Global Land Cover Classification Research Articles

Assessment of Land Cover Status and Change in the World and "the Belt and Road" Region from 2016 to 2020.

The assessment of land cover and changes will help to understand the temporal and spatial pattern of land cover in the world and the Belt and Road (B&R) region, and provide reference information for global sustainable development and the Belt and Road construction. In this paper, the 1 km global land cover classification maps of 2016 and 2020 with a high accuracy of 88% are mapped using the Moderate Resolution Imaging Spectroradiometer (MODIS) time series surface reflectance products. Based on the maps, the land cover status of the world and the Belt and Road region, the land cover change from 2016 to 2020, and the mutual transformation characteristics between various types, are analyzed. The research results indicate that from 2016 to 2020, the global change rates of cropland, forest, grassland, and impervious surface are 0.25%, 0.22%, 0.08% and 3.41%, respectively. In the Belt and Road region, the change rates of cropland, forest, grassland, and impervious surface are 0.42%, 0.60%, -0.55% and 2.98% respectively. The assessment results will help to clarify the spatial pattern of land cover change in the five years from 2016 to 2020, so as to provide valuable scientific information for the global realization of sustainable development goals and the construction of the B&R.

High-Confidence Sample Generation Technology and Application for Global Land-Cover Classification

Deep learning technology has become one of the most important technologies in remote sensing land classification applications. Its powerful sample-learning and information-mining abilities promote the continuous improvement of classification accuracy. A large volume of high-quality and representative sample data is the premise for the successful application of deep learning technology. Conventional methods of obtaining samples through manual delineation or surface surveys require a great deal of manpower and material resources. Therefore, the inability to obtain sufficient and widely representative high-quality samples is one of the key factors limiting the application of deep learning technology. In this study, the method of generating sample data obtains high-confidence classification results from a variety of existing high-quality classification products as deep learning samples, which are then used to support the application of deep learning technology for land-cover classification. When the three global land-cover classification products, FROM-GLC-2015, GLC_FCS30-2015, and GlobeLand30, have the same type of discrimination, the sample is considered a high-confidence sample. Based on this, a large volume of sample data widely distributed around the world was obtained. Using the extracted samples, a random forest classifier was trained using multiple types of information from the Landsat data, and land-cover classification was achieved. Application experiments were conducted in several typical regions, and the classification results were verified. The results showed that the classification accuracy of random forests under the support of samples generated using the sample extraction method proposed in this article was considerably higher than that of the three land-cover classification products.

Global Land Cover Mapping Using Annual Clear-sky Composites from FY3D/MERSI-II

ABSTRACT Medium Resolution Spectral Imager-II (MERSI-II) on board FY-3D satellite is a visible and infrared spectral imaging instrument featured with six 250-m spatial resolution channels (blue, green, red, NIR, 10.8 µm, 12.0 µm). Such unique spectral and spatial attributes provide significant opportunities and challenges for land cover (LC) mapping research. This paper reports the production of a 250-m spatial resolution global LC classification (under IGBP scheme) from MERSI-II NDVI operational products from May 2019 to April 2020. To extract robust features for discriminating LC types, we adopted three classic compositing criteria (MaxNDVI, Mean8NDVI, and Mean4BT) and two new compositing criteria (3minRedBlue and 3minRedNIR) to obtain high-quality annual composites. Random Forest (RF) algorithm was used for supervised learning, and 10-fold cross validation (CV) based on ‘site-splits’ was implemented to evaluate classification accuracy. The derived MERIS-II LC map was compared at pixel level with two widely used LC products (MODIS and VIIRS), and consistency among the three maps was investigated. Results showed that the higher resolution afforded by 250-m MERSI-II data in combination with a refined input features produced a global LC map with an overall accuracy of 79–86%. Most classes exhibited low commission and omission errors except for some classes of mixed nature (i.e. wetland, cropland/natural vegetation mosaic, savanna, woody savanna). The MERSI-II LC map generally agreed better with MODIS than with VIIRS, and the proportion of land pixels (excluding Antarctic) for which MERSI-II shows the same LC label with either MODIS or VIIRS is very large (0.70–0.82). Composite data contributed differently to total accuracy, with Mean8NDVI being the most spatial coherent and most contributive and MaxNDVI being the opposite. The other three composites (Mean4BT, 3minRedBlue, and 3minRedNIR) showed moderate spatial coherence and contributed to classification variedly by extracting information from different phenology phases. The unique 250-m blue and green bands of MERSI-II were found to be helpful to discriminate between wetland and needleleaf forest in high latitudes and between cities and grassland. This study confirms that with the RF algorithm (one of the state-of-art supervised machine learning algorithms) and image compositing techniques, accurate LC map can be derived from MERSI-II. Future work will attempt to include more bands and compositing criteria to further improve classification accuracy.

Developments in Land Use and Land Cover Classification Techniques in Remote Sensing: A Review

Studies on land use and land cover changes (LULCC) have been a great concern due to their contribution to the policies formulation and strategic plans in different areas and at different scales. The LULCC when intense and on a global scale can be catastrophic if not detected and monitored affecting the key aspects of the ecosystem’s functions. For decades, technological developments and tools of geographic information systems (GIS), remote sensing (RS) and machine learning (ML) since data acquisition, processing and results in diffusion have been investigated to access landscape conditions and hence, different land use and land cover classification systems have been performed at different levels. Providing coherent guidelines, based on literature review, to examine, evaluate and spread such conditions could be a rich contribution. Therefore, hundreds of relevant studies available in different databases (Science Direct, Scopus, Google Scholar) demonstrating advances achieved in local, regional and global land cover classification products at different spatial, spectral and temporal resolutions over the past decades were selected and investigated. This article aims to show the main tools, data, approaches applied for analysis, assessment, mapping and monitoring of LULCC and to investigate some associated challenges and limitations that may influence the performance of future works, through a progressive perspective. Based on this study, despite the advances archived in recent decades, issues related to multi-source, multi-temporal and multi-level analysis, robustness and quality, scalability need to be further studied as they constitute some of the main challenges for remote sensing.

Using Google Earth Engine to classify unique forest and agroforest classes using a mix of Sentinel 2a spectral data and topographical features: a Sri Lanka case study

Global land cover classifications may lead to the loss of important local and national nuances such as forest and agroforestry classes. These classes are important to local contexts because they contribute to sustainable land management systems. This paper demonstrates the application of Sentinel-2A satellite images, elevation data, and the Google Earth Engine platform to generate more detailed, specialist land cover classification for forestry classes important in Sri Lanka deriving ten spectral, 16 textural, and three topographical features from the input datasets. The random forest classification model discriminates vegetation types as forest, forest plantations, shrub, grassland, home garden, and cultivation with an overall accuracy of 94% and kappa value of 0.91. Results indicate the elevation feature contributes the most to discriminate forest and agroforestry classes, and red band (664.6 nm) textural metrics derived from grey-level co-occurrence matrix analysis are more useful for separating the home garden from other land cover classes.

A platform for land use and land cover data integration and trajectory analysis

Information on land use and land cover (LULC) is essential to support governments in making decisions about the impact of human activities on the environment, planning the use of natural resources, conserving biodiversity, and monitoring climate change. Nowadays, different initiatives systematically produce information on LULC dynamics, on global, national, and regional scales. Examples of open and global LULC data products are Global Land-cover Classification with a Fine Classification System, Copernicus Global Land Service, and Global Land Cover by European Space Agency (ESA). At the national and regional level in Brazil, we can cite the data sets produced by PRODES, TerraClass, MapBiomas, and IBGE. Although these initiatives provide rich collections of open LULC maps, there is still a gap in tools that facilitate the integration of these data sets. The integrated analysis of these collections requires considerable effort by researchers who have to download, organize and harmonize them in their local computers, facing with different spatiotemporal resolutions and classification systems containing distinct class numbers, names and meanings. Besides that, these collections are distributed in different data formats through files or web services. To minimize these efforts, we propose a platform that allows users to access LULC collections from distinct sources, map their distinct classification systems, and retrieve LULC trajectories associated with spatial locations by integrating these collections. Besides the platform architecture description, this paper presents a case study that demonstrates its use in the integration and analysis.

Rice Mapping Using a BiLSTM-Attention Model from Multitemporal Sentinel-1 Data

Timely and accurate rice distribution information is needed to ensure the sustainable development of food production and food security. With its unique advantages, synthetic aperture radar (SAR) can monitor the rice distribution in tropical and subtropical areas under any type of weather condition. This study proposes an accurate rice extraction and mapping framework that can solve the issues of low sample production efficiency and fragmented rice plots when prior information on rice distribution is insufficient. The experiment was carried out using multitemporal Sentinel-1A Data in Zhanjiang, China. First, the temporal characteristic map was used for the visualization of rice distribution to improve the efficiency of rice sample production. Second, rice classification was carried out based on the BiLSTM-Attention model, which focuses on learning the key information of rice and non-rice in the backscattering coefficient curve and gives different types of attention to rice and non-rice features. Finally, the rice classification results were optimized based on the high-precision global land cover classification map. The experimental results showed that the classification accuracy of the proposed framework on the test dataset was 0.9351, the kappa coefficient was 0.8703, and the extracted plots maintained good integrity. Compared with the statistical data, the consistency reached 94.6%. Therefore, the framework proposed in this study can be used to extract rice distribution information accurately and efficiently.

Classification and Evaluation of Extended PICS (EPICS) on a Global Scale for Calibration and Stability Monitoring of Optical Satellite Sensors

Historically stable areas across North Africa, known as pseudo invariant calibration sites (PICS), have been used as targets for the calibration and monitoring of optical satellite sensors. However, two major drawbacks exist for these sites: first is the dependency on a single location to be always invariant, and second is the limited amount of observation achieved using these sites. As a result, longer time periods are necessary to construct a dense dataset to assess the radiometric performance of on-orbit optical sensors and confirm that the change detected is sensor-specific rather than site-specific. This work presents a global land cover classification to obtain an extended pseudo invariant calibration site (EPICS) on a global scale using Landsat-8 Operational Land Imager (OLI) data. This technique provides multiple calibration sites across the globe, allowing for the building of richer datasets in a shorter time frame compared to the traditional approach (PICS), with the advantage of assessing the calibration and stability of the sensors faster, detecting possible changes sooner and correcting them accordingly. This work identified 23 World Reference System two (WRS-2) path/row locations around the globe as part of the global EPICS. These EPICS have the advantage of achieving multiple observations per day, with similar spectral characteristics compared to traditional PICS, while still producing a temporal coefficient of variation (ratio of temporal standard deviation and temporal mean) less than 4% for all bands, with some as low as 2.7%.

Detection of land cover changes in Baluchistan (shared between Iran, Pakistan, and Afghanistan) using the MODIS Land Cover Product

Land use and land cover (LULC) changes have been one of the most important and persistent factors recently causing changes in the Earth’s land. The present study aimed to detect land use and land cover (LULC) changes in Baluchistan, in Southwestern Asia, which is shared by the three countries of Iran, Pakistan, and Afghanistan, using satellite remote-sensing products. To this end, the global land cover classification provided for a period of 13 years from 2001 to 2013 by the MODIS Land Cover Type product (MCD12Q1) was used. The changes and dynamics of different land cover classes were investigated using net change analysis and cross-tabulating matrix analysis methods. The net change analysis showed that the most area of Baluchistan is covered by the barren or sparsely vegetated land cover (about 82%) and the shrubland (about 16%) classes. The dynamics analysis of different land cover classes also indicated that there were almost mutually inverse relationships between the different land cover classes in Baluchistan. Such mutual relationships were most common between the following pair classes: shrublands—bare and non-vegetated lands; grasslands—bare and non-vegetated land classes; croplands—bare and non-vegetated lands classes; and shrublands—grasslands. The most unstable land cover classes in this territory were forests, Savannas, and grassland classes. Also, the analysis of land cover changes in the period 2001–2013 provided no clear and accurate evidence of desertification and land degradation at this spatial scale in Baluchistan.

An Integrated Land Cover Mapping Method Suitable for Low-Accuracy Areas in Global Land Cover Maps

In land cover mapping, an area with complex topography or heterogeneous land covers is usually poorly classified and therefore defined as a low-accuracy area. The low-accuracy areas are important because they restrict the overall accuracy (OA) of global land cover classification (LCC) data generated. In this paper, low-accuracy areas in China (extracted from the MODIS global LCC maps) were taken as examples, identified as the regions having lower accuracy than the average OA of China. An integrated land cover mapping method targeting low-accuracy regions was developed and tested in eight representative low-accuracy regions of China. The method optimized procedures of image choosing and sample selection based on an existent visually-interpreted regional LCC dataset with high accuracies. Five algorithms and 16 groups of classification features were compared to achieve the highest OA. The support vector machine (SVM) achieved the highest mean OA (81.5%) when only spectral bands were classified. Aspect tended to attenuate OA as a classification feature. The optimal classification features for different regions largely depends on the topographic feature of vegetation. The mean OA for eight low-accuracy regions was 84.4% by the proposed method in this study, which exceeded the mean OA of most precedent global land cover datasets. The new method can be applied worldwide to improve land cover mapping of low-accuracy areas in global land cover maps.

CoSpace: Common Subspace Learning From Hyperspectral-Multispectral Correspondences

With a large amount of open satellite multispectral imagery (e.g., Sentinel-2 and Landsat-8), considerable attention has been paid to global multispectral land cover classification. However, its limited spectral information hinders further improving the classification performance. Hyperspectral imaging enables discrimination between spectrally similar classes but its swath width from space is narrow compared to multispectral ones. To achieve accurate land cover classification over a large coverage, we propose a cross-modality feature learning framework, called common subspace learning (CoSpace), by jointly considering subspace learning and supervised classification. By locally aligning the manifold structure of the two modalities, CoSpace linearly learns a shared latent subspace from hyperspectral-multispectral(HS-MS) correspondences. The multispectral out-of-samples can be then projected into the subspace, which are expected to take advantages of rich spectral information of the corresponding hyperspectral data used for learning, and thus leads to a better classification. Extensive experiments on two simulated HSMS datasets (University of Houston and Chikusei), where HS-MS data sets have trade-offs between coverage and spectral resolution, are performed to demonstrate the superiority and effectiveness of the proposed method in comparison with previous state-of-the-art methods.

Global hydro-climatic biomes identified via multitask learning

Abstract. The most widely used global land cover and climate classifications are based on vegetation characteristics and/or climatic conditions derived from observational data. However, these classification schemes do not directly stem from the characteristic interaction between the local climate and the biotic environment. In this work, we model the dynamic interplay between vegetation and local climate in order to delineate ecoregions that share a coherent response to hydro-climate variability. Our novel framework is based on a multitask learning approach that discovers the spatial relationships among different locations by learning a low-dimensional representation of predictive structures. This low-dimensional representation is combined with a clustering algorithm that yields a classification of biomes with coherent behaviour. Experimental results using global observation-based datasets indicate that, without the need to prescribe any land cover information, the identified regions of coherent climate–vegetation interactions agree well with the expectations derived from traditional global land cover maps. The resulting global hydro-climatic biomes can be used to analyse the anomalous behaviour of specific ecosystems in response to climate extremes and to benchmark climate–vegetation interactions in Earth system models.

Measuring land-use and land-cover change using the U.S. department of agriculture’s cropland data layer: Cautions and recommendations

Monitoring agricultural land is important for understanding and managing food production, environmental conservation efforts, and climate change. The United States Department of Agriculture's Cropland Data Layer (CDL), an annual satellite imagery-derived land cover map, has been increasingly used for this application since complete coverage of the conterminous United States became available in 2008. However, the CDL is designed and produced with the intent of mapping annual land cover rather than tracking changes over time, and as a result certain precautions are needed in multi-year change analyses to minimize error and misapplication. We highlight scenarios that require special considerations, suggest solutions to key challenges, and propose a set of recommended good practices and general guidelines for CDL-based land change estimation. We also characterize a problematic issue of crop area underestimation bias within the CDL that needs to be accounted for and corrected when calculating changes to crop and cropland areas. When used appropriately and in conjunction with related information, the CDL is a valuable and effective tool for detecting diverse trends in agriculture. By explicitly discussing the methods and techniques for post-classification measurement of land-cover and land-use change using the CDL, we aim to further stimulate the discourse and continued development of suitable methodologies. Recommendations generated here are intended specifically for the CDL but may be broadly applicable to additional remotely-sensed land cover datasets including the National Land Cover Database (NLCD), Moderate Resolution Imaging Spectroradiometer (MODIS)-based land cover products, and other regional, national, and global land cover classification maps.

Using the 500 m MODIS land cover product to derive a consistent continental scale 30 m Landsat land cover classification

Classification is a fundamental process in remote sensing used to relate pixel values to land cover classes present on the surface. Over large areas land cover classification is challenging particularly due to the cost and difficulty of collecting representative training data that enable classifiers to be consistent and locally reliable. A novel methodology to classify large volume Landsat data using high quality training data derived from the 500m MODIS land cover product is demonstrated and used to generate a 30m land cover classification for all of North America between 20°N and 50°N. Publically available 30m global monthly Web-enabled Landsat Data (GWELD) products generated from every available Landsat 7 ETM+ and Landsat 5 TM image for a three year period, that are defined aligned to the MODIS land products and are consistently pre-processed data (cloud-screened, saturation flagged, atmospherically corrected and normalized to nadir BRDF adjusted reflectance), were classified. The MODIS 500m land cover product was filtered judiciously, using only good quality pixels that did not change land cover class in 2009, 2010 or 2011, followed by automated selection of spatially corresponding 30m GWELD temporal metric values, to define a large training data set sampled across North America. The training data were sampled so that the class proportions were the same as the North America MODIS land cover product class proportions and corresponded to 1% of the 500m and <0.005% of the 30m pixels. Thirty nine GWELD temporal metrics for every 30m North America pixel location were classified using (a) a single random forest, and (b) a locally adaptive method with a random forest classifier derived and applied locally and the classification results spatially mosaicked together. The land cover classification results appeared geographically plausible and at synoptic scale were similar to the MODIS land cover product. Detailed visual inspection revealed that the locally adaptive random forest classifications and associated classification confidences were generally more coherent than the single random forest classification results. The level of agreement between the 30m classifications and the MODIS land cover product derived training data was assessed by bootstrapping the random forest implementation. The locally adaptive random forest classification had higher overall agreement (95.44%, 0.9443 kappa) than the single random forest classification (93.13%, 0.9195 kappa). The paper concludes with a discussion of future research including the potential for automated global land cover classification.

The Standardization and Harmonization of Land Cover Classification Systems towards Harmonized Datasets: A Review

A number of national, regional and global land cover classification systems have been developed to meet specific user requirements for land cover mapping exercises, independent of scale, nomenclature and quality. However, this variety of land-cover classification systems limits the compatibility and comparability of land cover data. Furthermore, the current lack of interoperability between different land cover datasets, often stemming from incompatible land cover classification systems, makes analysis of multi-source, heterogeneous land cover data for various applications a very difficult task. This paper provides a critical review of the harmonization of land cover classification systems, which facilitates the generation, use and analysis of land cover maps consistently. Harmonization of existing land cover classification systems is essential to improve their cross-comparison and validation for understanding landscape patterns and changes. The paper reviews major land cover classification standards according to different scales, summarizes studies on harmonizing land cover mapping, and discusses some research problems that need to be solved and some future research directions.

The first all-season sample set for mapping global land cover with Landsat-8 data

We report the world’s first all-season training and validation sample sets for global land cover classification with Landsat-8 data. Prior to this, such samples were only available at a single date primarily from the growing season. It is unknown how much limitation such a single-date sample has to mapping global land cover in other seasons of the year. To answer this question, we selected available Landsat-8 images from four seasons and collected training and validation samples from them. We compared the performances of training samples in different seasons using Random Forest algorithm. We found that the use of training samples from any individual season would result in the best overall classification accuracy when validated by samples in the same season. The global overall accuracy from combined best seasonal results was 67.2% when classifying the 11 Level-1 classes in the Finer Resolution Observation and Monitoring of Global Land Cover (FROM-GLC) classification system. The use of training samples from all seasons (named all-season training sample set hereafter) produced an overall accuracy of 67.0%. We also tested classification within 10° latitude 60° longitude zones using all-season training subsample within each zone and obtained an overall accuracy of 70.2%. This indicates that properly grouped subsamples in space can help improve classification accuracies. All the results in this study seem to suggest that it is possible to use an all-season training sample set to reach global optimality with universal applicability in classifying images acquired at any time of a year for global land cover mapping.

GLOBAL LAND COVER CLASSIFICATION USING MODIS SURFACE REFLECTANCE PROSUCTS

The objective of this study is to develop high accuracy land cover classification algorithm for Global scale by using multi-temporal MODIS land reflectance products. In this study, time-domain co-occurrence matrix was introduced as a classification feature which provides time-series signature of land covers. Further, the non-parametric minimum distance classifier was introduced for timedomain co-occurrence matrix, which performs multi-dimensional pattern matching for time-domain co-occurrence matrices of a classification target pixel and each classification classes. The global land cover classification experiments have been conducted by applying the proposed classification method using 46 multi-temporal(in one year) SR(Surface Reflectance) and NBAR(Nadir BRDF-Adjusted Reflectance) products, respectively. IGBP 17 land cover categories were used in our classification experiments. As the results, SR and NBAR products showed similar classification accuracy of 99%.

GLOBAL LAND COVER CLASSIFICATION USING MODIS SURFACE REFLECTANCE PROSUCTS

Abstract. The objective of this study is to develop high accuracy land cover classification algorithm for Global scale by using multi-temporal MODIS land reflectance products. In this study, time-domain co-occurrence matrix was introduced as a classification feature which provides time-series signature of land covers. Further, the non-parametric minimum distance classifier was introduced for timedomain co-occurrence matrix, which performs multi-dimensional pattern matching for time-domain co-occurrence matrices of a classification target pixel and each classification classes. The global land cover classification experiments have been conducted by applying the proposed classification method using 46 multi-temporal(in one year) SR(Surface Reflectance) and NBAR(Nadir BRDF-Adjusted Reflectance) products, respectively. IGBP 17 land cover categories were used in our classification experiments. As the results, SR and NBAR products showed similar classification accuracy of 99%.

Development and validation of the global surface type data product from S-NPP VIIRS

ABSTRACTAccurate representation of actual terrestrial surface types at regional to global scales is an important element for many applications. Based on National Aeronautics and Space Administration Moderate Resolution Imaging Spectroradiometer land cover algorithms, a global surface-type product from observations of the Visible Infrared Imaging Radiometer Suite (VIIRS) on board the Suomi National Polar-orbiting Partnership, provides consistent global land cover classification map for various studies, such as land surface modelling for numerical weather predictions, land management, biodiversity and hydrological modelling, and carbon and ecosystem studies. This letter introduces the development and validation of the VIIRS global surface-type product using the land cover classification scheme of the International Geosphere-Biosphere Programme. Surface reflectance data from VIIRS were composited into monthly data and then into annual metrics. The C5.0 decision tree classifier was used to determine the surface type for each pixel in a 1 km grid. To quantitatively evaluate accuracies of the new surface type product, a visual interpretation-based validation was performed in which high-resolution satellite images and other ancillary data were used as the reference. The validation result based on the large validation data set indicated that (78.64 ± 0.57)% overall classification accuracy was achieved.

Assessment of a Global Land Cover Classification Allocated Across the Landscape of Georgia (USA)

Aims: To assess the agreement of a globalland covermap to reference imagery when applied to a region (state) of the southern United States and to determine whether different sampling designs or the use of broader land class definitions can overcome problems associated with the inherent heterogeneity of land use in the region . Study Design:We assessed the agreement of theGlob Cover 2009 global, medium resolution land cover assignments within the State of Georgia to USDA NAIP reference imagery. We performed the assessment using two statistically random sampling methods: pixel-based and block-based sampling. We then grouped some land classes according to possibilities of agreementrelationships expressed by others, and assessed the agreement using these systems. Place and Duration of Study:State of Georgia (USA). Imagery and reference data acquired in 2009. Methodology: Sample: We examined 3,930 sample pixels or pixel blocks from 16 land cover classes. Each sample was allocated a land class in the GlobCover 2009 dat abase. Each sample was interpreted as a land class through photo -interpretation of