Traditional Maximum Likelihood Classifier Research Articles

Assessing forest cover changes in Dak Lak province (Central Highlands of Vietnam) from multi-temporal Landsat data and machine learning techniques

Dak Lak is a province in the Central Highlands region of Vietnam, with a large area of forests and forestry land. However, the forest cover has changed dramatically in recent times due to the influence of human activities and climate change. This article presents the results of assessing forest cover changes in Dak Lak province from Landsat satellite image data for the period 2000 – 2020. Three Landsat satellite image scenes, including Landsat 5 TM images taken in March 2000 and February 2010 and Landsat 8 OLI image taken in February 2020 are used to classify forest cover. Three common machine learning techniques, including Random Forest (RF), Support Vector Machine (SVM), Classification and Regression Tree (CART) and the traditional maximum likelihood classification algorithm are used to classify forest cover in the study area, thereby choosing the method with the highest accuracy. The results show that the RF algorithm has the highest accuracy in classifying forest cover from multi-temporal Landsat images by comparing the overall accuracy value and the Kappa coefficient. The obtained results are used to build forest cover change maps in the period 2000 - 2010, 2010 - 2020 and 2000 - 2020. The results received in the study provide information to help managers in monitoring and protecting forest resources.

Comparing Four Machine Learning Algorithms for Land Cover Classification in Gold Mining: A Case Study of Kyaukpahto Gold Mine, Northern Myanmar

Numerous studies have been undertaken to determine the optimal land use/cover classification algorithm. However, there have not been many studies that have compared and evaluated the performance of maximum likelihood (ML), random forest (RF), support vector machine (SVM), and classification and regression trees (CART) using ASTER imagery, especially in a mining district. Therefore, this study aims to investigate land use/cover (LULC) change over three decades (1990–2020), comparing the performance of the ML, RF, SVM, and CART machine learning algorithms. The Landsat and ASTER data were retrieved using Google Earth Engine (GEE). Traditional ML classification was performed on ArcGIS 10.2 software while RF, SVM, and CART classification were undertaken on GEE. Then, thematic accuracy assessments were conducted for the four algorithms and their performances were compared. The results showed that the largest changes in area occurred in forest cover that decreased from 37.8 to 27.3 km2 during the three decades. The remarkable expansion of gold mining occurred during 2005–2010 with the increases of 1.6%. The mining land rose by 2.9% during the study period whereas agricultural land increased significantly by 10.7% between 1990 and 2020. When comparing the four algorithms, the RF algorithm gives the highest accuracy with an overall accuracy of 95.85% while SVM follows RF with 91.69%. This study proved that RF is the best choice for optimal land use/cover classification, particularly in the mining district.

Detecting and mapping the spatial distribution of Chromoleana odorata invasions in communal areas of South Africa using Sentinel-2 multispectral remotely sensed data

Alien Invasive Plants (AIPs) have been proclaimed to be a significant non-climatic driver of global change. The Siam weed (Chromoleana odorata) is highly invasive in South African rangelands, with serious environmental, agricultural and socio-economic consequences. Therefore, a comprehensive analysis of C.odorata's spatial distribution is necessary for implementing relevant mitigation and management approaches to vulnerable and invaded landscapes. Remotely sensed data offer a viable opportunity for detecting and mapping AIPs spatial extents. Hence, this study sought to compare the value of algorithms (i.e. maximum likelihood and random forest) in detecting and mapping C.odorata's spatial distribution in relation to other land use land cover classes (LULC) using the freely available Sentinel-2 multispectral image data. The findings of the study revealed that the vegetation red edge and near-infrared spectral bands of Sentinel-2 multispectral imagery (MSI) sensor were the most important spectral variables for discriminating C.odorata from other land cover classes. The random forest algorithm yielded the highest overall classification accuracy of 83%, outperforming the traditional maximum likelihood classifier (75%) in classifying C.odorata's spatial distribution. Overall, results demonstrate that the combination of freely available Sentinel-2 MSI satellite data and machine learning random forest algorithm produces the highest accuracies for detecting the spatial distribution of C.odorata, particularly in heterogeneous environments. These results could be beneficial to sustainable rangeland management and the adoption of site-specific mitigation approaches in areas invaded by C.odorata.

Joint probability-based classifier based on vine copula method for land use classification of multispectral remote sensing data

Land use classification is fundamental both for monitoring and predicting regional development patterns and for planning and regulating land use. This research proposed a joint probability-based classifier for land use classification of multispectral remote sensing data and applied it to the Lake Baiyangdian region of North China. This classifier, based on the vine copula method, was suitable for dealing with the uncertainties of land classification and its random variables that did not necessarily obey predefined distributions. Comparison of the results obtained using the proposed classifier with those derived using the widely used maximum likelihood classifier indicated that the accuracy of land use classification of multispectral remote sensing data was higher with the proposed classifier. Compared with the contingency matrix of the maximum likelihood classifier, that of the vine copula classifier showed an increase in the producer’s (user’s) accuracy of rural land (shallow water) of 29.4% (30.0%). The proposed classifier increased the shallow water area and significantly reduced the area of rural land. The main reason was the maximum likelihood classifier had poor classification performance, misclassifying pixels of shallow water as rural land. The findings of this study demonstrated that the vine copula classifier performs better than the traditional maximum likelihood classifier and that its application could promote full utilization of remotely sensed data.

A Comparative Assessment of Ensemble-Based Machine Learning and Maximum Likelihood Methods for Mapping Seagrass Using Sentinel-2 Imagery in Tauranga Harbor, New Zealand

Seagrass has been acknowledged as a productive blue carbon ecosystem that is in significant decline across much of the world. A first step toward conservation is the mapping and monitoring of extant seagrass meadows. Several methods are currently in use, but mapping the resource from satellite images using machine learning is not widely applied, despite its successful use in various comparable applications. This research aimed to develop a novel approach for seagrass monitoring using state-of-the-art machine learning with data from Sentinel–2 imagery. We used Tauranga Harbor, New Zealand as a validation site for which extensive ground truth data are available to compare ensemble machine learning methods involving random forests (RF), rotation forests (RoF), and canonical correlation forests (CCF) with the more traditional maximum likelihood classifier (MLC) technique. Using a group of validation metrics including F1, precision, recall, accuracy, and the McNemar test, our results indicated that machine learning techniques outperformed the MLC with RoF as the best performer (F1 scores ranging from 0.75–0.91 for sparse and dense seagrass meadows, respectively). Our study is the first comparison of various ensemble-based methods for seagrass mapping of which we are aware, and promises to be an effective approach to enhance the accuracy of seagrass monitoring.

Winter wheat identification by integrating spectral and temporal information derived from multi-resolution remote sensing data

Timely crop acreage and distribution information are the basic data which drive many agriculture related applications. For identifying crop types based on remote sensing, methods using only a single image type have significant limitations. Current research that integrates fine and coarser spatial resolution images, using techniques such as unmixing methods, regression models, and others, usually results in coarse resolution abundance without sufficient detail within pixels, and limited attention has been paid to the spatial relationship between the pixels from these two kinds of images. Here we propose a new solution to identify winter wheat by integrating spectral and temporal information derived from multi-resolution remote sensing data and determine the spatial distribution of sub-pixels within the coarse resolution pixels. Firstly, the membership of pixels which belong to winter wheat is calculated using a 25-m resolution resampled Landsat Thematic Mapper (TM) image based on the Bayesian equation. Then, the winter wheat abundance (acreage fraction in a pixel) is assessed by using a multiple regression model based on the unique temporal change features from moderate resolution imaging spectroradiometer (MODIS) time series data. Finally, winter wheat is identified by the proposed Abundance-Membership (AM) model based on the spatial relationship between the two types of pixels. Specifically, winter wheat is identified by comparing the spatially corresponding 10×10 membership pixels of each abundance pixel. In other words, this method takes advantage of the relative size of membership in a local space, rather than the absolute size in the entire study area. This method is tested in the major agricultural area of Yiluo Basin, China, and the results show that acreage accuracy (Aa) is 93.01% and sampling accuracy (As) is 91.40%. Confusion matrix shows that overall accuracy (OA) is 91.4% and the kappa coefficient (Kappa) is 0.755. These values are significantly improved compared to the traditional Maximum Likelihood classification (MLC) and Random Forest classification (RFC) which rely on spectral features. The results demonstrate that the identification accuracy can be improved by integrating spectral and temporal information. Since the identification of winter wheat is performed in the space corresponding to each MODIS pixel, the influence of differences of environmental conditions is greatly reduced. This advantage allows the proposed method to be effectively applied in other places.

Ailanthus altissima mapping from multi-temporal very high resolution satellite images

This study presents the results of multi-seasonal WorldView-2 (WV-2) satellite images classification for the mapping of Ailanthus altissima (A. altissima), an invasive plant species thriving in a protected grassland area of Southern Italy. The technique used relied on a two-stage hybrid classification process: the first stage applied a knowledge-driven learning scheme to provide a land cover map (LC), including deciduous vegetation and other classes, without the need of reference training data; the second stage exploited a data-driven classification to: (i) discriminate pixels of the invasive species found within the deciduous vegetation layer of the LC map; (ii) determine the most favourable seasons for such recognition. In the second stage, when a traditional Maximum Likelihood classifier was used, the results obtained with multi-temporal July and October WV-2 images, showed an output Overall Accuracy (OA) value of ≈91%. To increase such a value, first a low-pass median filtering was used with a resulting OA of 99.2%, then, a Support Vector Machine classifier was applied obtaining the best A. altissima User’s Accuracy (UA) and OA values of 82.47% and 97.96%, respectively, without any filtering. When instead of the full multi-spectral bands set some spectral vegetation indices computed from the same months were used the UA and OA values decreased. The findings reported suggest that multi-temporal, very high resolution satellite imagery can be effective for A. altissima mapping, especially when airborne hyperspectral data are unavailable. Since training data are required only in the second stage to discriminate A. altissima from other deciduous plants, the use of the first stage LC mapping as pre-filter can render the hybrid technique proposed cost and time effective. Multi-temporal VHR data and the hybrid system suggested may offer new opportunities for invasive plant monitoring and follow up of management decision.

Land-cover classification of the Yellow River Delta wetland based on multiple end-member spectral mixture analysis and a Random Forest classifier

ABSTRACTAs an important ecosystem, wetlands play a crucial role in both regional and global environments. Accurate land-cover classification can facilitate the management and understanding of wetlands. Considering the timely and cost-effective characteristics of remote sensing, this technique was used to obtain land-cover information for the Yellow River Delta (YRD) wetland in this investigation. Landsat-8 Operational Land Imager (OLI) sensor data were selected for the data set in this study. A combined approach of multiple end-member spectral mixture analysis (MESMA) and Random Forest (RF) was developed for land-cover classification mapping of the YRD wetland. This study aimed (1) to determine whether the MESMA technique in combination with RF significantly improves the accuracy of classification in complex landscapes such as the YRD wetland, (2) to determine whether the RF classifier shows good performance in land-cover classification of the YRD wetland, and (3) to compare the proposed method with the traditional Maximum Likelihood Classifier (MLC). The proposed hybrid method showed good performance, with an overall accuracy of 89.5% and a kappa coefficient (κ) of 0.88. The inclusion of fractional information derived from MESMA can improve the classification accuracy by 2–3%. In addition, through a comparison with traditional maximum likelihood (ML) methodology, the effectiveness of the proposed approach was evaluated. Overall, the proposed approach in this study can relatively accurately delineate a land-cover classification map of the YRD wetland with Landsat-8 OLI remotely sensed data.

Classification of Complex Urban Fringe Land Cover Using Evidential Reasoning Based on Fuzzy Rough Set: A Case Study of Wuhan City

Urban fringe is the transition zone fine grained with urban and non-urban land cover types. The complex landscape mosaic in this area challenges the land cover classification based on the remote-sensing data. Spectral signatures are not efficient to discriminate all pixels into classes. To improve the recognition and handle the uncertainty, this paper provides a novel integrated approach, based on a fuzzy rough set and evidential reasoning (FRSER), for land cover classification in an urban fringe area. The approach is implemented on Landsat Operation Land Imager data covering the urban fringe area of Wuhan city, China. A fuzzy rough set is first used to define a decision table from multispectral imagery and ground reference data. Then the fuzzy rough information system is interpreted using the Dempster–Shafer theory, based on an evidential reasoning system. A final land cover classification with uncertainty is achieved by evidential reasoning. The results are compared with the traditional maximum likelihood classifier (MLC) and some rough set-based classifiers including classical rough set classifier (RS), fuzzy rough set classifier (FRS), and variable precision fuzzy rough set classifier (VPFRS). The better overall accuracy, user’s and producer’s accuracies, and the kappa coefficient, in comparison with the other classifiers, suggest that the proposed approach can effectively discriminate land cover types in urban fringe areas with high inter-class similarities and intra-class heterogeneity. It is also capable of handling the uncertainty in data processing, and the final land cover map comes with a degree of uncertainty. The proposed approach that can efficiently integrate the merits of both the fuzzy rough set and DS theory provides an efficient method for urban fringe land cover classification.

Using GF-2 Imagery and the Conditional Random Field Model for Urban Forest Cover Mapping

ABSTRACTGaofen-2 (GF-2), a Chinese new-generation satellite launched in August 2014, is providing high-resolution imagery for Earth observation. In this study, GF-2 imagery was employed for mapping forest cover in the core area of Tongzhou district, Beijing, China. The study analysed the performance of GF-2 data for identifying urban forest using a contextual classification model conditional random field (CRF) with Gabor texture features. The results show that the proposed method outperforms the traditional maximum likelihood classifier (MLC) by improving the producer’s accuracy of conifer and hardwood forest from 86.61% to 92.41%, and 86.59% to 91.57%, respectively. Overall, 87.43% of the area classified as forest by GF-2 classification spatially corresponded to areas of the reference forest map. The mapping results suggest that GF-2 imagery in concert with an efficient classification algorithm can be recommended for urban forest monitoring.

Mapping the land-cover distribution in arid and semiarid urban landscapes with Landsat Thematic Mapper imagery

Mapping the land-cover distribution in arid and semiarid urban landscapes using medium spatial resolution imagery is especially difficult due to the mixed-pixel problem in remotely sensed data and the confusion of spectral signatures among bare soils, sparse density shrub lands, and impervious surface areas (ISAs hereafter). This article explores a hybrid method consisting of linear spectral mixture analysis (LSMA), decision tree classifier, and cluster analysis for mapping land-cover distribution in two arid and semiarid urban landscapes: Urumqi, China, and Phoenix, USA. The Landsat Thematic Mapper (TM) imagery was unmixed into four endmember fraction images (i.e. high-albedo object, low-albedo object, green vegetation (GV), and soil) using the LSMA approach. New variables from these fraction images and TM spectral bands were used to map seven land-cover classes (i.e. forest, shrub, grass, crop, bare soil, ISA, and water) using the decision tree classifier. The cluster analysis was further used to modify the classification results. QuickBird imagery in Urumqi and aerial photographs in Phoenix were used to assess classification accuracy. Overall classification accuracies of 86.0% for Urumqi and 88.7% for Phoenix were obtained, much higher accuracies than those utilizing the traditional maximum likelihood classifier (MLC). This research demonstrates the necessity of using new variables from fraction images to distinguish between ISA and bare soils and between shrub and other vegetation types. It also indicates the different effects of spatial patterns of land-cover composition in arid and semiarid landscapes on urban land-cover classification.

Classification of remote sensed data using Artificial Bee Colony algorithm

The present study employs the traditional swarm intelligence technique in the classification of satellite data since the traditional statistical classification technique shows limited success in classifying remote sensing data. The traditional statistical classifiers examine only the spectral variance ignoring the spatial distribution of the pixels corresponding to the land cover classes and correlation between various bands. The Artificial Bee Colony (ABC) algorithm based upon swarm intelligence which is used to characterise spatial variations within imagery as a means of extracting information forms the basis of object recognition and classification in several domains avoiding the issues related to band correlation. The results indicate that ABC algorithm shows an improvement of 5% overall classification accuracy at 6 classes over the traditional Maximum Likelihood Classifier (MLC) and Artificial Neural Network (ANN) and 3% against support vector machine.

UAV Remote Sensing for Urban Vegetation Mapping Using Random Forest and Texture Analysis

Unmanned aerial vehicle (UAV) remote sensing has great potential for vegetation mapping in complex urban landscapes due to the ultra-high resolution imagery acquired at low altitudes. Because of payload capacity restrictions, off-the-shelf digital cameras are widely used on medium and small sized UAVs. The limitation of low spectral resolution in digital cameras for vegetation mapping can be reduced by incorporating texture features and robust classifiers. Random Forest has been widely used in satellite remote sensing applications, but its usage in UAV image classification has not been well documented. The objectives of this paper were to propose a hybrid method using Random Forest and texture analysis to accurately differentiate land covers of urban vegetated areas, and analyze how classification accuracy changes with texture window size. Six least correlated second-order texture measures were calculated at nine different window sizes and added to original Red-Green-Blue (RGB) images as ancillary data. A Random Forest classifier consisting of 200 decision trees was used for classification in the spectral-textural feature space. Results indicated the following: (1) Random Forest outperformed traditional Maximum Likelihood classifier and showed similar performance to object-based image analysis in urban vegetation classification; (2) the inclusion of texture features improved classification accuracy significantly; (3) classification accuracy followed an inverted U relationship with texture window size. The results demonstrate that UAV provides an efficient and ideal platform for urban vegetation mapping. The hybrid method proposed in this paper shows good performance in differentiating urban vegetation mapping. The drawbacks of off-the-shelf digital cameras can be reduced by adopting Random Forest and texture analysis at the same time.

Application of Support Vector Machines for Landuse Classification Using High-Resolution RapidEye Images: A Sensitivity Analysis

The classification accuracy of remotely sensed data and its sensitivity to classification algorithms have a critical importance for the geospatial community, as classified images provide the base layers for many applications and models. Support Vector Machines (SVMs), a non-parametric statistical learning method that has recently been used in numerous applications in image processing. The SVMs need user-defined parameters and each parameter has different impact on kernels hence the classification accuracy of SVMs is based upon the choice of the parameters and kernels. The objective of this study is to investigate the sensitivity of SVM architecture including internal parameters and kernel types on landuse classification accuracy of RapidEye imagery for the study area in Turkey. Four types of kernels (linear, polynomial, radial basis function, and sigmoid) were used for the SVM classification. A total of 63 different models were developed and implemented for sensitivity analysis of SVM architecture. The traditional Maximum Likelihood Classification (MLC) method was also performed for comparison. The classification accuracies of the best model for each kernel type and MLC are 85.63%, 83.94%, 83.94%, 83.82% and 81.64% for polynomial, linear, radial basis function, sigmoid kernels and MLC, respectively. The results suggest that the choice of model parameters and kernel types play an important role on SVMs classification accuracy. Best model of polynomial kernel outperformed all SVMs models and gave the highest classification accuracy of 85.63% with RapidEye imagery.

Land Cover Classification of Landsat Data with Phenological Features Extracted from Time Series MODIS NDVI Data

Temporal-related features are important for improving land cover classification accuracy using remote sensing data. This study investigated the efficacy of phenological features extracted from time series MODIS Normalized Difference Vegetation Index (NDVI) data in improving the land cover classification accuracy of Landsat data. The MODIS NDVI data were first fused with Landsat data via the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) algorithm to obtain NDVI data at the Landsat spatial resolution. Next, phenological features, including the beginning and ending dates of the growing season, the length of the growing season, seasonal amplitude, and the maximum fitted NDVI value, were extracted from the fused time series NDVI data using the TIMESAT tool. The extracted data were integrated with the spectral data of the Landsat data to improve classification accuracy using a maximum likelihood classifier (MLC) and support vector machine (SVM) classifier. The results indicated that phenological features had a statistically significant effect on improving the land cover classification accuracy of single Landsat data (an approximately 3% increase in overall classification accuracy), especially for vegetation type discrimination. However, the phenological features did not improve on statistical measures including the maximum, the minimum, the mean, and the standard deviation values of the time series NDVI dataset, especially for human-managed vegetation types. Regarding different classifiers, SVM could achieve better classification accuracy than the traditional MLC classifier, but the improvement in accuracy obtained using advanced classifiers was inferior to that achieved by involving the temporally derived features for land cover classification.

Comparative analysis of mapping burned areas from landsat TM images

Remote sensing is a major source of mapping the burned area caused by forest fire. The focus in this application is to map a single class of interest, i.e. burned area. In this study, three different data combinations were classified using different classifiers and quantitatively compared. The adopted classifiers are Support Vector Data Description (SVDD), a one-class classifier, Binary classifier Support Vector Machines (SVMs) and traditional Maximum Likelihood classifier (ML). At first, the Principal Component Analysis (PCA) was applied to extract the best possible features form the original multispectral image (OMI) and calculated spectral indices (SI). Then the resulting subset of features was applied to the classifiers. The comparative study has undertaken to find firstly, the best possible set of features (data combination) and secondly, an effective classifier to map the burned areas. The best possible set of features was attained by data combination- II (i.e., OMI information). Furthermore, the results of the SVM showed the high classification accuracies than ML. Experimental results demonstrate that even though the SVDD for mapping the burned areas doesn't showed the higher classification accuracy than SVM, but it shows the suitability for the cases with few or poorly represented labelled samples available. The parameters should be further optimized through the use of intelligent training for improving the accuracy of SVDD.

Classification of ETM+ Remote Sensing Image Based on Hybrid Algorithm of Genetic Algorithm and Back Propagation Neural Network

The back propagation neural network (BPNN) algorithm can be used as a supervised classification in the processing of remote sensing image classification. But its defects are obvious: falling into the local minimum value easily, slow convergence speed, and being difficult to determine intermediate hidden layer nodes. Genetic algorithm (GA) has the advantages of global optimization and being not easy to fall into local minimum value, but it has the disadvantage of poor local searching capability. This paper uses GA to generate the initial structure of BPNN. Then, the stable, efficient, and fast BP classification network is gotten through making fine adjustments on the improved BP algorithm. Finally, we use the hybrid algorithm to execute classification on remote sensing image and compare it with the improved BP algorithm and traditional maximum likelihood classification (MLC) algorithm. Results of experiments show that the hybrid algorithm outperforms improved BP algorithm and MLC algorithm.

Urban land-use, land-cover classification through watershed segmentation in the V–I–S feature space

This article introduces an innovative approach using marker-controlled watershed segmentation (WS) in the Vegetation–Impervious Surface–Soil (V–I–S) feature space for urban land-use and land-cover (LULC) classification. The complement (e.g., the inverse) of the V–I–S feature space image shows depressions, which can be treated as topographic watersheds and they correspond to LULC classes. WS partitions the complement of V–I–S feature space image into LULC regions based on user-specified initial markers. These regions are then labelled with specific LULC classes, which lead to the subsequent LULC classification. The results showed that WS can achieve substantial classification accuracy. Compared to the traditional maximum likelihood classifier (MLC), our method requires less user input while improving the classification accuracy. A sensitivity test of the WS-based method on the location of the initial markers also is provided.

AUTOMATIC TRAINING SITE SELECTION FOR AGRICULTURAL CROP CLASSIFICATION: A CASE STUDY ON KARACABEY PLAIN, TURKEY

Abstract. This study implements a traditional supervised classification method to an optical image composed of agricultural crops by means of a unique way, selecting the training samples automatically. Panchromatic (1m) and multispectral (4m) Kompsat-2 images (July 2008) of Karacabey Plain (~100km2), located in Marmara region, are used to evaluate the proposed approach. Due to the characteristic of rich, loamy soils combined with reasonable weather conditions, the Karacabey Plain is one of the most valuable agricultural regions of Turkey. Analyses start with applying an image fusion algorithm on the panchromatic and multispectral image. As a result of this process, 1m spatial resolution colour image is produced. In the next step, the four-band fused (1m) image and multispectral (4m) image are orthorectified. Next, the fused image (1m) is segmented using a popular segmentation method, Mean- Shift. The Mean-Shift is originally a method based on kernel density estimation and it shifts each pixel to the mode of clusters. In the segmentation procedure, three parameters must be defined: (i) spatial domain (hs), (ii) range domain (hr), and (iii) minimum region (MR). In this study, in total, 176 parameter combinations (hs, hr, and MR) are tested on a small part of the area (~10km2) to find an optimum segmentation result, and a final parameter combination (hs=18, hr=20, and MR=1000) is determined after evaluating multiple goodness measures. The final segmentation output is then utilized to the classification framework. The classification operation is applied on the four-band multispectral image (4m) to minimize the mixed pixel effect. Before the image classification, each segment is overlaid with the bands of the image fused, and several descriptive statistics of each segment are computed for each band. To select the potential homogeneous regions that are eligible for the selection of training samples, a user-defined threshold is applied. After finding those potential regions, the training pixels are automatically selected and labelled. Thereafter, those training pixels are utilized in a traditional Maximum Likelihood Classification to classify five crop types namely; corn, tomato/pepper, rice, sugar beet, and wheat. The accuracy of the classification is evaluated in pixel-based manner with the help of a reference map including crop information of the area. Promising results are achieved for pixel-based approach. Based on the error matrices used in the evaluation, overall accuracy of the pixel-based analysis is computed as 89.31%. Similar to the overall accuracies, high individual class accuracies are obtained as well. The results point out that automatically collecting the training samples by extracting representative homogenous areas significantly increases the speed of the classification and minimizes the human interaction. The results also confirm that the proposed approach is highly appropriate for the extraction of representative homogenous training areas.

Land cover change analysis of a Mediterranean area in Spain using different sources of data: Multi-seasonal Landsat images, land surface temperature, digital terrain models and texture

The main objective of this study was to apply a method for the mapping and analysis of land cover changes in a Mediterranean area in southern Spain (Granada Province). The province of Granada is a complex and very heterogeneous area made up of numerous land covers that are difficult to map due to spectral similarities.The inherent difficulty of the mapping of areas with the abovementioned characteristics was addressed by choosing a supervised classification algorithm called random forest and, in addition, by obtaining and incorporating new variables that allowed an improved land cover characterization: multi-seasonal spectral variables corresponding to different stages of land cover phenological development, variables linked to environmental gradients (digital terrain models and land surface temperature) and spatial variability structure textural measures. The same level of accuracy was obtained from the combined use of satellite images with digital terrain models or textural measures. However, the inclusion of the land surface temperature had a more moderate effect and only improved the mapping of some of the land-covers.A post-classification change analysis was conducted by comparing two supervised classifications obtained from the application of the classifier to a selection of different spectral, terrain and textural variables of images from 1998 and 2004. The global accuracy of the change map was 86% (matching percentage). The applied method resulted in a difference in the mapping accuracy of 33% in relation to the method based on the traditional maximum-likelihood classifier and considering only the spectral variables from the Landsat TM-5 satellite sensor.