Land Use/Land Cover Mapping Using Multitemporal Sentinel-2 Imagery and Four Classification Methods—A Case Study from Dak Nong, Vietnam

Automated Mapping of Land Cover Type within International Heterogenous Landscapes Using Sentinel-2 Imagery with Ancillary Geospatial Data

A near-global framework for automated training data generation and land cover classification using shallow machine learning with low-density time series imagery does not exist. This study presents a methodology to map nine-class, six-class, and five-class land cover using two dates (winter and non-winter) of a Sentinel-2 granule across seven international sites. The approach uses a series of spectral, textural, and distance decision functions combined with modified ancillary layers (such as global impervious surface and global tree cover) to create binary masks from which to generate a balanced set of training data applied to a random forest classifier. For the land cover masks, stepwise threshold adjustments were applied to reflectance, spectral index values, and Euclidean distance layers, with 62 combinations evaluated. Global (all seven scenes) and regional (arid, tropics, and temperate) adaptive thresholds were computed. An annual 95th and 5th percentile NDVI composite was used to provide temporal corrections to the decision functions, and these corrections were compared against the original model. The accuracy assessment found that the regional adaptive thresholds for both the two-date land cover and the temporally corrected land cover could accurately map land cover type within nine-class (68.4% vs. 73.1%), six-class (79.8% vs. 82.8%), and five-class (80.1% vs. 85.1%) schemes. Lastly, the five-class and six-class models were compared with a manually labeled deep learning model (Esri), where they performed with similar accuracies (five classes: Esri 80.0 ± 3.4%, region corrected 85.1 ± 2.9%). The results highlight not only performance in line with an intensive deep learning approach, but also that reasonably accurate models can be created without a full annual time series of imagery.

Limiting the Collection of Ground Truth Data for Land Use and Land Cover Maps with Machine Learning Algorithms

Land use and land cover (LULC) classification maps help understand the state and trends of agricultural production and provide insights for applications in environmental monitoring. One of the major downfalls of the LULC technique is inherently linked to its need for ground truth data to cross-validate maps. This paper aimed at evaluating the efficiency of machine learning (ML) in limiting the use of ground truth data for LULC maps. This was accomplished by (1) extracting reliable LULC information from Sentinel-2 and Landsat-8 s images, (2) generating remote sensing indices used to train ML algorithms, and (3) comparing the results with ground truth data. The remote sensing indices that were tested include the difference vegetation index (DVI), the normalized difference vegetation index (NDVI), the normalized built-up index (NDBI), the urban index (UI), and the normalized bare land index (NBLI). Extracted vegetation indices were evaluated on three ML algorithms, namely, random forest (RF), k-nearest neighbour (K-NN), and k dimensional-tree (KD-Tree). The accuracy of these algorithms was assessed with standard statistical measures and ground truth data randomly collected in Prince Edward Island, Canada. Results showed that high kappa coefficient values were achieved by K-NN (82% and 74%), KD-Tree (80% and 78%), and RF (83% and 73%) for Sentinel-2A and Landsat-8 imagery, respectively. RF was a better classifier than K-NN and KD-Tree and had the highest overall accuracy with Sentinel-2A satellite images (92%). This approach provides the basis for limiting the collection of ground truth data and thus reduces the labour cost, time, and resources needed to collect ground truth data for LULC maps.

Can a Hierarchical Classification of Sentinel-2 Data Improve Land Cover Mapping?

Monitoring of land cover plays an important role in effective environmental management, assessment of natural resources, environmental protection, urban planning and sustainable development. Increasing demand for accurate and repeatable information on land cover and land cover changes causes rapid development of the advanced, machine learning algorithms dedicated to land cover mapping using satellite images. Free and open access to Sentinel-2 data, characterized with high spatial and temporal resolution, increased the potential to map and to monitor land surface with high accuracy and frequency. Despite a considerable number of approaches towards land cover classification based on satellite data, there is still a challenge to clearly separate complex land cover classes, for example grasslands, arable land and wetlands. The aim of this study is to examine, whether a hierarchal classification of Sentinel-2 data can improve the accuracy of land cover mapping and delineation of complex land cover classes. The study is conducted in the Lodz Province, in central Poland. The pixel-based land cover classification is carried out using the machine learning Random Forest (RF) algorithm, based on a time series of Sentinel-2 imagery acquired in 2020. The following nine land cover classes are mapped: sealed surfaces, woodland broadleaved, woodland coniferous, shrubs, permanent herbaceous (grassy cover), periodically herbaceous (i.e., arable land), mosses, non-vegetated (bare soil) and water bodies. The land cover classification is conducted following two approaches: (1) flat, where all land cover classes are classified together, and (2) hierarchical, where the stratification is applied to first separate the most stable land cover classes and then classifying the most problematic once. The national databases served as the source of the reference sampling plots for the classification process. The process of selection and verification of the reference sampling plots is performed automatically. To assess the stability of the classification models the classification processes are performed iteratively. The results of this study confirmed that the hierarchical approach gave more accurate results compared to the commonly used flat approach. The median of the overall accuracy (OA) of the hierarchical classification was higher by 3–9 percentage points compared to the flat one. Of interest, the OA of the hierarchical classification reached 0.93–0.99, whereas the flat approach reached 0.90. Individual classes are also better classified in the hierarchical approach.

From Land Cover Map to Land Use Map: A Combined Pixel-Based and Object-Based Approach Using Multi-Temporal Landsat Data, a Random Forest Classifier, and Decision Rules

It is essential to produce land cover maps and land use maps separately for different purposes. This study was conducted to generate such maps in Binh Duong province, Vietnam, using a novel combination of pixel-based and object-based classification techniques and geographic information system (GIS) analysis on multi-temporal Landsat images. Firstly, the connection between land cover and land use was identified; thereafter, the land cover map and land use function regions were extracted with a random forest classifier. Finally, a land use map was generated by combining the land cover map and the land use function regions in a set of decision rules. The results showed that land cover and land use were linked by spectral, spatial, and temporal characteristics, and this helped effectively convert the land cover map into a land use map. The final land cover map attained an overall accuracy (OA) = 93.86%, with producer’s accuracy (PA) and user’s accuracy (UA) of its classes ranging from 73.91% to 100%. Meanwhile, the final land use map achieved OA = 93.45%, and the UA and PA ranged from 84% to 100%. The study demonstrated that it is possible to create high-accuracy maps based entirely on free multi-temporal satellite imagery that promote the reproducibility and proactivity of the research as well as cost-efficiency and time savings.

Machine Learning-Based Processing Proof-of-Concept Pipeline for Semi-Automatic Sentinel-2 Imagery Download, Cloudiness Filtering, Classifications, and Updates of Open Land Use/Land Cover Datasets

Land use and land cover are continuously changing in today’s world. Both domains, therefore, have to rely on updates of external information sources from which the relevant land use/land cover (classification) is extracted. Satellite images are frequent candidates due to their temporal and spatial resolution. On the contrary, the extraction of relevant land use/land cover information is demanding in terms of knowledge base and time. The presented approach offers a proof-of-concept machine-learning pipeline that takes care of the entire complex process in the following manner. The relevant Sentinel-2 images are obtained through the pipeline. Later, cloud masking is performed, including the linear interpolation of merged-feature time frames. Subsequently, four-dimensional arrays are created with all potential training data to become a basis for estimators from the scikit-learn library; the LightGBM estimator is then used. Finally, the classified content is applied to the open land use and open land cover databases. The verification of the provided experiment was conducted against detailed cadastral data, to which Shannon’s entropy was applied since the number of cadaster information classes was naturally consistent. The experiment showed a good overall accuracy (OA) of 85.9%. It yielded a classified land use/land cover map of the study area consisting of 7188 km2 in the southern part of the South Moravian Region in the Czech Republic. The developed proof-of-concept machine-learning pipeline is replicable to any other area of interest so far as the requirements for input data are met.

Monitoring oil spill in Norilsk, Russia using satellite data

This paper studies the oil spill, which occurred in the Norilsk and Taimyr region of Russia due to the collapse of the fuel tank at the power station on May 29, 2020. We monitored the snow, ice, water, vegetation and wetland of the region using data from the Multi-Spectral Instruments (MSI) of Sentinel-2 satellite. We analyzed the spectral band absorptions of Sentinel-2 data acquired before, during and after the incident, developed true and false-color composites (FCC), decorrelated spectral bands and used the indices, i.e. Snow Water Index (SWI), Normalized Difference Water Index (NDWI) and Normalized Difference Vegetation Index (NDVI). The results of decorrelated spectral bands 3, 8, and 11 of Sentinel-2 well confirmed the results of SWI, NDWI, NDVI, and FCC images showing the intensive snow and ice melt between May 21 and 31, 2020. We used Sentinel-2 results, field photographs, analysis of the 1980-2020 daily air temperature and precipitation data, permafrost observations and modeling to explore the hypothesis that either the long-term dynamics of the frozen ground, changing climate and environmental factors, or abnormal weather conditions may have caused or contributed to the collapse of the oil tank.

RUESVMs: An Ensemble Method to Handle the Class Imbalance Problem in Land Cover Mapping Using Google Earth Engine

Timely and accurate Land Cover (LC) information is required for various applications, such as climate change analysis and sustainable development. Although machine learning algorithms are most likely successful in LC mapping tasks, the class imbalance problem is known as a common challenge in this regard. This problem occurs during the training phase and reduces classification accuracy for infrequent and rare LC classes. To address this issue, this study proposes a new method by integrating random under-sampling of majority classes and an ensemble of Support Vector Machines, namely Random Under-sampling Ensemble of Support Vector Machines (RUESVMs). The performance of RUESVMs for LC classification was evaluated in Google Earth Engine (GEE) over two different case studies using Sentinel-2 time-series data and five well-known spectral indices, including the Normalized Difference Vegetation Index (NDVI), Green Normalized Difference Vegetation Index (GNDVI), Soil-Adjusted Vegetation Index (SAVI), Normalized Difference Built-up Index (NDBI), and Normalized Difference Water Index (NDWI). The performance of RUESVMs was also compared with the traditional SVM and combination of SVM with three benchmark data balancing techniques namely the Random Over-Sampling (ROS), Random Under-Sampling (RUS), and Synthetic Minority Over-sampling Technique (SMOTE). It was observed that the proposed method considerably improved the accuracy of LC classification, especially for the minority classes. After adopting RUESVMs, the overall accuracy of the generated LC map increased by approximately 4.95 percentage points, and this amount for the geometric mean of producer’s accuracies was almost 3.75 percentage points, in comparison to the most accurate data balancing method (i.e., SVM-SMOTE). Regarding the geometric mean of users’ accuracies, RUESVMs also outperformed the SVM-SMOTE method with an average increase of 6.45 percentage points.