Burnt area delineation from a uni-temporal perspective based on Landsat TM imagery classification using Support Vector Machines

Information on burnt area is of critical importance in many applications as for example in assessing the disturbance of natural ecosystems due to a fire or in proving important information to policy makers on the land cover changes for establishing restoration policies of fire-affected regions. Such information is commonly obtained through remote sensing image thematic classification and a wide range of classifiers have been suggested for this purpose. The objective of the present study has been to investigate the use of Support Vector Machines (SVMs) classifier combined with multispectral Landsat TM image for obtaining burnt area mapping. As a case study a typical Mediterranean landscape in Greece was used, in which occurred one of the most devastating fires during the summer of 2007. Accuracy assessment was based on the classification overall statistical accuracy results and also on comparisons of the derived burnt area estimates versus validated estimates from the Risk-EOS Burnt Scar Mapping service. Results from the implementation of the SVM using diverse kernel functions showed an average overall classification accuracy of 95.87% and a mean kappa coefficient of 0.948, with the burnt area class always clearly separable from all the other classes used in the classification scheme. Total burnt area estimate computed from the SVM was also in close agreement with that from Risk-EOS (mean difference of less than 1%). Analysis also indicated that, at least for the studied here fire, the inclusion of the two middle infrared spectral bands TM5 and TM7 of TM sensor as well as the selection of the kernel function in SVM implementation have a negligible effect in both the overall classification performance and in the delineation of total burnt area. Overall, results exemplified the appropriateness of the spatial and spectral resolution of the Landsat TM imagery combined with the SVM in obtaining rapid and cost-effective post-fire analysis. This is of considerable scientific and practical value, given the present open access to the archived and new observations from this satellite radiometer globally.     

基于SVM的资源三号测绘卫星影像分类

以江苏省宜兴市为研究区,利用支持向量机(SVM)方法对资源三号测绘卫星影像进行了分类,其总分类精度为97.76%,Kappa精度为0.968 7。为了评价算法的适用性,同时应用最大似然法与最小距离法对同一影像进行分类测试,支持向量机分类法精度高于其他2种方法,可以满足土地覆盖分类调查需求。     

In-season crop classification using elements of the Kennaugh matrix derived from polarimetric RADARSAT-2 SAR data

Accurate spatio-temporal classification of crops is of prime importance for in-season crop monitoring. Synthetic Aperture Radar (SAR) data provides diverse physical information about crop morphology. In the present work, we propose a day-wise and a time-series approach for crop classification using full-polarimetric SAR data. In this context, the 4 × 4 real Kennaugh matrix representation of a full-polarimetric SAR data is utilized, which can provide valuable information about various morphological and dielectric attributes of a scatterer. The elements of the Kennaugh matrix are used as the parameters for the classification of crop types using the random forest and the extreme gradient boosting classifiers.The time-series approach uses data patterns throughout the whole growth period, while the day-wise approach analyzes the PolSAR data from each acquisition into a single data stack for training and validation. The main advantage of this approach is the possibility of generating an intermediate crop map, whenever a SAR acquisition is available for any particular day. Besides, the day-wise approach has the least climatic influence as compared to the time series approach. However, as time-series data retains the crop growth signature in the entire growth cycle, the classification accuracy is usually higher than the day-wise data.Within the Joint Experiment for Crop Assessment and Monitoring (JECAM) initiative, in situ measurements collected over the Canadian and Indian test sites and C-band full-polarimetric RADARSAT-2 data are used for the training and validation of the classifiers. Besides, the sensitivity of the Kennaugh matrix elements to crop morphology is apparent in this study. The overall classification accuracies of 87.75% and 80.41% are achieved for the time-series data over the Indian and Canadian test sites, respectively. However, for the day-wise data, a ∼6% decrease in the overall accuracy is observed for both the classifiers.     

Spatial prediction of rainfall-induced shallow landslides using hybrid integration approach of Least-Squares Support Vector Machines and differential evolution optimization: a case study in Central Vietnam

This study represents a hybrid intelligence approach based on the differential evolution optimization and Least-Squares Support Vector Machines for shallow landslide prediction, named as DE–LSSVM_SLP. The LSSVM is used to establish a landslide prediction model whereas the DE is adopted to search the optimal tuning parameters of the LSSVM model. In this research, a GIS database with 129 historical landslide records in the Quy Hop area (Central Vietnam) has been collected to establish the hybrid model. The receiver operating characteristic (ROC) curve and area under the curve (AUC) were used to assess the performance of the newly constructed model. Experimental results show that the proposed model has high performances with approximately 82% of AUCs on both training and validating datasets. The model’s results were compared with those obtained from other methods, Support Vector Machines, Multilayer Perceptron Neural Networks, and J48 Decision Trees. The result comparison demonstrates that the DE–LSSVM_SLP deems best suited for the dataset at hand; therefore, the proposed model can be a promising tool for spatial prediction of rainfall-induced shallow landslides for the study area.     

Object-based crop classification using multi-temporal SPOT-5 imagery and textural features with a Random Forest classifier

In this study, an object-based image analysis (OBIA) approach was developed to classify field crops using multi-temporal SPOT-5 images with a random forest (RF) classifier. A wide range of features, including the spectral reflectance, vegetation indices (VIs), textural features based on the grey-level co-occurrence matrix (GLCM) and textural features based on geostatistical semivariogram (GST) were extracted for classification, and their performance was evaluated with the RF variable importance measures. Results showed that the best segmentation quality was achieved using the SPOT image acquired in September, with a scale parameter of 40. The spectral reflectance and the GST had a stronger contribution to crop classification than the VIs and GLCM textures. A subset of 60 features was selected using the RF-based feature selection (FS) method, and in this subset, the near-infrared reflectance and the image acquired in August (jointing and heading stages) were found to be the best for crop classification.     

A comparison of Support Vector Machines and Bayesian algorithms for landslide susceptibility modelling

Abstract

In this study, the main goal is to compare the predictive capability of Support Vector Machines (SVM) with four Bayesian algorithms namely Naïve Bayes Tree (NBT), Bayes network (BN), Naïve Bayes (NB), Decision Table Naïve Bayes (DTNB) for identifying landslide susceptibility zones in Pauri Garhwal district (India). First, landslide inventory map was built using 1295 historical landslide data, then in total sixteen influencing factors were selected and tested for landslide susceptibility modelling. Performance of the model was evaluated and compared using Statistical based index methods, Area under the Receiver Operating Characteristic (ROC) curve named AUC, and Chi-square method. Analysis results show that that the SVM has the highest prediction capability, followed by the NBT, DTNBT, BN and NB, respectively. Thus, this study confirms that the SVM is one of the benchmark models for the assessment of susceptibility of landslides.     

Comparison of two feature selection methods for the separability analysis of intertidal sediments with spectrometric datasets in the German Wadden Sea

The spectral features of intertidal sediments are all influenced by the same biophysical properties, such as water, salinity, grain size or vegetation and therefore they are hard to separate by using only multispectral sensors. This could be shown by a previous study of Jung et al. (2015). A more detailed analysis of their characteristic spectral feature has to be carried out to understand the differences and similarities. Spectrometry data (i.e., hyperspectral sensors), for instance, have the opportunity to measure the reflection of the landscape as a continuous spectral pattern for each pixel of an image built from dozen to hundreds of narrow spectral bands. This reveals a high potential to measure unique spectral responses of different ecological conditions (Hennig et al., 2007).In this context, this study uses spectrometric datasets to distinguish between 14 different sediment classes obtained from a study area in the German Wadden Sea. A new feature selection method is proposed (Jeffries-Matusita distance bases feature selection; JMDFS), which uses the Euclidean distance to eliminate the wavelengths with the most similar reflectance values in an iterative process. Subsequent to each iteration, the separation capability is estimated by the Jeffries-Matusita distance (JMD). Two classes can be separated if the JMD is greater than 1.9 and if less than four wavelengths remain, no separation can be assumed. The results of the JMDFS are compared with a state-of-the-art feature selection method called ReliefF. Both methods showed the ability to improve the separation by achieving overall accuracies greater than 82%. The accuracies are 4%–13% better than the results with all wavelengths applied. The number of remaining wavelengths is very diverse and ranges from 14 to 213 of 703. The advantage of JMDFS compared with ReliefF is clearly the processing time. ReliefF needs 30 min for one temporary result. It is necessary to repeat the process several times and to average all temporary results to achieve a final result. In this study 50 iterations were carried out, which makes four days of processing. In contrast, JMDFS needs only 30 min for a final result.     

An evaluation of Guided Regularized Random Forest for classification and regression tasks in remote sensing

New Earth observation missions and technologies are delivering large amounts of data. Processing this data requires developing and evaluating novel dimensionality reduction approaches to identify the most informative features for classification and regression tasks. Here we present an exhaustive evaluation of Guided Regularized Random Forest (GRRF), a feature selection method based on Random Forest. GRRF does not require fixing a priori the number of features to be selected or setting a threshold of the feature importance. Moreover, the use of regularization ensures that features selected by GRRF are non-redundant and representative. Our experiments based on various kinds of remote sensing images, show that GRRF selected features provides similar results to those obtained when using all the available features. However, the comparison between GRRF and standard random forest features shows substantial differences: in classification, the mean overall accuracy increases by almost 6% and, in regression, the decrease in RMSE almost reaches 2%. These results demonstrate the potential of GRRF for remote sensing image classification and regression. Especially in the context of increasingly large geodatabases that challenge the application of traditional methods.     

A systematic comparison of different object-based classification techniques using high spatial resolution imagery in agricultural environments

Geographic Object-Based Image Analysis (GEOBIA) is becoming more prevalent in remote sensing classification, especially for high-resolution imagery. Many supervised classification approaches are applied to objects rather than pixels, and several studies have been conducted to evaluate the performance of such supervised classification techniques in GEOBIA. However, these studies did not systematically investigate all relevant factors affecting the classification (segmentation scale, training set size, feature selection and mixed objects). In this study, statistical methods and visual inspection were used to compare these factors systematically in two agricultural case studies in China. The results indicate that Random Forest (RF) and Support Vector Machines (SVM) are highly suitable for GEOBIA classifications in agricultural areas and confirm the expected general tendency, namely that the overall accuracies decline with increasing segmentation scale. All other investigated methods except for RF and SVM are more prone to obtain a lower accuracy due to the broken objects at fine scales. In contrast to some previous studies, the RF classifiers yielded the best results and the k-nearest neighbor classifier were the worst results, in most cases. Likewise, the RF and Decision Tree classifiers are the most robust with or without feature selection. The results of training sample analyses indicated that the RF and adaboost. M1 possess a superior generalization capability, except when dealing with small training sample sizes. Furthermore, the classification accuracies were directly related to the homogeneity/heterogeneity of the segmented objects for all classifiers. Finally, it was suggested that RF should be considered in most cases for agricultural mapping.     

Impact of the spatial resolution of climatic data and soil physical properties on regional corn yield predictions using the STICS crop model

The assimilation of Earth observation (EO) data into crop models has proven to be an efficient way to improve yield prediction at a regional scale by estimating key unknown crop management practices. However, the efficiency of prediction depends on the uncertainty associated with the data provided to crop models, particularly climatic data and soil physical properties. In this study, the performance of the STICS (Simulateur mulTIdisciplinaire pour les Cultures Standard) crop model for predicting corn yield after assimilation of leaf area index derived from EO data was evaluated under different scenarios. The scenarios were designed to examine the impact of using fine-resolution soil physical properties, as well as the impact of using climatic data from either one or four weather stations across the region of interest. The results indicate that when only one weather station was used, the average annual yield by producer was predicted well (absolute error <5%), but the spatial variability lacked accuracy (root mean square error = 1.3 t ha⁻¹). The model root mean square error for yield prediction was highly correlated with the distance between the weather stations and the fields, for distances smaller than 10 km, and reached 0.5 t ha⁻¹ for a 5-km distance when fine-resolution soil properties were used. When four weather stations were used, no significant improvement in model performance was observed. This was because of a marginal decrease (30%) in the average distance between fields and weather stations (from 10 to 7 km). However, the yield predictions were improved by approximately 15% with fine-resolution soil properties regardless of the number of weather stations used. The impact of the uncertainty associated with the EO-derived soil textures and the impact of alterations in rainfall distribution were also evaluated. A variation of about 10% in any of the soil physical textures resulted in a change in dry yield of 0.4 t ha⁻¹. Changes in rainfall distribution between two abundant rainfalls during the growing season led to a significant change in yield (0.5 t ha⁻¹ on average). Our results highlight the importance of using fine-resolution gridded daily precipitation data to capture spatial variations of rainfall as well as using fine-resolution soil properties instead of coarse-resolution soil properties from the Canadian soil dataset, especially for regions with high pedodiversity.