Comparison of Machine Learning Methods Applied to SAR Images for Forest Classification in Mediterranean Areas

Spatial Autocorrelation Analysis of Land Use and Ecosystem Service Value in the Huangshui River Basin at the Grid Scale

The Huangshui River Basin is one of the most densely populated areas on the Qinghai–Tibet Plateau and is characterized by a high level of human activity. The contradiction between ecological protection and socioeconomic development has become increasingly prominent; determining how to achieve the balanced and coordinated development of the Huangshui River Basin is an important task. Thus, this study used the Google Earth Engine (GEE) cloud-computing platform and Sentinel-1/2 data, supplemented with an ALOS digital elevation model (ALOS DEM) and field survey data, and combined a remote sensing classification method, grid method, and ecosystem service value (ESV) evaluation method to study the spatial correlation and interaction between land use (LU) and ESV in the Huangshui River Basin. The following results were obtained: (1) on the GEE platform, Sentinel-1/2 active and passive remote sensing data, combined with the gradient tree-boosting algorithm, can efficiently produce highly accurate LU data with a spatial resolution of 10 m in the Huangshui River Basin; the overall accuracy (OA) reached 88%. (2) The total ESV in the Huangshui River Basin in 2020 was CNY 33.18 billion (USD 4867.2 million), of which woodland and grassland were the main contributors to ESV. In the Huangshui River Basin, the LU type, LU degree, and ESV have significant positive spatial correlations, with urban and agricultural areas showing an H-H agglomeration in terms of LU degree, with woodlands, grasslands, reservoirs, and wetlands showing an H-H agglomeration in terms of ESV. (3) There is a significant negative spatial correlation between the LU degree and ESV in the Huangshui River Basin, indicating that the enhancement of the LU degree in the basin could have a negative spatial spillover effect on the ESV of surrounding areas. Thus, green development should be the future direction of progress in the Huangshui River Basin, i.e., while maintaining and expanding the land for ecological protection and restoration, and the LU structure should be actively adjusted to ensure ecological security and coordinated and sustainable socioeconomic development in the Basin.

Preliminary Classification of Selected Farmland Habitats in Ireland Using Deep Neural Networks

Ireland has a wide variety of farmlands that includes arable fields, grassland, hedgerows, streams, lakes, rivers, and native woodlands. Traditional methods of habitat identification rely on field surveys, which are resource intensive, therefore there is a strong need for digital methods to improve the speed and efficiency of identification and differentiation of farmland habitats. This is challenging because of the large number of subcategories having nearly indistinguishable features within the habitat classes. Heterogeneity among sites within the same habitat class is another problem. Therefore, this research work presents a preliminary technique for accurate farmland classification using stacked ensemble deep convolutional neural networks (DNNs). The proposed approach has been validated on a high-resolution dataset collected using drones. The image samples were manually labelled by the experts in the area before providing them to the DNNs for training purposes. Three pre-trained DNNs customized using the transfer learning approach are used as the base learners. The predicted features derived from the base learners were then used to train a DNN based meta-learner to achieve high classification rates. We analyse the obtained results in terms of convergence rate, confusion matrices, and ROC curves. This is a preliminary work and further research is needed to establish a standard technique.

Assessing Forest Type and Tree Species Classification Using Sentinel-1 C-Band SAR Data in Southern Sweden

The multitemporal acquisition of images from the Sentinel-1 satellites allows continuous monitoring of a forest. This study focuses on the use of multitemporal C-band synthetic aperture radar (SAR) data to assess the results for forest type (FTY), between coniferous and deciduous forest, and tree species (SPP) classification. We also investigated the temporal stability through the use of backscatter from multiple seasons and years of acquisition. SAR acquisitions were pre-processed, histogram-matched, smoothed, and temperature-corrected. The normalized average backscatter was extracted for interpreted plots and used to train Random Forest models. The classification results were then validated with field plots. A principal component analysis was tested to reduce the dimensionality of the explanatory variables, which generally improved the results. Overall, the FTY classifications were promising, with higher accuracies (OA of 0.94 and K = 0.86) than the SPP classification (OA of 0.66 and K = 0.54). The use of merely winter images (OA = 0.89) reached, on average, results that were almost as good as those using of images from the entire year. The use of images from a single winter season reached a similar result (OA = 0.87). We conclude that multiple Sentinel-1 images acquired in winter conditions are feasible to classify forest types in a hemi-boreal Swedish forest.

Semantic segmentation of PolSAR image data using advanced deep learning model

Urban area mapping is an important application of remote sensing which aims at both estimation and change in land cover under the urban area. A major challenge being faced while analyzing Synthetic Aperture Radar (SAR) based remote sensing data is that there is a lot of similarity between highly vegetated urban areas and oriented urban targets with that of actual vegetation. This similarity between some urban areas and vegetation leads to misclassification of the urban area into forest cover. The present work is a precursor study for the dual-frequency L and S-band NASA-ISRO Synthetic Aperture Radar (NISAR) mission and aims at minimizing the misclassification of such highly vegetated and oriented urban targets into vegetation class with the help of deep learning. In this study, three machine learning algorithms Random Forest (RF), K-Nearest Neighbour (KNN), and Support Vector Machine (SVM) have been implemented along with a deep learning model DeepLabv3+ for semantic segmentation of Polarimetric SAR (PolSAR) data. It is a general perception that a large dataset is required for the successful implementation of any deep learning model but in the field of SAR based remote sensing, a major issue is the unavailability of a large benchmark labeled dataset for the implementation of deep learning algorithms from scratch. In current work, it has been shown that a pre-trained deep learning model DeepLabv3+ outperforms the machine learning algorithms for land use and land cover (LULC) classification task even with a small dataset using transfer learning. The highest pixel accuracy of 87.78% and overall pixel accuracy of 85.65% have been achieved with DeepLabv3+ and Random Forest performs best among the machine learning algorithms with overall pixel accuracy of 77.91% while SVM and KNN trail with an overall accuracy of 77.01% and 76.47% respectively. The highest precision of 0.9228 is recorded for the urban class for semantic segmentation task with DeepLabv3+ while machine learning algorithms SVM and RF gave comparable results with a precision of 0.8977 and 0.8958 respectively.

Earth Observation and Biodiversity Big Data for Forest Habitat Types Classification and Mapping

In the light of the “Biological Diversity” concept, habitats are cardinal pieces for biodiversity quantitative estimation at a local and global scale. In Europe EUNIS (European Nature Information System) is a system tool for habitat identification and assessment. Earth Observation (EO) data, which are acquired by satellite sensors, offer new opportunities for environmental sciences and they are revolutionizing the methodologies applied. These are providing unprecedented insights for habitat monitoring and for evaluating the Sustainable Development Goals (SDGs) indicators. This paper shows the results of a novel approach for a spatially explicit habitat mapping in Italy at a national scale, using a supervised machine learning model (SMLM), through the combination of vegetation plot database (as response variable), and both spectral and environmental predictors. The procedure integrates forest habitat data in Italy from the European Vegetation Archive (EVA), with Sentinel-2 imagery processing (vegetation indices time series, spectral indices, and single bands spectral signals) and environmental data variables (i.e., climatic and topographic), to parameterize a Random Forests (RF) classifier. The obtained results classify 24 forest habitats according to the EUNIS III level: 12 broadleaved deciduous (T1), 4 broadleaved evergreen (T2) and eight needleleaved forest habitats (T3), and achieved an overall accuracy of 87% at the EUNIS II level classes (T1, T2, T3), and an overall accuracy of 76.14% at the EUNIS III level. The highest overall accuracy value was obtained for the broadleaved evergreen forest equal to 91%, followed by 76% and 68% for needleleaved and broadleaved deciduous habitat forests, respectively. The results of the proposed methodology open the way to increase the EUNIS habitat categories to be mapped together with their geographical extent, and to test different semi-supervised machine learning algorithms and ensemble modelling methods.

Classification of Nemoral Forests with Fusion of Multi-Temporal Sentinel-1 and 2 Data

Mapping forest extent and forest cover classification are important for the assessment of forest resources in socio-economic as well as ecological terms. Novel developments in the availability of remotely sensed data, computational resources, and advances in areas of statistical learning have enabled the fusion of multi-sensor data, often yielding superior classification results. Most former studies of nemoral forests fusing multi-sensor and multi-temporal data have been limited in spatial extent and typically to a simple classification of landscapes into major land cover classes. We hypothesize that multi-temporal, multi-sensor data will have a specific strength in the further classification of nemoral forest landscapes owing to the distinct seasonal patterns in the phenology of broadleaves. This study aimed to classify the Danish landscape into forest/non-forest and further into forest types (broadleaved/coniferous) and species groups, using a cloud-based approach based on multi-temporal Sentinel 1 and 2 data and a random forest classifier trained with National Forest Inventory (NFI) data. Mapping of non-forest and forest resulted in producer accuracies of 99% and 90%, respectively. The mapping of forest types (broadleaf and conifer) within the forested area resulted in producer accuracies of 95% for conifer and 96% for broadleaf forest. Tree species groups were classified with producer accuracies ranging 34–74%. Species groups with coniferous species were the least confused, whereas the broadleaf groups, especially Quercus species, had higher error rates. The results are applied in Danish national accounting of greenhouse gas emissions from forests, resource assessment, and assessment of forest biodiversity potentials.

Fusion and Correction of Multi-Source Land Cover Products Based on Spatial Detection and Uncertainty Reasoning Methods in Central Asia

Land cover products are an indispensable data source in land surface process research, and their accuracy directly affects the reliability of related research. Due to the differences in factors such as satellite sensors, the temporal–spatial resolution of remote sensing images, and landcover interpretation technologies, various recently released land cover products are inconsistent, and their accuracy is usually insufficient to meet application requirements. This study, therefore, established a fusion and correction method for multi-source landcover products by combining them with landcover statistics from the Food and Agriculture Organization of the United Nations (FAO), introducing a spatial consistency discrimination technique, and applying an improved Dempster-Shafer evidence fusion method. The five countries in Central Asia were used for a method application and verification assessment. The nine products selected (CCI-LC, CGLS, FROM-GLC, GLCNMO, MCD12Q, GFSAD30, PALSAR, GSWD, and GHS-BUILT) were consistent in time and covered the study area. Based on the interpretation of 1437 high-definition image verification areas, the overall accuracy of the fusion landcover result was 85.32%, and the kappa coefficient was 0.80, which was better than that of the existing comprehensive products. The spatial consistency fusion method had the advantage of an improved statistical fitting, with an overall similarity statistic of 0.999. The improved Dempster-Shafer evidence theory fusion method had an accuracy that was 4.86% higher than the spatial consistency method, and the kappa coefficient increased by 0.07. Combining these two methods improved the consistency of the multi-source data fusion and correction method established in this paper and will also provide more reliable basic data for future research in Central Asia.

Comparative Analysis of Edge Information and Polarization on SAR-to-Optical Translation Based on Conditional Generative Adversarial Networks

To accurately describe dynamic vegetation changes, high temporal and spectral resolution data are urgently required. Optical images contain rich spectral information but are limited by poor weather conditions and cloud contamination. Conversely, synthetic-aperture radar (SAR) is effective under all weather conditions but contains insufficient spectral information to recognize certain vegetation changes. Conditional adversarial networks (cGANs) can be adopted to transform SAR images (Sentinel-1) into optical images (Landsat8), which exploits the advantages of both optical and SAR images. As the features of SAR and optical remote sensing data play a decisive role in the translation process, this study explores the quantitative impact of edge information and polarization (VV, VH, VV&VH) on the peak signal-to-noise ratio, structural similarity index measure, correlation coefficient (r), and root mean squared error. The addition of edge information improves the structural similarity between generated and real images. Moreover, using the VH and VV&VH polarization modes as the input provides the cGANs with more effective information and results in better image quality. The optimal polarization mode with the addition of edge information is VV&VH, whereas that without edge information is VV. Near-infrared and short-wave infrared bands in the generated image exhibit higher accuracy (r > 0.8) than visible light bands. The conclusions of this study could serve as an important reference for selecting cGANs input features, and as a potential reference for the applications of cGANs to the SAR-to-optical translation of other multi-source remote sensing data.

Modified Red Blue Vegetation Index for Chlorophyll Estimation and Yield Prediction of Maize from Visible Images Captured by UAV

The vegetation index (VI) has been successfully used to monitor the growth and to predict the yield of agricultural crops. In this paper, a long-term observation was conducted for the yield prediction of maize using an unmanned aerial vehicle (UAV) and estimations of chlorophyll contents using SPAD-502. A new vegetation index termed as modified red blue VI (MRBVI) was developed to monitor the growth and to predict the yields of maize by establishing relationships between MRBVI- and SPAD-502-based chlorophyll contents. The coefficients of determination (R2s) were 0.462 and 0.570 in chlorophyll contents' estimations and yield predictions using MRBVI, and the results were relatively better than the results from the seven other commonly used VI approaches. All VIs during the different growth stages of maize were calculated and compared with the measured values of chlorophyll contents directly, and the relative error (RE) of MRBVI is the lowest at 0.355. Further, machine learning (ML) methods such as the backpropagation neural network model (BP), support vector machine (SVM), random forest (RF), and extreme learning machine (ELM) were adopted for predicting the yields of maize. All VIs calculated for each image captured during important phenological stages of maize were set as independent variables and the corresponding yields of each plot were defined as dependent variables. The ML models used the leave one out method (LOO), where the root mean square errors (RMSEs) were 2.157, 1.099, 1.146, and 1.698 (g/hundred grain weight) for BP, SVM, RF, and ELM. The mean absolute errors (MAEs) were 1.739, 0.886, 0.925, and 1.356 (g/hundred grain weight) for BP, SVM, RF, and ELM, respectively. Thus, the SVM method performed better in predicting the yields of maize than the other ML methods. Therefore, it is strongly suggested that the MRBVI calculated from images acquired at different growth stages integrated with advanced ML methods should be used for agricultural- and ecological-related chlorophyll estimation and yield predictions.

Near Real-Time Automated Early Mapping of the Perimeter of Large Forest Fires from the Aggregation of VIIRS and MODIS Active Fires in Mexico

In contrast with current operational products of burned area, which are generally available one month after the fire, active fires are readily available, with potential application for early evaluation of approximate fire perimeters to support fire management decision making in near real time. While previous coarse-scale studies have focused on relating the number of active fires to a burned area, some local-scale studies have proposed the spatial aggregation of active fires to directly obtain early estimate perimeters from active fires. Nevertheless, further analysis of this latter technique, including the definition of aggregation distance and large-scale testing, is still required. There is a need for studies that evaluate the potential of active fire aggregation for rapid initial fire perimeter delineation, particularly taking advantage of the improved spatial resolution of the Visible Infrared Imaging Radiometer (VIIRS) 375 m, over large areas and long periods of study. The current study tested the use of convex hull algorithms for deriving coarse-scale perimeters from Moderate Resolution Imaging Spectroradiometer (MODIS) and Visible Infrared Imaging Radiometer Suite (VIIRS) active fire detections, compared against the mapped perimeter of the MODIS collection 6 (MCD64A1) burned area. We analyzed the effect of aggregation distance (750, 1000, 1125 and 1500 m) on the relationships of active fire perimeters with MCD64A1, for both individual fire perimeter prediction and total burned area estimation, for the period 2012–2108 in Mexico. The aggregation of active fire detections from MODIS and VIIRS demonstrated a potential to offer coarse-scale early estimates of the perimeters of large fires, which can be available to support fire monitoring and management in near real time. Total burned area predicted from aggregated active fires followed the same temporal behavior as the standard MCD64A1 burned area, with potential to also account for the role of smaller fires detected by the thermal anomalies. The proposed methodology, based on easily available algorithms of point aggregation, is susceptible to be utilized both for near real-time and historical fire perimeter evaluation elsewhere. Future studies might test active fires aggregation between regions or biomes with contrasting fuel characteristics and human activity patterns against medium resolution (e.g., Landsat and Sentinel) fire perimeters. Furthermore, coarse-scale active fire perimeters might be utilized to locate areas where such higher-resolution imagery can be downloaded to improve the evaluation of fire extent and impact.

Machine Learning Classification Ensemble of Multitemporal Sentinel-2 Images: The Case of a Mixed Mediterranean Ecosystem

Land cover type classification still remains an active research topic while new sensors and methods become available. Applications such as environmental monitoring, natural resource management, and change detection require more accurate, detailed, and constantly updated land-cover type mapping. These needs are fulfilled by newer sensors with high spatial and spectral resolution along with modern data processing algorithms. Sentinel-2 sensor provides data with high spatial, spectral, and temporal resolution for the in classification of highly fragmented landscape. This study applies six traditional data classifiers and nine ensemble methods on multitemporal Sentinel-2 image datasets for identifying land cover types in the heterogeneous Mediterranean landscape of Lesvos Island, Greece. Support vector machine, random forest, artificial neural network, decision tree, linear discriminant analysis, and k-nearest neighbor classifiers are applied and compared with nine ensemble classifiers on the basis of different voting methods. kappa statistic, F1-score, and Matthews correlation coefficient metrics were used in the assembly of the voting methods. Support vector machine outperformed the base classifiers with kappa of 0.91. Support vector machine also outperformed the ensemble classifiers in an unseen dataset. Five voting methods performed better than the rest of the classifiers. A diversity study based on four different metrics revealed that an ensemble can be avoided if a base classifier shows an identifiable superiority. Therefore, ensemble approaches should include a careful selection of base-classifiers based on a diversity analysis.

Comparative Assessment of Machine Learning Methods for Urban Vegetation Mapping Using Multitemporal Sentinel-1 Imagery

Mapping of green vegetation in urban areas using remote sensing techniques can be used as a tool for integrated spatial planning to deal with urban challenges. In this context, multitemporal (MT) synthetic aperture radar (SAR) data have not been equally investigated, as compared to optical satellite data. This research compared various machine learning methods using single-date and MT Sentinel-1 (S1) imagery. The research was focused on vegetation mapping in urban areas across Europe. Urban vegetation was classified using six classifiers—random forests (RF), support vector machine (SVM), extreme gradient boosting (XGB), multi-layer perceptron (MLP), AdaBoost.M1 (AB), and extreme learning machine (ELM). Whereas, SVM showed the best performance in the single-date image analysis, the MLP classifier yielded the highest overall accuracy in the MT classification scenario. Mean overall accuracy (OA) values for all machine learning methods increased from 57% to 77% with speckle filtering. Using MT SAR data, i.e., three and five S1 imagery, an additional increase in the OA of 8.59% and 13.66% occurred, respectively. Additionally, using three and five S1 imagery for classification, the F1 measure for forest and low vegetation land-cover class exceeded 90%. This research allowed us to confirm the possibility of MT C-band SAR imagery for urban vegetation mapping.