Remote Sensing Land Cover Research Articles

A Hierarchical Coarse–Fine Adaptive Fusion Network for the Joint Classification of Hyperspectral and LiDAR Data

Hyperspectral image (HSI) and light detection and ranging (LiDAR) data have gained significant attention due to their excellent complementarity, which can be combined to improve the accuracy of land cover classification. However, there are still many challenges, such as unrelated physical characteristics, different data structures, and a lack of labeled samples. Many methods fail to leverage the full potential of multi-source data, particularly hierarchical complementary information. To address these problems, a hierarchical coarse–fine adaptive (HCFA) fusion network with dynamic convolution and a transformer is proposed for multi-source remote sensing land cover classification. Although fusing hierarchical information can improve the classification accuracy of the model, improper hierarchical feature selection and optimization may negatively affect the classification results. Therefore, a coarse–fine mutual learning strategy is proposed to dynamically fuse hierarchical information. Additionally, the disparity between multi-source data continues to prevent the realization of effective fusion. To tackle this challenge, cross-tokenization and cross-token attention are implemented to enhance information interaction. Furthermore, to improve the model representation with limited computational cost, we combine the advantages of dynamic convolution with a transformer. Validation on three standard datasets demonstrates that HCFA achieves high accuracy with just 1% of the training set while maintaining low computational costs.

Investigating the effects of spatial scales on social vulnerability index: A hybrid uncertainty and sensitivity analysis approach combined with remote sensing land cover data.

Investigating the effects of spatial scales on the uncertainty and sensitivity analysis of the social vulnerability index (SoVI) model output is critical, especially for spatial scales finer than the census block group or census block. This study applied the intelligent dasymetric mapping approach to spatially disaggregate the census tract scale SoVI model into a 300-m grids resolution SoVI map in Davidson County, Nashville. Then, uncertainty analysis and variance-based global sensitivity analysis were conducted on two scales of SoVI models: (a) census tract scale; (b) 300-m grids scale. Uncertainty analysis results indicate that the SoVI model has better confidence in identifying places with a higher socially vulnerable status, no matter the spatial scales in which the SoVI is constructed. However, the spatial scale of SoVI does affect the sensitivity analysis results. The sensitivity analysis suggests that for census tract scale SoVI, the indicator transformation and weighting scheme are the two major uncertainty contributors in the SoVI index modeling stages. While for finer spatial scales like the 300-m grid's resolution, the weighting scheme becomes the uttermost dominant uncertainty contributor, absorbing uncertainty contributions from indicator transformation.

Spatial distribution of wildfire threat in the far north: exposure assessment in boreal communities

Increased wildfire activity has raised concerns among communities about how to assess and prepare for this threat. There is a need for wildfire hazard assessment approaches that capture local variability to inform decisions, produce results understood by the public, and are updatable in a timely manner. We modified an existing approach to assess decadal wildfire hazards based primarily on ember dispersal and wildfire proximity, referencing landscape changes from 1984 through 2014. Our modifications created a categorical flammability hazard scheme, rather than dichotomous, and integrated wildfire exposure results across spatial scales. We used remote sensed land cover from four historical decadal points to create flammability hazard and wildfire exposure maps for three arctic communities (Anchorage and Fairbanks, Alaska and Whitehorse, Yukon). Within the Fairbanks study area, we compared 2014 flammability hazard, wildfire exposure, and FlamMap burn probabilities among burned (2014–2023) and unburned areas. Unlike burn probabilities, there were significantly higher in exposure values among burned and unburned locations (Wilcoxon; p < 0.001) and exposure rose as flammability hazard classes increased (Kruskal–Wallis; p < 0.001). Very high flammability hazard class supported 75% of burned areas and burns tended to occur in areas with 60% exposure or greater. Areas with high exposure values are more prone to burn and thus desirable for mitigation actions. By working with wildfire practitioners and communities, we created a tool that rapidly assesses wildfire hazards and is easily modified to help identify and prioritize mitigation activities.

Consistency Analysis of Multi-Source Remote Sensing Land Cover Products in Arid Regions—A Case Study of Xinjiang

Arid regions are considered to be among the most ecologically fragile and highly sensitive to environmental change globally, and land use and land cover conditions in the region directly influence large-scale ecosystem processes. Currently, thanks to diverse remote sensing platforms, geographers have developed an array of land cover products. However, there are differences between these products due to variations in spatio-temporal resolutions. In this context, assessing the accuracy and consistency of different land cover products is crucial for rationalizing the selection of land cover products to study global or regional environmental changes. In this study, Xinjiang Uygur Autonomous Region (XUAR) is taken as the study area, and the consistency and performance (type area deviation, spatial consistency, accuracy assessment, and other indexes) of the five land cover products (GlobeLand30, FROM_GLC30, CLCD, GLC_FCS30, and ESRI) were compared and analyzed. The results of the study show that (1) the GlobeLand30 product has the highest overall accuracy in the study area, with an overall accuracy of 84.06%, followed by ESA with 75.57%, while CLCD has the lowest overall accuracy of 70.05%. (2) The consistency between GlobeLand30 and CLCD (area correlation coefficient of 0.99) was higher than that among the other products. (3) Among the five products, the highest consistency was found for water bodies and permanent snow and ice, followed by bare land. In contrast, the consistency of these five products for grassland and forest was relatively low. (4) The full-consistency area accounts for 49.01% of the total study area. They were mainly distributed in areas with relatively homogeneous land cover types, such as the north and south of the Tianshan Mountains, which are dominated by bare land and cropland. In contrast, areas of inconsistency make up only 0.03% and are mostly found in heterogeneous areas, like the transitional zones with mixed land cover types in the Altai Mountains and Tianshan Mountains, or in areas with complex terrain. In terms of meeting practical user needs, GlobeLand30 offers the best comprehensive performance. GLC_FCS30 is more suitable for studies related to forests, while FROM_GLC30 and ESRI demonstrate greater advantages in identifying permanent ice and snow, whereas the performance of CLCD is generally average.

Evaluation of different machine learning algorithms for LULC classification in heterogeneous landscape by using remote sensing and GIS techniques

Remote sensing land cover classification plays a crucial role in detecting changes, urbanization planning, mapping and monitoring land cover on earth surface. It is very challenging to get accurate result in remote sensing data because different classifiers are very much area dependent. Different classifiers such as decision tree (DT), K-nearest neighbor (KNN), artificial neural network (ANN), support vector machine (SVM), boosted decision tree (BDT), random forest (RF), classification and regression tree (CART) and maximum likelihood classifiers (MLC), have different accuracies for the same classes. Several studies have utilized remote sensing and GIS tools to investigate changes in land use and land cover (LULC) using different classifiers. Seasonal rivers which should be classified as water bodies are mostly classified as urban area with the conventional land cover classification schemes because spectral reflectance of these river bodies is similar to urban area due to stones present in their river bed. It is very difficult to distinguish between these river beds (which are mostly found in various districts of Uttarakhand, India) and urban area in remote sensing images. In this paper, we present a new method to distinguish these river beds with the urban area and to separate other classes easily. First of all different spectral indices such as NDVI, NDBI, EVI and MNDWI are extracted from a high resolution Sentinel-2 MSI image then these indices are integrated with Sentinel-2 MSI image for classification of different land cover classes by using four different machine learning classifiers such as RF, SVM, DT and CART. The obtained results confirm the performance strength of the suggested method as there is much improvement in accuracy.

Characterizing Uncertainty and Enhancing Utility in Remotely Sensed Land Cover Using Error Matrices Localized in Canonical Correspondence Analysis Ordination Space

In response to uncertainty in remotely sensed land cover products, there is continuing research on accuracy assessment and analysis. Given reference sample data, accuracy indicators are commonly estimated based on error matrices, from which areal extents of different cover types are also estimated. There are merits to explore the ways utilities of land cover products may be further enhanced beyond map face values and conventional area estimation. This paper presents an integrative method (CCAErrMat) for uncertainty characterization and utility enhancement. This works through reference-map cover type co-occurrence analyses based on error matrices localized in canonical correspondence analysis (CCA) ordination space rather than in geographic space to overcome the sparsity of reference sample data. The aforementioned co-occurrence analyses facilitate quantification of accuracy indicators, identification of correctly classified and perfectly misclassified pixels, and prediction of reference class probabilities, all at individual pixels. Moreover, these predicted reference class probabilities are used as auxiliary variables to formulate model-assisted area estimation, further enhancing map utilities. Extensions to CCAErrMat are also investigated as a way to bypass the pre-computing of map class occurrence pattern indices as candidate explanatory variables for CCAErrMat, leading to two variant methods: CCACCAErrMat and CNNCCAErrMat. A case study based in Wuhan municipality, central China was undertaken to compare the proposed method against alternative methods, including CCA-separate and CNN-separate. The advantages of CCAErrMat and CCACCAErrMat were confirmed. The proposed method is recommendable for characterizing uncertainty and enhancing utilities in land cover maps by analyzing locally constrained error matrices. The method is also cost-effective in terms of reference sample data, as requirements for them are similar to those for conventional accuracy assessments.

Continuous Long Time Series Monitoring of Urban Construction Land in Supporting the SDG 11.3.1—A Case Study of Nanning, Guangxi, China

Enhancing the spatio-temporal dynamic monitoring of the construction land of emerging major cities provides the basis for understanding the effects of human activities on the earth’s ecological environment changes, which is regarded as an important way to improve human wellbeing and to achieve the UN SDGs. This paper, which uses Nanning as an example, adopts 30 m annual remote sensing land cover datasets from 1990–2021 to elaborately analyze the temporal evolution and spatial expansion of construction land and to evaluate the sustainable development level based on SDG 11.3.1. The conclusions of this paper are as follows: (1) For Nanning City, during 1990–2021, construction land area increased from 54.77 km2 to 326.33 km2—a nearly five-fold increase. Spreading expansion and finger-like expansion along roads played a leading role during the development of the construction land in Nanning. Liangqing and Yongning are the future directions of urban development. The LCRPGR of Nanning declined with one peak period around 1995, and the LCRPGR was relatively stable with a high intensity of land use. (2) For the other emerging major cities, continuous long duration data are more suitable for urban construction land monitoring compared with traditional sparse time-series monitoring. The rich information derived from continuous long duration data can help decision makers to formulate and implement more comprehensive measures. The research paradigm adopted in this study can be applied in other cities. The newly developed urban districts will have foreseeable high sustainable development risks, and urban development strategies at the international, national, and urban levels could reduce those risks and promote the realization of UN SDGs from different angles and with different intensities.

RSSFormer: Foreground Saliency Enhancement for Remote Sensing Land-Cover Segmentation.

High spatial resolution (HSR) remote sensing images contain complex foreground-background relationships, which makes the remote sensing land cover segmentation a special semantic segmentation task. The main challenges come from the large-scale variation, complex background samples and imbalanced foreground-background distribution. These issues make recent context modeling methods sub-optimal due to the lack of foreground saliency modeling. To handle these problems, we propose a Remote Sensing Segmentation framework (RSSFormer), including Adaptive TransFormer Fusion Module, Detail-aware Attention Layer and Foreground Saliency Guided Loss. Specifically, from the perspective of relation-based foreground saliency modeling, our Adaptive Transformer Fusion Module can adaptively suppress background noise and enhance object saliency when fusing multi-scale features. Then our Detail-aware Attention Layer extracts the detail and foreground-related information via the interplay of spatial attention and channel attention, which further enhances the foreground saliency. From the perspective of optimization-based foreground saliency modeling, our Foreground Saliency Guided Loss can guide the network to focus on hard samples with low foreground saliency responses to achieve balanced optimization. Experimental results on LoveDA datasets, Vaihingen datasets, Potsdam datasets and iSAID datasets validate that our method outperforms existing general semantic segmentation methods and remote sensing segmentation methods, and achieves a good compromise between computational overhead and accuracy. Our code is available at https://github.com/Rongtao-Xu/RepresentationLearning/tree/main/RSSFormer-TIP2023.

Influence of surrounding land cover on marsh-breeding birds: Implications for wetland restoration and conservation planning

Marsh-breeding birds are valuable components of healthy ecosystems and are useful indicators of successful wetland restorations. The occurrence of these species, however, is influenced by the surrounding landscape. To aid decision-makers, we used data from the Great Lakes Marsh Monitoring Program of Birds Canada at 521 sites across four time periods (2000–2001, 2005–2006, 2010–2011, 2015–2016) throughout southern Ontario, Canada, to quantify how initial occupancy and subsequent local extinction or colonization of 13 marsh-breeding bird indicator species (or species groups) was influenced by the composition of, and concurrent changes in, surrounding remotely-sensed land cover within 200–6,400 m. For six species (46 %) initial occupancy was higher or extinction was lower where surrounding wetland/open water land cover was higher. By contrast, initial occupancy was lower, extinction was higher, and/or colonization was lower: 1) where surrounding anthropogenic (predominantly urban) land cover was higher (four species, 31 %), 2) in Great Lakes coastal compared to inland landscapes (five species, 38 %), and 3) where loss of surrounding wetland/open water land cover increased, depending on whether surrounding wetland/open water land cover was initially low or high (seven species, 54 %). We recommend that decision-makers consider the influence of the surrounding landscape during conservation planning and when measuring success of wetland restorations based on marsh-breeding bird indicator species, particularly in Great Lakes coastal landscapes and landscapes with high urban land cover and/or low wetland/open water land cover.

Landscape Pattern Consistency Assessment of 10 m Land Cover Products in Different Ecological Zoning Contexts of Sichuan Province, China

The research on land cover and its changing value to the ecology environment and ecosystem service is of great importance. Understanding the landscape patterns and accuracy in remote sensing land cover data from multiple sources under eco-regionalization is important for relevant research under eco-regionalization. Land cover and land use in different ecological regionalization contexts influence massive ecosystem processes on a global scale, and many ecosystem models are dependent on accurate land cover information. It is, therefore, important to assess the available global land cover products based on different ecological zoning contexts and to understand the differences between them so that different researchers can apply them in a rational way. This study took Sichuan Province as an experimental case. Based on the three methods of spatial superposition, landscape index, and validation sample, we discussed and then analyzed the consistency of landscape patterns for the three 10 m global land cover data under different ecogeographic zones. The results showed that the spatial consistency of FROM-GLC, ESA, and ESRI land cover data were high under the ecological zoning of Palearctic0518 (PA0518) and PA1017, while the spatial pattern was less consistent for the three land cover data under the ecological zoning of PA0509 and PA0437. The fully consistent areas for the three data were 44,420.9 km2 and 53,368.9 km2, respectively. The results of the quantitative analysis of the landscape index showed significant differences in the degree of landscape fragmentation, landscape shape complexity, and the connectivity among landscape patches of several land cover data were significantly different under different ecological zones. Based on the three kinds of independent validation samples to achieve the absolute accuracy of evaluation, the overall accuracy of the FROM-GLC, ESA, and ESRI land cover data was less than 60%, and future drawing still needs to further improve the regional land cover data mapping accuracy under different ecological zones in order to carry out ecological environment monitoring, land ecological security evaluation, and related research to provide a reference.

MSANet: an improved semantic segmentation method using multi-scale attention for remote sensing images

ABSTRACT With the development of deep learning technology in the field of computer vision, the land cover classification of remote sensing images has gradually developed. In this paper, a deep learning algorithm for land cover classification of remote sensing images is proposed. Two dual-attention mechanism modules with multi-scale spatial attention and channel attention are designed and integrated into the DeepLabv3+ network, called multi-scale convolutional attention network (MCAN) and multi-scale atrous convolutional attention network (MACAN). Besides, Inception-Squeeze-and-Excitation network (ISNet) is added. So, the network can focus on the parts of remote sensing images that need attention at different scales, and it can improve the accuracy of remote sensing semantic segmentation. The experimental results show that the optimized network can achieve a Mean Intersection over Union (mIoU) of 74.03% on the DeepGlobe datasets, which is 14.55%, 13.47%, and 5.73% higher than that of UNet, SegNet, and DeepLabv3+. It also achieved good results on Vaihingen datasets. This result indicates that the optimized network can effectively improve the accuracy of remote sensing land cover classification.

Spatial machine-learning model diagnostics: a model-agnostic distance-based approach

While significant progress has been made towards explaining black-box machine-learning (ML) models, there is still a distinct lack of diagnostic tools that elucidate the spatial behaviour of ML models in terms of predictive skill and variable importance. This contribution proposes spatial prediction error profiles (SPEPs) and spatial variable importance profiles (SVIPs) as novel model-agnostic assessment and interpretation tools for spatial prediction models with a focus on prediction distance. Their suitability is demonstrated in two case studies representing a regionalization task in an environmental-science context, and a classification task from remotely-sensed land cover classification. In these case studies, the SPEPs and SVIPs of geostatistical methods, linear models, random forest, and hybrid algorithms show striking differences but also relevant similarities. Limitations of related cross-validation techniques are outlined, and the case is made that modelers should focus their model assessment and interpretation on the intended spatial prediction horizon. The range of autocorrelation, in contrast, is not a suitable criterion for defining spatial cross-validation test sets. The novel diagnostic tools enrich the toolkit of spatial data science and may improve ML model interpretation, selection, and design.

Phytoplankton bloom stages estimated from chlorophyll pigment proportions suggest delayed summer production in low sea ice years in the northern Bering Sea.

Decreased sea ice cover in the northern Bering Sea has altered annual phytoplankton phenology owing to an expansion of open water duration and its impact on ocean stratification. Limitations of satellite remote sensing such as the inability to detect bloom activity throughout the water column, under ice, and in cloudy conditions dictate the need for shipboard based measurements to provide more information on bloom dynamics. In this study, we adapted remote sensing land cover classification techniques to provide a new means to determine bloom stage from shipboard samples. Specifically, we used multiyear satellite time series of chlorophyll a to determine whether in-situ blooms were actively growing or mature (i.e., past-peak) at the time of field sampling. Field observations of chlorophyll a and pheophytin (degraded and oxidized chlorophyll products) were used to calculate pheophytin proportions, i.e., (Pheophytin/(Chlorophyll a + Pheophytin)) and empirically determine whether the bloom was growing or mature based on remotely sensed bloom stages. Data collected at 13 north Bering Sea stations each July from 2013–2019 supported a pheophytin proportion of 28% as the best empirical threshold to distinguish a growing vs. mature bloom stage. One outcome was that low vs. high sea ice years resulted in significantly different pheophytin proportions in July; in years with low winter-to-spring ice, more blooms with growing status were observed, compared to later stage, more mature blooms following springs with abundant seasonal sea ice. The detection of growing blooms in July following low ice years suggests that changes in the timing of the spring bloom triggers cascading effects on mid-summer production.

A novel interval-valued data driven type-2 possibilistic local information c-means clustering for land cover classification

In the process of land cover classification, existing fuzzy clustering is not enough to describe the high-order fuzzy uncertainty, while possibilistic clustering has serious parameter dependence and cluster consistency, which makes them unable to effectively deal with the phenomena of “different objects with the same spectrum” and “different objects with the same spectrum”. Hence, this paper proposes a robust type-2 possibilistic C-means clustering with local information and interval-valued data model for remote sensing land cover classification. Firstly, according to the local neighborhood variance, remote sensing information is modeled as interval-valued data. Secondly, existing possibilistic C-means clustering is modified and an enhanced possibilistic C-means clustering is obtained. Once again, it is used to clustering interval-valued data and a single weighting exponent type-2 possibilistic C-means clustering with double distance measures is constructed. Finally, to further improve the robustness of clustering method, local information is embedded into the objective function of this enhanced type-2 possibilistic C-means clustering and a novel robust possibilistic clustering-related algorithm for remote sensing information classification is proposed. Experimental results show that the proposed algorithm outperforms existing state of the art type-2 clustering-related algorithms, and is of great significance to the interpretation of remote sensing images.

Relation between Climate, Agricultural Land Cover and Bird Richness in a Mediterranean Natura 2000 Wetland

In this study we explore for the first time the combined effect of climate variables and remote sensing land cover indicators on bird richness in a representative wetland location of the Thermaikos gulf Natura 2000 protected area in Northern Greece. In particular the association between bird diversity and climate, as well as remote sensed land cover indices was explored for seven successive seasons using correlation analysis and a Cox-Box transformed multivariate linear model. Three climate variables were tested, mean temperature, rain level and mean relative humidity and three land cover indices, the Normalized difference index (NDVI), the atmospheric resistance vegetation index (ARVI) and an agricultural band combination index (ABCI). Among the environmental drivers explored, temperature, rain levels and ABCI were significantly correlated to bird richness in contrast to NDVI and ARVI which showed a lower correlation, while relative humidity displayed the poorest correlation. Thus, temperature, rain levels and agricultural intensity significantly influenced bird composition. Additionally, the multivariable linear model indicates that temperature, rain levels and ABCI have a statistically significant effect (p&lt;0.05) on bird species richness accounting for 73,02% of data variability. Based on the overall model results and the related 3D contour plot simulation we can conclude that in general, bird species richness increases with an increase in temperature and Rain, as well as with a decrease in agricultural intensity (ABCI). Understanding the factors that can affect bird biodiversity of species is of great importance in ecosystem management and species conservation.

Collaborative Extraction of Paddy Planting Areas with Multi-Source Information Based on Google Earth Engine: A Case Study of Cambodia

High-precision spatial mapping of paddy planting areas is very important for food security risk assessment and agricultural monitoring. Previous studies have mainly been based on multi-source satellite imagery, the fusion of Synthetic Aperture Radar (SAR) with optical data, and the combined use of multi-scale and multi-source sensors. However, there have been few studies on paddy spatial mapping using collaborative multi-source remote sensing product information, especially in tropical regions such as Southeast Asia. Therefore, based on the Google Earth Engine (GEE) platform, in this study, Cambodia, which is dominated by agriculture, was taken as the study area, and an extraction scheme for paddy planting areas was developed from collaborative multi-source information, including multi-source remote sensing images (Sentinel-1 and Sentinel-2), multi-source remote sensing land cover products (GFSAD30SEACE, GLC_FCS30-2015, FROM_GLC2015, SERVIR MEKONG, and GUF), paddy phenology information, and topographic features. Evaluation and analysis of the extraction results and the SERVIR MEKONG and ESACCI-LC paddy products revealed that the accuracy of the paddy planting areas extracted using the proposed method is the highest, with an overall accuracy of 89.90%. The results of the proposed method are better than those of the other products in terms of the outline of the paddy planting areas and the description of the road information. The results of this study provide a reference for future high-precision paddy mapping.

Comparison of Three Ten Meter Land Cover Products in a Drought Region: A Case Study in Northwestern China

The ecological and natural conditions in drought regions are harsh. Water shortages and land desertification are prominent features of these regions. The land cover in these regions has a large impact on global climate change, as well as on ecological protection and construction. To make rational and sustainable use of land resources, it is crucial to quickly grasp the accuracy and spatial distribution differences of multi-source remote sensing land cover products in drought regions. Therefore, taking northwestern China as the study area, in this study, the accuracy and spatial pattern distribution differences of three high-resolution (10 m) land cover products, namely, the Finer Resolution Observation and Monitoring of Global Land Cover (FROM-GLC), European Space Agency (ESA), and Environmental Systems Research Institute (ESRI) products, were compared and analyzed via area composition similarity, spatial pattern consistency, and absolute accuracy assessment for three validation samples. The results show that the FROM-GLC product had the highest overall accuracy, ranging from 53.81% to 73.45%. The ESRI product had the lowest overall accuracy, ranging from 35.90% to 64.16%. The spatial consistencies of the three products were low, accounting for 46.26% of the total area, and they were mostly distributed in a single area (mainly bare land and forest). The low accuracy for grassland, bare land, shrubland, and other vegetation types was the primary reason for the large differences between the three products. Future research should focus on improving the mapping accuracy for these vegetation types. Accuracies for water and cropland of the three products were consistent, and, thus, the FROM-GLC, ESA, and ESRI products can be used as auxiliary data in research related to water resources and cropland resources in drought regions.

A Methodology to Generate Integrated Land Cover Data for Land Surface Model by Improving Dempster-Shafer Theory

Land cover type is a key parameter for simulating surface processes in many land surface models (LSMs). Currently, the widely used global remote sensing land cover products cannot meet the requirements of LSMs for classification systems, physical definition, data accuracy, and space-time resolution. Here, a new fusion method was proposed to generate land cover data for LSMs by fusing multi-source remote sensing land cover data, which was based on improving Dempster-Shafer evidence theory with mathematical models and knowledge rules optimization. The new method has the ability to deal with seriously disagreement information, thereby improving the robustness of the theory. The results showed the new method can reduce the disagreement between input data and realized the conversion of multiple land cover classification systems to into a single land cover classification system. China Fusion Land Cover data (CFLC) in 2015 generated by the new method maintained the classification accuracy of the China land use map (CNLULC), which is based on visual image interpretation and further enriched land cover classes of input data. Compared with Geo-Wiki observations in 2015, the overall accuracy for CFLC is higher than other two global land cover data. Compared with the observations, the 0–10 cm soil moisture simulated by the CFLC in Noah–MP LSM during the growing season in 2014 had better performance than that simulated by initial land cover data and MODIS land cover data. Our new method is highly portable and generalizable to generate higher quality land cover data with a specific land cover classification system for LSMs by fusing multiple land cover data, providing a new approach to land cover mapping for LSMs.

Adopt a Pixel 3 km: A Multiscale Data Set Linking Remotely Sensed Land Cover Imagery With Field Based Citizen Science Observation

Adopt a Pixel 3 km: A Multiscale Data Set Linking Remotely Sensed Land Cover Imagery With Field Based Citizen Science Observation

Deep neural network ensembles for remote sensing land cover and land use classification

ABSTRACT With the advancement of satellite technology, a considerable amount of very high-resolution imagery has become available to be used for the Land Cover and Land Use (LCLU) classification task aiming to categorize remotely sensed images based on their semantic content. Recently, Deep Neural Networks (DNNs) have been widely used for different applications in the field of remote sensing and they have profound impacts; however, improvement of the generalizability and robustness of the DNNs needs to be progressed further to achieve higher accuracy for a variety of sensing geometries and categories. We address this problem by deploying three different Deep Neural Network Ensemble (DNNE) methods and creating a comparative analysis for the LCLU classification task. DNNE enables improvement of the performance of DNNs by ensuring the diversity of the models that are combined. Thus, enhances the generalizability of the models and produces more robust and generalizable outcomes for LCLU classification tasks. The experimental results on NWPU-RESISC45 and AID datasets demonstrate that utilizing the aggregated information from multiple DNNs leads to an increase in classification performance, achieves state-of-the-art, and promotes researchers to make use of DNNE.