Search

Precise estimation of Sea Surface Salinity (SSS) is of prime importance to know marine physical and biochemical processes. In this manuscript we have utilized the Random Forest (RF) and Support Vector Regression (SVR) to estimate the SSS concentration. Particularly, we have utilized the global SSS values provided by the Copernicus Marine Services. We have also extracted the Aqua MODIS and Sentinel-3 data from the Google Earth Engine platform. The complete range of SSS varies approximately from 6 psu to 38 psu which enables the global application of the dataset. We have also studied the importance of different wavelength bands and their significance to SSS using the RF model. It is found that wavelengths around 400 nm have their highest significance due to the sensitiveness to Chromophoric Dissolved Organic Matter (CDOM) and other related constituents. For the regression analysis we have also utilized the Laplacian kernel in the SVR. It is found that the SVR with Laplace kernel outperforms the RF and SVR with Radial Basis Function (RBF) kernel. The trained algorithms are used to estimate the SSS content over the Bay of Bengal and the Arabian Sea regions which show interesting variations in SSS content due to diverse climatological conditions. [ABSTRACT FROM AUTHOR]

Imagery acquired by the Moderate-resolution Imaging Spectroradiometer (MODIS) provides a global archive of dailyNormalized Difference Snow Index (NDSI) at 500 m nominal resolution since the year 2000. While Sentinel-2 (S2) NDSI provides an increased spatial resolution of 20 m since the year 2015, the temporal resolution amounts to only 5 days and thus lacks the high temporal resolution of MODIS. Efforts to combine NDSI datasets for an increased temporal and spatial resolution have so far focused on the deriving binary snow cover maps or combining data from other sensors. In contrast, we produce fine scale (20 m) fractional snow cover (FSC) by downscaling MODIS NDSI to S2 resolution. Random forest regression predicts S2 NDSI based on dynamic features (MODIS NDSI, day-of-year) and static, topographic features for an alpine study site. Subsequently, FSC is derived from S2 NDSI. Cross-validation results in R2 of 0.795 and RMSE of 0.155 for FSC and outperforms common resampling methods. Multi-annual S2 NDSI metrics are able to slightly improve model accuracy. Our results suggest that combining topographical data and low-resolution NDSI allows to produce daily, high-resolution S2 NDSI and FSC and improve fine scale characterization of snow cover dynamics in mountain landscapes. [ABSTRACT FROM AUTHOR]

The invasion and expansion of Spartina alterniflora have led to a certain degree of degradation in the ecological environment of coastal wetlands. To discuss the impacts of Spartina alterniflora expansion on the tidal channels, we took the third core area of the Dafeng Elk National Nature Reserve in Yancheng, Jiangsu Province as the study area, retrieved the spatial distribution of Spartina alterniflora by random forest algorithm based on long time series of Landsat and Sentinel-2 images, analysed the expansion models. The tidal channels were obtained by the image segmentation based on MSAVI and visual interpretation and the spatiotemporal evolution traits were finally analysed. The results revealed that the area and perimeter of Spartina alterniflora are continuously increasing, with a wide-ranging invasion. The coverage area peaks at 13.4%, making it the dominant species in the study area. When excluding the influence of human activities, a significant negative correlation emerges between the invasion rate and compactness of Spartina alterniflora and the spatio-temporal evolution of tidal channels. Faster invasion rates and greater compactness correspond to a deceleration in tidal channel morphology development. Additionally, we also found that the growth of Spartina alterniflora has a positive influence on the growth and reproduction of elk. [ABSTRACT FROM AUTHOR]

Quickly and accurately obtaining information on rice planting areas is crucial for maintaining food security and social stability. Based on the time series of Sentinel-1A images, this study proposes an automatic rice mapping method called the Dynamic Threshold Time-Weighted Dynamic Time Warping method (DT-TWDTW). This method consists of two parts. First, rice samples are extracted automatically by a dynamic threshold method based on the unique flooding signal before and after rice transplanting. Then, the Time-Weighted Dynamic Time Warping (TWDTW) method is used for rice mapping based on the cumulative cost matrix between unknown pixels and rice samples. The results show that the DT-TWDTW method can achieve satisfactory mapping accuracy with an overall accuracy of 94.6%, higher than random forest (89.9%), and support vector machine (91.7%). This study provides a new approach and perspective for remote sensing monitoring of rice, especially for some regions without prior phenological knowledge and collection of the rice samples. [ABSTRACT FROM AUTHOR]

Efficiently obtaining tobacco planting area is significant for rationally allocating tobacco resources and realizing the balance between supply and demand. However, tobacco fields are characterized by fragmentation in hilly areas, which brings various challenges to extracting the tobacco planting area. A 16-square-kilometre tobacco planting region in Xuancheng City, Anhui Province, China, was selected as the study area in this paper. First, the single-temporal full-feature sets (STFFS), a time-series full-feature set (TFFS) and a time-series optimal-feature set (TOFS) were constructed from multi-period Sentinel-2 images, respectively. Then, we applied Random Forest (RF), Support Vector Machine (SVM), Neural Network Classification (NNC) and Maximum Likelihood Classification (MLC) to classify the feature sets and compare the classification accuracy. The experimental results demonstrate that: (1) The best growth stage of tobacco remote sensing recognition is the mulching film phase of the spherical plant stage (ST1 period). (2) The classification accuracy indicates that TOFS outperforms the STFFS. (3) TOFS can still maintain the same classification accuracy as TFFS with fewer feature dimensions. (4) The RF classifier has great stability and robustness, which achieve an overall accuracy (OA) of 90.65%, 93.50%, and 93.09% based on the feature sets of ST1 period, TFFS, and TOFS, respectively. [ABSTRACT FROM AUTHOR]

Species-level classification of mangroves provides important inputs for conservation, rehabilitation and understanding of ecosystem functions. The hyperspectral sensor, Airborne Visible InfraRed Imaging Spectrometer-New Generation (AVIRIS-NG), holds promises for species-level discrimination by virtue of its coverage across a wider spectrum at very high spatial resolution. Using the continuum removal (CR) technique and absorption band depth (ABD), this study applied Random Forest (RF) model to classify the distribution of three species (Heritiera fomes, Excoecaria agallocha and Avicennia officinalis) and two of their combinations (Heritiera fomes-Excoecaria agallocha and Avicennia officinalis-Excoecaria agallocha). The classified map demonstrated good accuracy (overall accuracy = 88%; kappa coefficient = 0.84) using ABD as an independent variable. The important wavelengths (972, 1172, 1177 nm) identified for mangrove species discrimination correspond to water absorption bands. This characteristic may be replicated for species-level classification of other mangrove forests with similar species. [ABSTRACT FROM AUTHOR]

In this letter, a novel method for change detection is proposed using neighbourhood structure correlation. Because structure features are insensitive to the intensity differences between bi-temporal images, we perform the correlation analysis on structure features rather than intensity information. First, we extract the structure feature maps by using multi-orientated gradient information. Then, the structure feature maps are used to obtain the Neighbourhood Structural Correlation Image (NSCI), which can represent the context structure information. In addition, we introduce a measure named matching error, which can be used to improve neighbourhood information. Subsequently, a change detection model based on the random forest is constructed. The NSCI features and matching error (ME) are together used as the model inputs for training and prediction. Finally, the decision tree voting is used to produce the change detection result. To evaluate the performance of the proposed method, it is compared with three state-of-the-art change detection methods. The experimental results on two datasets demonstrate the effectiveness and robustness of the proposed method. [ABSTRACT FROM AUTHOR]

Shrubs are typical vegetation on the Qinghai – Tibet Plateau, and their spatial distribution pattern is crucial for understanding the regional ecological environment. Existing global land cover products cannot represent shrubs well in this area, and the relationship between shrubs and the corresponding environmental factors is still unclear. In this study, using the Dogxung Zangbo River Basin as an example, Sentinel-2 images were applied to estimate the regional shrub fraction image, using the random forest regression model trained with samples generated from fine-resolution Jilin-1 images. The estimated shrub fraction image was further analysed using an elevation dataset to explore the impact of topography on the spatial distribution of shrubs. The results showed that using Sentinel-2 images acquired in both plant growing and non-growing seasons can better distinguish shrubs from grass than using only one season of data, and the coefficient of determination and root-mean-square error derived from this model were 0.821 and 0.087, respectively. The shrub distribution was strongly correlated with topography. In the study area, areas with a southeast or south orientation and large slope are more suitable for shrub growth. [ABSTRACT FROM AUTHOR]

We used the secondary hyperspectral data set (leaf reflectance of maize) taken in the spectral range from 350 to 2500 nm in a field (under low and high nitrogen conditions) and glasshouse experiment. Water, chlorophyll, nitrogen (N), phosphorus (P) and potassium (K) contents of maize leaves taken from these experiments were measured using the standard methods. Six machine learning regression algorithms (Random Forest, Support Vector Regression, k-Nearest Neighbours, Multilayer Perceptron, Gradient Boosting Regression and Partial Least Square Regression) were used for development of the models to predict these parameters from leaf reflectance data. Each plant parameter was estimated with a different machine learning algorithm. Explainable artificial intelligence methods were used to identify the optimum wavelengths for each parameter. The wavelengths in the short-wave infrared (SWIR) region were found optimum for estimating the water content across the three N regimes, and the red-edge band for chlorophyll. The optimum wavelengths for estimating N content in leaves were in the green spectral region under low N status, near-infrared (NIR) and SWIR regions under greenhouse and SWIR region under high N status. The important wavelengths for estimating P in maize leaves were 1088, 1262 and 1263 nm under low N status and the SWIR range under high N status and greenhouse conditions. The SWIR band was useful for estimating the K content in leaves. [ABSTRACT FROM AUTHOR]

Using remote-sensing technology to accurately map the composition and distribution of tree species is vital for sustainable forest resource management. Sentinel-2 data with the dense time-series observations enable to identify tree species. However, few studies clarify the differences in classification using Sentinel-2 images in natural forest and planted forest. Two study areas with different forest environments (planted forest and natural forest) were selected to evaluate the potential of Sentinel-2 imagery. Our results show that red-edge band, short-wave infrared (SWIR) band and vegetation indices (VIs), such as Red-Edge Normalized Difference Vegetation Index (ReNDVI), Soil-Adjusted Vegetation Index (SAVI), and Enhanced Vegetation Index (EVI), have important effects on tree species classification, especially in the growing season. By using two machine learning algorithms (support vector machine [SVM] and random forest [RF]), the results show that the classification accuracy for planted forests (91.27% in SVM and 88.35% in RF) significantly exceeds that of natural forests (84.34% in SVM and 81.03% in RF). This accuracy difference may be related to the spatial heterogeneity inside the forest and the surrounding environmental implications. Although the multi-temporal Sentinel-2 images produce satisfactory accuracy for classifying tree species, further research is needed to improve the classification accuracy. [ABSTRACT FROM AUTHOR]

Recent increase in the frequency and spatial extent of wildfires motivates the quick recognition of the affected soil properties over large areas. Digital Soil Mapping is a valuable approach to map soil attributes based on remote sensing and field observations. We predicted the spatial distribution of post-fire litter composition in a 40,600 ha basin burned on the 2017 wildfire of Chile. Remotely sensed data of topography, vegetation structure and spectral indices (SI) were used as predictors of random forest (RF) models. Litter sampled in 60 hillslopes after the fire provided training and validation data. Predictors selected by the Variable Selection Using Random Forests (VSURF) algorithm resulted in models for litter composition with acceptable accuracy (coefficient of determination, R2 = 0.51–0.64, Normalized Root Mean Square Error, NRMSE = 16.9–22.1, percentage bias, pbias = −0.35%-0.5%). Modelled litter parameters decrease in concentration respect to the degree of burn severity, and the pre-fire biomass. Because pre-fire vegetation was conditioned by land cover and by a previous (2 years old) wildfire event, our results highlight the cumulative effect of severe wildfires in the depletion of litter composition. [ABSTRACT FROM AUTHOR]

North American ash species (Fraxinus spp.) are under dire threat from the invasive pest, emerald ash borer (Agrilus planipennis, EAB), and mapping ash trees is of utmost significance for conservation and prevention efforts. We developed remote sensing techniques to identify ash trees at the individual tree level in a mixed-species forest in Maine, USA, using hyperspectral bands combined with derived spectral vegetation indices (SVIs), texture metrics using Random Forest (RF) and Support Vector Machine (SVM) algorithms. The pixel-based SVM-Optimized model proved to be the most accurate classification method with the lowest number of input variables including six Minimum Noise Fraction (MNF)-reduced bands, four SVIs and two texture variables, achieving 78.2% overall accuracy and 81.0% ash Producer's Accuracy. The technique presented in this study could be used to map ash trees throughout Maine and other states with similar forest types for future preservation efforts. [ABSTRACT FROM AUTHOR]

Detailed soil maps contribute to the understanding of agricultural and food value chains. PlanetScope satellite constellations have high spatial and temporal resolution and could be used to develop Synthetic Soil Image (SYSI) at specific dates of field crops calendar. This work aimed to derive a SYSI from PlanetScope time series and estimate soil clay content, an important attribute related to several soil functions. The normalization process was applied to make the radiometries of the different Doves' images compatible and, subsequently, land cover classification was performed to select the bare soil class, obtaining the SYSI-Planet. The clay content was predicted for 2337 topsoil samples using the SYSI-Planet and Random Forest algorithm. The normalization of PlanetScope images were based on reference images, which mitigated the radiometric differences from acquisition dates and different radiometric calibrations. SYSI-Planet resulted in an excellent clay content prediction, resulting in a coefficient of determination (R2) equal to 0.93 and Root-Mean-Square Error (RMSE) equal to 70.90. The predicted clay mapped revealed spatial patterns with a high spatial resolution (~3 m), which can enhance precise agricultural management. [ABSTRACT FROM AUTHOR]

This letter proposes to use two k-nearest neighbours (k-NN) based algorithms namely: randomized k-NN and weighted k-NN for feature selection with two hyperspectral datasets. Two embedded feature selection methods, extreme gradient boosting and random forest classifier and one wrapper-based approach using support vector machine classifier, were used for comparison in terms of classification accuracy with selected features. For all classifications, a support vector machine classifier was used. Comparison of results in terms of classification accuracy with the selected features suggests better performance by randomized k-NN algorithm in comparison to weighted k-NN and other feature selection approaches used in this study. Results with both dataset also indicate that selected features by randomized k-NN based feature selection approach achieve higher accuracy than that with full datasets. [ABSTRACT FROM AUTHOR]

Despite significant improvement in Moderate Resolution Imaging Spectroradiometer (MODIS) aerosol optical depth (AOD) retrieval, high-resolution–high-accuracy AOD retrieval remains a challenging task. This study utilizes machine learning for AOD retrieval of MODIS data. The global long-time-series data of AERONET sites and corresponding MODIS data in time and space were used as sample training data for aerosol retrieval via use of the random forest (RF) approach. The accuracy and stability of retrieval were ensured by processing AERONET site data, performing time–space matching between different data types, and determining related parameters in the RF model. MODIS data use bands 1–7 of the top-of-atmosphere reflectance (TOA) – extending from the visible to near-infrared radiation spectra – along with the corresponding observation geometry data, global land surface satellite (GLASS) albedo dataset, and normalized difference vegetation index (NDVI) data. The proposed method facilitates realizationthe of high-accuracy aerosol retrieval. Furthermore, significant enhancement in the efficiency of aerosol inversion is an added advantage. [ABSTRACT FROM AUTHOR]

In this study, Sentinel-2 optical satellite imagery was acquired over the Peace Athabasca Delta and assessed for its open water classification capabilities using an object-oriented deep learning algorithm. The workflow involved segmenting the satellite data into meaningful image objects, building a Convolutional Neural Network (CNN), training the CNN, and lastly applying the CNN, resulting in probability heat maps of open water (with score values ranging from 0–1). Using the vector segmentation, heat maps were then iteratively assigned final class labels ('open water' or 'other') based on various probability thresholding. The ensuing open water classifications were assessed against a large validation dataset, and a highest overall accuracy of 96.2% (0.912 kappa coefficient) was achieved, with an open water producer's accuracy of 98.1%. These results were then compared against a Random Forest (RF) classification, and results indicated that the CNN algorithm outperforms RF in this study site. Additionally, an important component of this study was the optimization of several CNN configurations, including patch size and learning rate; the latter which plays a critical role in model adaptation. The optimized object-oriented CNN and associated results can be used to provide resource managers with accurate surface water extent maps at 10 m resolution. [ABSTRACT FROM AUTHOR]

The information on the topography of the seabed included in data from airborne lidar bathymetry can be used for the detection of changes occurring at the bottom of the basin and for the detection of objects deposited on it. Processing of full waveform data enables obtaining data on the water surface and the identification of the underwater situation. The classification process based on the Random Forest (RF) algorithm is presented using data from lidar bathymetry. The classification was performed in two independent approaches using input vector consisting of 16 features. In the first approach, the entire point cloud was classified, in the second the point cloud did not contain points of the water surface. In the classification, traits based on the full waveform and resulting from point cloud geometry were used. The quantitative efficiency of classification was verified through error matrices. The obtained efficiency (100% water surface, 99.9% seabed and 60% objects) of the point classification of the objects enables the possibility for using the RF algorithm for detection of objects on the seabed. In comparison to Support Vector Machines, the RF algorithm has better results in the detection of points on the objects in point cloud with water surface. [ABSTRACT FROM AUTHOR]

Efficient and reliable wetland mapping is essential for wetland resource management and conservation planning. Repeated attempts at classifying dense wetland vegetation has often failed to produce a high-quality product because of poor penetration of electromagnetic signals that cannot detect sub-surface water, and because of associated redundant information. In this study, a two-step classification procedure was developed for classifying typical coastal plain wetlands from multi-temporal Advanced Land Observing Satellite (ALOS)/Phased Array type L-band Synthetic Aperture Radar (PALSAR)-1 data. First, a new statistical strategy for feature selection was proposed by combining manual and automatic selection according to the backscattering mechanism differences between various land-cover types. Then the Random Forest (RF) classifier was applied to evaluate the established feature combinations for wetland classification. The experimental results show that the method described has high classification accuracies over the test site, with overall accuracies and kappa coefficients (κ) of above 89.79% and 0.88, respectively. This indicates that the presented technique is effective in dealing with high-dimensional feature space and may lead to significant improvements in multi-temporal Synthetic Aperture Radar (SAR) image-based wetland classification. [ABSTRACT FROM AUTHOR]

In this letter, a novel edge-preserving filtering based approach is proposed for feature extraction of hyperspectral images, which consists of the following steps. First, the dimension of the hyperspectral image is reduced with an averaging based method. Then, the resulting features are obtained by performing edge-preserving filtering on the dimension reduced image, in which a learned edge detection map serves as one of the major cues in the filtering process. The advantage of the proposed method is that it makes full use of the learned edge information in the feature extraction process, and thus, able to improve the performance with respect to other traditional feature extraction methods. Experiments on two real hyperspectral data sets demonstrate the outstanding performance of the proposed method especially when the number of training samples is limited. [ABSTRACT FROM AUTHOR]

Polarimetric Synthetic Aperture Radar (PolSAR) imagery can provide valuable observables at different frequencies for classification tasks. In this paper, we assessed separability rate of various polarimetric features in three frequencies of X-, C-, and L- bands. To this end, Jeffries–Matusita distance was firstly used to measure separability of each polarimetric feature in each frequency band. Random Forest classifier was then applied to map various land cover classes in study area. The classification outputs indicated that C-band results were better and more reliable than L-band results and L-band results were subsequently better than X-band results. These results were perfectly compatible with the results obtained by the separability analysis of multifrequency PolSAR features. [ABSTRACT FROM PUBLISHER]

In this study, a sea ice mapping model based on Random Forest (RF), a rule-based machine learning approach, has been developed for the Korea Multi-Purpose Satellite-5 (KOMPSAT-5) Synthetic Aperture Radar (SAR) data in Enhanced Wide swath mode obtained from 6 August to 9 September 2015 in the Chukchi Sea. A total of 12 texture features derived from backscattering intensity and the gray-level co-occurrence matrix were used as input variables for sea ice mapping. The RF model produced a sea ice map with a grid spacing of 125 m, demonstrating excellent performance in the classification of sea ice and open water with an overall accuracy of 99.2% and a kappa coefficient of 98.5%. Sea ice concentration (SIC) retrieved from the RF-derived sea ice maps was compared with that from ice charts. The mean and median values of the differences between the SICs derived from the RF model and the ice charts were −8.85% and −8.38%, respectively. Such difference was attributed to both the uncertainty in the ice charts and classification error of the RF model. [ABSTRACT FROM AUTHOR]

With an ever growing need to classify multispectral images, the accuracy of the classification becomes a matter of concern, especially when mapping heterogeneous environments such as urban areas. Not all of the features of a high-resolution multispectral image would equally contribute to the formulation of an ideal hypothesis, some of which might be redundant and might hinder the classification itself. Random forests (RF), an ensemble of decision trees, would assist in the removal of such extraneous features. While RF has proved to be one of the most efficient classifiers for spectral imagery, it holds untapped potential which could be honed to yield much more accurate prediction rates. This article discusses the application of AdaBoosted random forest (ABRF) to classify landcover segments from multispectral satellite or aerial imagery. A stage-wise optimization procedure is applied to achieve the best possible performance. It involves genetic feature selection and strategic empirical calibration to determine the ideal set of parameters. The application of boosting to RF resulted in the increase in the overall accuracy from 84.42% to 88.8% with an increase in kappa coefficient from 0.804 to 0.859. [ABSTRACT FROM AUTHOR]

The classification maps are required for the management and the estimation of agricultural disaster compensation; however, those techniques have yet to be established. Some supervised learning models may allow accurate classification. In this study, the Random Forest (RF) classifier and the classification and regression tree (CART) were applied to evaluate the potential of multi-temporal TerraSAR-X dual-polarimetric data, on the StripMap mode, for the classification of crop type. Furthermore, comparisons of the two algorithms and polarizations were carried out. In the study area, beans, beet, grasslands, maize, potato and winter wheat were cultivated, and these crop types were classified using the data set acquired in 2009. The classification results of RF were superior to those of CART, and the overall accuracies were 0.91–0.93. [ABSTRACT FROM AUTHOR]

Potential data sets for landcover classification, such as Landsat (or pre-processed data such as the National Land Cover Dataset (NLCD)), are often too coarse for fine-scale research needs or are cost-prohibitive (Quickbird, Ikonos and Geoeye). Repeated attempts at classifying high spatial resolution data, National Agricultural Imagery Program (NAIP) imagery, based on traditional techniques, such as a maximum likelihood supervised classification, have failed to produce a product with sufficient accuracy. We used the ensemble Random Forests (RFs) classifier to classify landcover at 1 m resolution using 2009 NAIP imagery in south-eastern Wyoming. We classified riparian areas within a 225 km2study area, at 1 m spatial resolution, using RFs with emphasis on riparian corridors that yielded a land cover map with overall accuracy of 81% and a kappa coefficient of 79%. Users’ accuracy of important riparian vegetation species, aspen, riparian grasses and willow were 79%, 81% and 83%, respectively. Techniques presented in this paper, which exploit free NAIP imagery for landcover classification, represent an inexpensive and reliable alternative to purchasing commercial imagery when high spatial resolution landcover data is needed. [ABSTRACT FROM AUTHOR]