate spatial information for land cover classification, which is particularly crucial when analyzing complex built-up environments. However, the application of HR and VHR images to large-scale and detailed land cover mapping is always constrained by the intricacy of land cover classification models, the exorbitant cost of collecting training samples, and geographical changes or acquisition conditions. To overcome this limitation, we propose an unsupervised domain adaptation (UDA) with contrastive learning-based discriminative feature augmentation (CLDFA) for RS image classification. In detail, our method first utilizes contrastive learning (CL) through a memory bank in order to memorize sample features and improve model performance, where the approach employs an end-to-end Siamese network and incorporates dynamic pseudo-label assignment and class-balancing strategies for adaptive domain joint learning. By transferring classification models trained on a source domain (SD) to an unlabeled target domain (TD), our proposed UDA method enables large-scale land cover mapping. We conducted experiments using a massive five billion-pixels dataset as the SD and tested the HR and VHR RS images of five typical Chinese cities as the TD and applied the method on the completely unlabeled world view 3 (WV3) image of Urumqi city. The experimental results demonstrate that our method excels in large-scale HR and VHR RS image classification tasks, highlighting the advantages of semantic segmentation based on end-to-end deep convolutional neural networks (DCNNs).

MAPPING BEE-KEEPING FOREST PLANTS FROM MEDIUM SPATIAL RESOLUTION MULTISPECTRAL SATELLITE DATA

Abstract. Apiculture is one of the main branches of agriculture and crucial to rural development since it provides farmers with unique products like honey, wax, pollen, royal jelly, propolis and bee venom. Due to pollination services provided by honeybees, environmental sustainability and diversity are also increased, as well as crop production. In Greece, approximately 2.000.000 bee colonies (second in the E.U.) are reported with a relatively high density per km2. There are more than 20.000 beekeeping operators that produce about 20.000 tons of honey every year, while more than 65% of Greek honey is produced from honey dew. To this end, this study aims to identify and map major honeybee flora in the Greek mainland and the Greek islands (Fir forests, Pine forests and Oak forests) for the year 2019, in order to examine best regions to deploy honeybee colonies. In particular, a classification framework for mapping the main honeybee flora is introduced that is exploiting annual moderate-resolution satellite multi-temporal data. Additionally, a methodology is presented to generate a coarser training dataset by utilizing a high-spatial resolution, detailed land cover map. This process specifically focuses on the integration of honeybee flora classes that are not present in the land cover map but are of significant great importance for honey flora mapping. The goal was to enable large-scale classification without the computational resource constraints typically associated with such national scale frameworks. Experimental results are quite promising with the quantitative validation indicating overall accuracy of more than 85%.

Enabling country-scale land cover mapping with meter-resolution satellite imagery

High-resolution satellite images can provide abundant, detailed spatial information for land cover classification, which is particularly important for studying the complicated built environment. However, due to the complex land cover patterns, the costly training sample collections, and the severe distribution shifts of satellite imageries caused by, e.g., geographical differences or acquisition conditions, few studies have applied high-resolution images to land cover mapping in detailed categories at large scale. To fill this gap, we present a large-scale land cover dataset, Five-Billion-Pixels. It contains more than 5 billion labeled pixels of 150 high-resolution Gaofen-2 (4 m) satellite images, annotated in a 24-category system covering artificial-constructed, agricultural, and natural classes. In addition, we propose a deep-learning-based unsupervised domain adaptation approach that can transfer classification models trained on labeled dataset (referred to as the source domain) to unlabeled data (referred to as the target domain) for large-scale land cover mapping. Specifically, we introduce an end-to-end Siamese network employing dynamic pseudo-label assignment and class balancing strategy to perform adaptive domain joint learning. To validate the generalizability of our dataset and the proposed approach across different sensors and different geographical regions, we carry out land cover mapping on five megacities in China and six cities in other five Asian countries severally using: PlanetScope (3 m), Gaofen-1 (8 m), and Sentinel-2 (10 m) satellite images. Over a total study area of 60,000 km2, the experiments show promising results even though the input images are entirely unlabeled. The proposed approach, trained with the Five-Billion-Pixels dataset, enables high-quality and detailed land cover mapping across the whole country of China and some other Asian countries at meter-resolution.

Detailed Land Cover and Land Use Mapping, A New Approach: Case Study, Usumacinta Watershed, Mexico

Detailed Land Cover and Land Use Mapping, A New Approach: Case Study, Usumacinta Watershed, Mexico

Deep learning in land-use classification and geostatistics in soil pH mapping: a case study at Lakehead University Agricultural Research Station, Thunder Bay, Ontario, Canada

Grain agriculture is one of the main economic activities in Thunder Bay. However, farming faces many challenges: shorter growing seasons, severe weather conditions, climate change, and limited services and economic support. Lakehead University Agricultural Research Station (LUARS) is continuously working toward improving farming in Northern Ontario and testing different agricultural initiatives. Precision agriculture (PA) is one of the strategic means to improve Northern Ontario farming and sustain agricultural operations. Hence, PA requires accurate and up-to-date baseline information; the aim of this study is to develop baseline spatial information (land cover and soil pH variability maps) for the LUARS using remote sensing technology. A set of aerial images were acquired from the remotely piloted aircraft system (RPAS) and created an orthophoto (7-cm spatial resolution). A detailed land cover map was produced using a deep learning algorithm. We collected soil samples from three different sections (plots) of the area to analyze soil pH variations. Different geostatistical methods [ordinary kriging, Simple kriging, and empirical Bayesian kriging (EBK)] and deterministic interpolation techniques: local polynomial interpolation (LPI) and inverse distance weighted (IDW) were tested to map the spatial distribution of soil pH. A land cover classification achieved an overall accuracy of 73% and a Kappa coefficient of 0.70. Most of the homogeneous agricultural plots indicated higher producer’s and user’s classification accuracies. There was no linear correlation between soil sample data and vegetation indices derived from RPAS images to test the linear regression for soil pH mapping. EBK, LPI, and IDW produced soil pH maps with higher accuracies for Secs. 1–3, respectively.

Mapping native and non-native vegetation in the Brazilian Cerrado using freely available satellite products

Native vegetation across the Brazilian Cerrado is highly heterogeneous and biodiverse and provides important ecosystem services, including carbon and water balance regulation, however, land-use changes have been extensive. Conservation and restoration of native vegetation is essential and could be facilitated by detailed landcover maps. Here, across a large case study region in Goiás State, Brazil (1.1 Mha), we produced physiognomy level maps of native vegetation (n = 8) and other landcover types (n = 5). Seven different classification schemes using different combinations of input satellite imagery were used, with a Random Forest classifier and 2-stage approach implemented within Google Earth Engine. Overall classification accuracies ranged from 88.6–92.6% for native and non-native vegetation at the formation level (stage-1), and 70.7–77.9% for native vegetation at the physiognomy level (stage-2), across the seven different classifications schemes. The differences in classification accuracy resulting from varying the input imagery combination and quality control procedures used were small. However, a combination of seasonal Sentinel-1 (C-band synthetic aperture radar) and Sentinel-2 (surface reflectance) imagery resulted in the most accurate classification at a spatial resolution of 20 m. Classification accuracies when using Landsat-8 imagery were marginally lower, but still reasonable. Quality control procedures that account for vegetation burning when selecting vegetation reference data may also improve classification accuracy for some native vegetation types. Detailed landcover maps, produced using freely available satellite imagery and upscalable techniques, will be important tools for understanding vegetation functioning at the landscape scale and for implementing restoration projects.

Spatio-temporal patterns of flooded areas in the lower part of the Sana river basin (Bosnia and Herzegovina)

Floods are the most frequent and devastating natural hazard event in Bosnia and Herzegovina. The detected increase in extreme precipitation over the study area in the last period has altered flood event patterns due to climate changes. Higher frequency of flood events and lack of flood protection infrastructure has a severe impact on socio-economic sectors and natural ecosystems. This paper focuses on the identification of flooded areas for each single flooding event in the lower part of the Sana river basin during the period 2016-2020. For delineating flooded areas, both radar and optical satellite imagery were used. Data obtained after processing remote sensing images were overlaid with a detailed land cover map in order to get insight into flooded land cover types. From temporal aspects, floods are most common during the spring season. They are usually caused by rapid snowmelt and prolonged excessive precipitation. Considering spatial aspects, flooded areas vary from 110 to 522 hectares in the study area. Over 95% of the flooded areas are arable land, meadows and pastures. Most affected settlements by floods are urban and suburban area of Prijedor, Gomjenica, Hambarine, Rakovcani, Rizvanovici, Brezicani, Donja Dragotinja, Vitasavci, Svodna, Blagaj Rijeka and urban area of Novi Grad. The applied methodological approach represents a starting point for further investigation of flooded areas in the Sana basin and data obtained by this analysis can be used in water management, spatial planning and emergency planning.

Assessing Impact of Multiple Fires on a Tropical Peat Swamp Forest Using High and Very High-Resolution Satellite Images

Tropical peat swamp forests, found mainly in Southeast Asia, have been threatened by recurring El Niño fires. Repeated burnings form a complex and heterogeneous landscape comprising a mosaic of burned patches of different fire frequencies, requiring fine-scale assessment to understand their impact. We examined the impact of the El Niño fires of 1998 and 2003 on a tropical peat swamp forest in northern Borneo, with the combined use of high and very high-resolution satellite images. Object-based and pixel-based classifications were compared to classify a QuickBird image. Burned patches of different fire frequencies were derived based on unsupervised classification of the principal components of multitemporal Normalized Difference Water Index (NDWI) data. The results show that the object-based classification was more accurate than the pixel-based classification for generating a detailed land cover map. Fire frequency had a severe impact on the number of burned patches and the residual forest cover. Larger patch area retained more residual forest cover for the burned patches. Forest structure of burned-twice patches was more severely altered compared to burned-once patches. Two burned-once patches had a relatively promising recovery potential by natural regeneration due to higher residual forest cover, a vast number of large trees, and aboveground biomass. Except for the largest patch, rehabilitation seemed inevitable for burned-twice patches. This approach can be applied to assess the impact of multiple fires on other forest types for better post-fire forest management.

Integration of Sentinel optical and radar data for mapping smallholder coffee production systems in Vietnam

Perennial commodity crops, such as coffee, often play a large role globally in agricultural markets and supply chains and locally in livelihoods, poverty reduction, and biodiversity. Yet, the production of spatial information on these crops are often overlooked in favor of annual food crops. Remote sensing detection of coffee faces a particular set of challenges due to persistent cloud cover in the tropical “coffee belt,” hilly topography in coffee growing regions, diversity of coffee growing systems, and spectral similarity to other tree crops and agricultural land. Looking at the major coffee growing region in Dak Lak, Vietnam, we integrate multi-temporal 10 m optical Sentinel-2 and Sentinel-1 SAR data in order to map three coffee production systems: i) open-canopy sun coffee, ii) intercropped and other shaded coffee and iii) newly planted or young coffee.Leveraging Google Earth Engine (GEE), we compute five sets of features in order to best enhance separability between coffee and other land cover and within coffee production systems. The features include Sentinel-2 dry and wet season composites, Sentinel-1 texture features, Sentinel-1 spatiotemporal metrics, and topographic features. Using a random forest classification algorithm, we produce a 9-class land cover map including our three coffee production classes and a binary coffee/non-coffee map. The binary map has an overall accuracy of 89% and the three coffee production systems have user accuracies of 65, 56, 71% for sun coffee, intercropped coffee and newly planted coffee, respectively. This is a first effort at large-scale distinction of within-crop production styles and has implications across many applications. The binary coffee map can be used as a high-resolution crop mask, whereas the detailed land cover map can inform monitoring of deforestation dynamics, biodiversity, sustainability certification and implementation of climate adaptation strategies. This work offers a scalable approach to integrating optical and radar Sentinel data for production of spatially explicit agricultural information and contributes particularly to tree crop and agroforestry mapping, which often is overlooked in between agricultural and forestry sciences.

JOINT LAND COVER AND CROP TYPE MAPPING USING MULTI-TEMPORAL SENTINEL-2 DATA FROM VARIOUS ENVIRONMENTAL ZONES IN GREECE

Abstract. In this work, we elaborate on the gained insights from various classification experiments towards detailed land cover mapping over four representative regions of different environmental characteristics in Greece. In particular, the proposed methodology exploits Sentinel-2 data at an annual basis, for the joint classification of 35 land cover and crop type classes. A number of pre-processing steps were employed on the satellite data, in order to address atmospheric and geometric effects, as well as clouds and pertinent shadows. Several classification set-ups were designed and performed using either time series of spectral features or temporal features. The latter consisted of statistical metrics, derived from the spectral time series, and therefore were significantly reduced in dimension. Experiments using the Random Forest algorithm were performed by building several per-tile models, as well as cross- regional models based on training data from all considered regions/tiles. Overall classification accuracy rates exceeded 90% for most experiments. Further analysis on the experimental results highlighted that crop types were classified more accurately when using the spectral time series features, compared to the temporal ones. Classification accuracy for non-crop classes proved much less affected by the type of employed features. The inclusion of auxiliary data layers was beneficial in all cases, both for overall and for per-class accuracy metrics. Qualitative evaluation on the predicted maps further affirmed the efficiency of the developed methodology.

Quality assurance and assessment framework for land cover maps validation in the Copernicus Hot Spot Monitoring activity

ABSTRACT The Copernicus High-Resolution Hot Spot Monitoring activity (C-HSM) delivers a global dataset of Key Landscapes for Conservation (KLC), which are characterized by pronounced anthropogenic pressures that require high mapping accuracy. Detailed land cover and land cover change map products are freely available through the activity and include extensive map production accuracy assessments. Without a complete understanding of the map products’ spatial, temporal and logical consistencies, quality or quantified confidence levels, usability is reduced and can affect stakeholder decision-making and the implementation of sustainable solutions. For the quantitative accuracy assessment, a stratified random sampling approach was implemented where special emphasis was placed on (i) allocation of sampling units for rare land cover change categories; (ii) effective and accurate labelling of large numbers of sampling units; (iii) accuracy and area estimation in one consistent approach; and (iv) derivation of confidence intervals for all accuracy measures and area estimates. To handle correlations, large uncertainties, and complex probability density functions, bootstrapping was applied instead of analytical equations, which are based on normality assumptions. This paper presents the Quality Assurance and Quality Control framework applied to validate all the C-HSM thematic map products. The Kundelungu-Upemba KLC product results are presented as our use case.

DECISIONAL TREE MODELS FOR LAND COVER MAPPING AND CHANGE DETECTION BASED ON PHENOLOGICAL BEHAVIORS. APPLICATION CASE: LOCALIZATION OF NON-FULLY-EXPLOITED AGRICULTURAL SURFACES IN THE EASTERN PART OF THE HAOUZ PLAIN IN THE SEMI-ARID CENTRAL MOROCCO

Abstract. Great effort has been recently employed for the development of a modern and competitive agriculture in Morocco, growth in the agricultural sector is determined largely through the realization of thousands of new projects, and the support of the smallholder farmers at a national scale. Modernization of irrigation systems, and enlargement of the extent and spatial distribution of irrigated areas holds the key to increase annual productions. In this context, we established a unique procedure for monitoring the agricultural surfaces not fully exploited in terms of potential and production, in the semi-arid zone of the Haouz plain, central Morocco. We derived Normalized Difference Vegetation Index (NDVI) time series from Sentinel-2 (S2) and Landsat 8 (L8) high spatial resolution satellite images from 2016 to 2018. Seasonal phenological changes and land-cover dynamics, in addition to elevation models and landscape slopes, helped determine periods and thresholds suitable for classes separability, and establish a set of rules to be implemented in a Decision tree classifier model for a detailed land-cover mapping of the last three years. The agricultural zone was successfully separated from mountains and hills, and the derived maps of the three years yielded satisfying result with an OAthat reached above 91% for quite detailed landscape-type information. The outputs of this work hold promise to provide valuable information for planners, decision-makers and regional offices, to help smallholder farmers. Although this approach has been developed at regional-scale, it holds the potential to be adapted to larger scales, with the appropriate selection of land-cover types, and carful adjustment in the threshold values.

Object-based island hierarchical land cover classification using unmanned aerial vehicle multitype data

The unmanned aerial vehicle (UAV) is an emerging technology applied recently in land cover classification, owing to its ability to acquire very high-resolution spatial data, that has provided an effective means for detailed land cover mapping, especially for a small island area. Selecting suitable UAV-acquired data and exploring the combined use of UAV multitype data are of significance for island mapping. Nine classification models were established through a fusion method of visible, multispectral, and light detection and ranging (LIDAR) data acquired by UAVs. A two-level hierarchical land cover classification (level 1 and level 2) of the Donkey Island in China was performed using geographic object-based image analysis with random forest classifier. We investigated the performance of land cover classification models containing different sets of features (spectral, height, intensity, and shape features extracted from UAV data) and evaluated the importance of various features. The results demonstrate that the overall accuracy (OA) of the models generally increase with decreasing detail and the amount of information entering the classification process. The OA achieved range from 82.08% to 92.54% and 74.12% to 85.08% across the nine models for level 1 and level 2, respectively. The best result was achieved with a model combining all features based on multispectral and LIDAR data. Height and intensity information input significantly affect the quality of classification models positively, with height apparently more significant than LIDAR information. When comparing different features, spectral features prominently assist in discriminating land cover classes. The importance of height and intensity features to classification accuracy varies for the classification models, showing greater importance in models based on visible data.

Urban aliens and threatened near-naturals: Land-cover affects the species richness of alien- and threatened species in an urban-rural setting

Urbanisation has strong effects on biodiversity patterns, but impacts vary among species groups and across spatial scales. From a local biodiversity management perspective, a more general understanding of species richness across taxonomic groups is required. This study aims to investigate how fine-scale land-cover variables influence species richness patterns of locally threatened and alien species. The study was performed in Trondheim, Norway, covering a steep urbanisation gradient. Spatially correlated Generalised Linear Mixed Effects Models predicting the number of all-, threatened-and alien species by taxon, habitat, habitat heterogeneity and mean aspect within 500 m×500 m grid cells were constructed. The habitat categories were based on detailed land-cover maps. The highest number of threatened species was found in habitats relatively less affected by humans, whereas the number of alien species were only dependent on taxonomic group and spatial correlation. It is shown that land-cover variables within an administrative border can be used to make predictions on species richness within overarching species groups. Recommendations to biodiversity management agencies are to ensure protection of natural habitats to favour locally threatened species, and closely monitor urban areas to mitigate the introduction and spread of alien species.

Introducing APiC for regionalised land cover mapping on the national scale using Sentinel-2A imagery

Overcoming the obstacle of frequent cloud coverage in optical remote sensing data is essential for monitoring dynamic land surface processes from space. APiC, a novel adaptable pixel-based compositing and classification approach, is especially designed to use high resolution spatio-temporal space-borne data.Here, pixel-based compositing is used separately for training data and prediction data. First, cloud-free pixels covered by reference data are used within adapted composite periods to compile a training dataset. The compiled training dataset contains samples of spectral reflectances for respective land cover classes at each composite period. For land cover prediction, pixel-based compositing is then applied region-wide. Multiple prediction models are used based on temporal subsets of the compiled training dataset to dynamically account for cloud coverage at pixel level. Thus we present a data-driven classification approach which is applicable in regions with different weather conditions, species composition and phenology.The capability of our method is demonstrated by mapping 19 land cover classes across Germany for the year 2016 based on Sentinel-2A data. Since climatic conditions and thus plant phenology change on a large scale, the classification was carried out separately in six landscape regions of different biogeographical characteristics. The study drew on extensive ground validation data provided by the federal states of Germany.For each landscape region, composite periods of different lengths have been established, which differ regionally in their temporal arrangement as well as in their total number, emphasising the advantage of a flexible regionalised classification procedure. Using a random forest classifier and evaluating outcomes with independent reference data, an overall accuracy of 88% was achieved, with particularly high classification accuracy of around 90% for the major land cover types. We found that class imbalances have significant influence on classification accuracy. Based on multiple temporal subsets of the compiled training dataset, over 10,000 random forest models were calculated and their performance varied considerably across and within landscape regions. The calculated importance of composite periods show that a high temporal resolution of the compiled training dataset is necessary to better capture the different phenology of land cover types.In this study we demonstrate that APiC, due to its data-driven nature, is a very flexible compositing and classification approach making efficient use of dense satellite time series in areas with frequent cloud coverage. Hence, regionalisation can be given greater focus in future broad-scale classifications in order to facilitate better integration of small-scale biophysical conditions and achieve even better results in detailed land cover mapping.

Transpiration from subarctic deciduous woodlands: Environmental controls and contribution to ecosystem evapotranspiration

AbstractPotential land–climate feedbacks in subarctic regions, where rapid warming is driving forest expansion into the tundra, may be mediated by differences in transpiration of different plant functional types. Here, we assess the environmental controls of overstorey transpiration and its relevance for ecosystem evapotranspiration in subarctic deciduous woodlands. We measured overstorey transpiration of mountain birch canopies and ecosystem evapotranspiration in two locations in northern Fennoscandia, having dense (Abisko) and sparse (Kevo) overstories. For Kevo, we also upscale chamber‐measured understorey evapotranspiration from shrubs and lichen using a detailed land cover map. Subdaily evaporative fluxes were not affected by soil moisture and showed similar controls by vapour pressure deficit and radiation across sites. At the daily timescale, increases in evaporative demand led to proportionally higher contributions of overstorey transpiration to ecosystem evapotranspiration. For the entire growing season, the overstorey transpired 33% of ecosystem evapotranspiration in Abisko and only 16% in Kevo. At this latter site, the understorey had a higher leaf area index and contributed more to ecosystem evapotranspiration compared with the overstorey birch canopy. In Abisko, growing season evapotranspiration was 27% higher than precipitation, consistent with a gradual soil moisture depletion over the summer. Our results show that overstorey canopy transpiration in subarctic deciduous woodlands is not the dominant evaporative flux. However, given the observed environmental sensitivity of evapotranspiration components, the role of deciduous trees in driving ecosystem evapotranspiration may increase with the predicted increases in tree cover and evaporative demand across subarctic regions.

The Spatial and Temporal Land Cover Patterns of the Qazaly Irrigation Zone in 2003–2018: The Case of Syrdarya River’s Lower Reaches, Kazakhstan

In this study, the spatial and temporal patterns of the land cover were monitored within the Qazaly irrigation zone located in the deltaic zone of the Syrdarya river in the surroundings of the former Aral Sea. A 16-day MODIS (Moderate Resolution Imaging Spectroradiometer) Aqua NDVI (Normalized Difference Vegetation Index) data product with a spatial resolution of 250 meters was used for this purpose, covering the period between 2003 and 2018. Field survey results obtained in 2018 were used to build a sample dataset. The random forests supervised classification machine learning algorithm was used to map land cover, which produced good results with an overall accuracy of about 0.8. Statistics on land cover change were calculated and analyzed. The correctness of obtained classes was checked with Landsat 8 (OLI, The Operational Land Imager) images. Detailed land cover maps, including rice cropland, were derived. During the observation period, the rice croplands increased, while the generally irrigated area decreased.

Geographic Object Based Image Analysis of WorldView-3 Imagery for Urban Hydrologic Modelling at the Catchment Scale

China’s Sponge City initiative will involve widespread installation of new stormwater infrastructure including green roofs, permeable pavements and rain gardens in at least 30 cities. Hydrologic modelling can support the planning of Sponge Cities at the catchment scale, however, highly detailed spatial data for model input can be challenging to compile from the various authorities, or, if available, may not be sufficiently detailed or updated. Remote sensing methods show great promise for mitigating this challenge due to their ability to efficiently classify satellite images into categories relevant to a specific application. In this study Geographic Object Based Image Analysis (GEOBIA) was applied to WorldView-3 satellite imagery (2017) to create a detailed land cover map of an urban catchment area in Beijing. While land cover classification results based on a Bayesian machine learning classifier alone provided an overall land cover classification accuracy of 63%, the subsequent inclusion of a series of refining rules in combination with supplementary data (including elevation and parcel delineations), yielded the significantly improved overall accuracy of 76%. Results of the land cover classification highlight the limitations of automated classification based on satellite imagery alone and the value of supplementary data and additional rules to refine classification results. Catchment scale hydrologic modelling based on the generated land cover results indicated that 61 to 82% of rainfall volume could be captured for a range of 24 h design storms under varying degrees of Sponge City implementation.

Monitoring opencast mine restorations using Unmanned Aerial System (UAS) imagery

Open-pit mine is still an unavoidable activity but can become unsustainable without the restoration of degraded sites. Monitoring the restoration after extractive activities is a legal requirement for mine companies and public administrations in many countries, involving financial provisions for environmental liabilities. The objective of this contribution is to present a rigorous, low-cost and easy-to-use application of Unmanned Aerial Systems (UAS) for supporting opencast mining and restoration monitoring, complementing the inspections with very high (<10 cm) spatial resolution multispectral imagery, and improving any restoration documentation with detailed land cover maps. The potential of UAS as a tool to control restoration works is presented in a calcareous quarry that has undergone different post-mining restoration actions in the last 20 years, representing 4 reclaimed stages. We used a small (<2 kg) drone equipped with a multispectral sensor, along with field spectroradiometer measurements that were used to radiometrically correct the UAS sensor data. Imagery was processed with photogrammetric and Remote Sensing and Geographical Information Systems software, resulting in spectral information, vegetation and soil indices, structural information and land cover maps. Spectral data and land cover classification, which were validated through ground-truth plots, aided in the detection and quantification of mine waste dumping, bare soil and other land cover extension. Moreover, plant formations and vegetation development were evaluated, allowing a quantitative, but at the same time visual and intuitive comparison with the surrounding reference systems. The protocol resulting from this research constitutes a pipeline solution intended for the implementation by public administrations and privates companies for precisely evaluating restoration dynamics in an expedient manner at a very affordable budget. Furthermore, the proposed solution prevents subjective interpretations by providing objective data, which integrate new technologies at the service of scientists, environmental managers and decision makers.

Detailed Land Cover Mapping from Multitemporal Landsat-8 Data of Different Cloud Cover

Detailed, accurate and frequent land cover mapping is a prerequisite for several important geospatial applications and the fulfilment of current sustainable development goals. This paper introduces a methodology for the classification of annual high-resolution satellite data into several detailed land cover classes. In particular, a nomenclature with 27 different classes was introduced based on CORINE Land Cover (CLC) Level-3 categories and further analysing various crop types. Without employing cloud masks and/or interpolation procedures, we formed experimental datasets of Landsat-8 (L8) images with gradually increased cloud cover in order to assess the influence of cloud presence on the reference data and the resulting classification accuracy. The performance of shallow kernel-based and deep patch-based machine learning classification frameworks was evaluated. Quantitatively, the resulting overall accuracy rates differed within a range of less than 3%; however, maps produced based on Support Vector Machines (SVM) were more accurate across class boundaries and the respective framework was less computationally expensive compared to the applied patch-based deep Convolutional Neural Network (CNN). Further experimental results and analysis indicated that employing all multitemporal images with up to 30% cloud cover delivered relatively higher overall accuracy rates as well as the highest per-class accuracy rates. Moreover, by selecting 70% of the top-ranked features after applying a feature selection strategy, slightly higher accuracy rates were achieved. A detailed discussion of the quantitative and qualitative evaluation outcomes further elaborates on the performance of all considered classes and highlights different aspects of their spectral behaviour and separability.