Geographic Information System (GIS)
Mohammad Karimi; Parastoo Pilehforooshha; Ali Safari
Abstract
Extended Abstract Introduction:The exploration and preparation of the potential map of mineral reserves requires the use of various methods and techniques, based on the geological and mining knowledge of the investigated area, and the use of predictive models of mineral potential (Bonham-Carter, ...
Read More
Extended Abstract Introduction:The exploration and preparation of the potential map of mineral reserves requires the use of various methods and techniques, based on the geological and mining knowledge of the investigated area, and the use of predictive models of mineral potential (Bonham-Carter, 1994; Carranza et al., 2008a). According to the investigations, the common models of map integration that are used in the discovery of mineral reserves in the initial exploration stage include index overlap model, fuzzy operators, weighted indicators and smart methods such as random forests and artificial networks. Determining the values of weights and scores that show the relative importance of the effective factors is the primary requirement in combining the maps and preparing the mineral potential map (Agterberg, 1992; Brown et al., 2000).The purpose of this research is to prepare a potential map of copper deposits in Dehj-Bazman region using two methods of random forest and support vector machine. In addition, in order to compare the potential map of porphyry copper reserves resulting from the random forest method, the support vector machine method and the knowledge-based methods of index overlap and fuzzy logic were used.Materials & Methods:The area studied in this research is a part of the magmatic belt of Kerman region, known as the Dehj-Sardouye belt. The information layers controlling mineralization in Dehj-Bazman area include rock units, structures, alterations, geochemistry, geophysics and copper deposits. In practical applications of machine learning algorithms, mineral potential mapping is essentially a bimodal classification problem, such that each undiscovered area is classified as prospective or non-prospective according to some combination of mapping criteria (Zuo, 2011). The final results are a set of predictive maps that show target areas with high ore formation potential.In order to model, training was done. Before training the random forest model, the input data set and the target variable should be prepared and then the model should be trained. The target variables for entering the random forest model and support vector machine were determined as deposit points (values of 1) and non-deposit points (values of 0). Then the genetic algorithm was used to adjust the parameters.Evaluation of the predictive performance of random forest model and support vector machine can be described by the ambiguity matrix. In this matrix, there are four components, which are defined as: (1) a deposit sample that is correctly classified as a deposit (TP); (2) a deposit sample incorrectly classified as a non-deposit sample (FN), (3) a non-deposit sample correctly classified as a non-deposit sample (TN), and (4) a non-deposit sample that is wrongly classified as a deposit sample (FP) (Liu et al., 2005; Tien Bui et al., 2016): (8) (9) (10) (11) (12) After training and evaluating different models, the best model was obtained by adjusting different parameters and it was used to integrate factor maps in order to predict areas with high potential of porphyry copper deposits. Also, knowledge-based methods of fuzzy logic and index overlap were used to combine factor maps to compare with the results of intelligent methods.Results & Discussion:At this stage, the desired information layers were collected and prepared in the GIS environment, and then factor maps were prepared. Accuracy, sensitivity, specificity, predicted positive value, predicted negative value, kappa index and OOB error were used to evaluate the performance of random forest model and support vector machine. Also, the importance of the predictor variables in the random forest model was evaluated through the mean decrease in accuracy and the mean decrease in node impurity or the Gini impurity index (Breiman, 2001). According to the results, the most important predictor in the random forest model is the geochemical map, while the structures factor has the least impact in predicting the preparation of the mineral potential map with the final random forest model.In the potential maps of porphyry copper deposits obtained from two methods of random forest and support vector machine, the target areas cover 14% of the studied area, in which there are 92% and 87% of known deposits, respectively. Finally, the efficiency of machine learning methods and knowledge-based methods were compared. In order to produce porphyry copper potential map with knowledge-based methods, the judgment of expert experts was used to assign weights to each criterion map. For this purpose, weights of 0.3, 0.25, 0.25, 0.1, 0.1 were assigned to produce maps of alteration factor, geochemistry, geology, geophysics and structures respectively. In the potential map obtained from the method of index overlap and fuzzy logic (fuzzy sum), the areas predicted as copper mines cover 16 and 17 percent of the studied area, respectively, in which 83 and 79 percent of the existing mines are located.Conclusion:This research was conducted with the aim of evaluating and comparing the effectiveness of random forest method and support vector machine method and knowledge-based methods to prepare porphyry copper potential map of Dehaj-Bozman region of Kerman province. Based on the results, the random forest model works well in the field of porphyry copper potential map preparation with geochemical, geophysical, geological, alteration and structures datasets. In addition, the random forest algorithm can estimate the importance of factor maps.The results of this research show that the geochemical factor map is the most important and the structure factor map is the least important in predicting the data-driven model of random forests. This estimate of importance is consistent with geological knowledge about porphyry copper mineralization in Dehj-Buzman region. In order to produce porphyry copper potential map with knowledge-based methods, the judgment of expert experts was used to assign weights to each criterion map. According to the obtained results, the performance of the random forest model is better than the vector machine model, and also, the performance of the support vector machine model is better than the knowledge-based methods.
Mohammad Ghasem Torkashvand; Mostafa Mousapour
Abstract
Extended Abstract
Introduction
The snow cover is one of the quickest changing phenomena on the earth that considerably affects the climate, amount of radiation, the balance of energy between atmosphere and earth, hydrology cycle and biogeochemical as well as human activities. Precise estimate ...
Read More
Extended Abstract
Introduction
The snow cover is one of the quickest changing phenomena on the earth that considerably affects the climate, amount of radiation, the balance of energy between atmosphere and earth, hydrology cycle and biogeochemical as well as human activities. Precise estimate of snow cover is regarded as one of the fundamental operations in precipitation. Thus, monitoring the snow-covered surfaces hasspecial importancefor the perspective of climatic, ecologic and hydrologic studies. The researchers believe that remote sensing data can lead to better assess from the snow-covered areas than traditional topography methods. Therefore, nowadays, in efficient management of water resources, remote sensing data aims to achieve exact information on snow-covered areasis applying operationally. Satellites are suitable tools to measure the mentioned areas since high snow reflection creates a good contrast with other natural surfaces except clouds. This research is conducted to compare the performance of Cornell functions of support vector and object-oriented Fuzzy operators in estimating the desired areas in Almabolaq Mountain, Asadabad.
Material & Methods
The data used in this research are the bands with 10 m spatial segregation of 2B Sentinel satellite including bands 2, 3, 4 and 8 on 6th March 2020. To classify Cornell functions of support vector machine and compute their accuracy, ENVI software was implemented. The eCognation software was used to partition and categorize those with the same object-oriented Fuzzy operators. Separating similar spectral sets and classifying those with the same spectral behaviour are regarded as satellite information classification. In other words, categorizing the photo pixels, and allocating one pixel to one class or phenomenon are the mentioned classification. Support vector machinethat is one of the most common classifiers in learning machine, which divides data using an optimum separation super plate. One of the important advantages of support vector machine is the ability to deal with high dimensional data using almost less training samples for remote sensing applications. Objective analysis is an advanced technique of image processing which is used to assess the digital images and typical conflicts of basic pixel classification based on different methods. Traditionally, pixel-based analysis can be done by available data of each pixel whereas object-based analysis considers a set of similar pixels called objects or image objects. It regards adjacent pixels with the same information value as one distinct unit called piece or segment. In fact, pieces are the areas produced by one or few homogeneous criteria in one or few dimensions of a specific space, so that the pieces have extra spectral information in each band, mean, maximum and minimum amounts, variance, etc. as compared to single pixels. Combining the object-oriented and Fuzzy methods provides the classification of image pieces with a specific membership degree. In this process, image pieces with different membership degrees are classified in more than one class, so according to the membership degree, image piece classification is done leading to the increased final precision.
Results & Discussion
In this research, after preparing satellite images in SNAP software using Sen2Cor, radiometric correction was conducted on the images. To prepare the classification map of Cornell functions of support vector machine, TIFF satellite images were called by ENVI software. Using the shape file of the case study, the area cutting operation was done. Afterwards, two classes of snow and non-snow regions were created to pick up the training points, so based on imagery processing, training points were specified for each class. To classify support vector machine algorithm, linear, polynomial, radialand sigmoid Cornell functions were applied,soclassification maps were separately produced. To draw the classification map of object-oriented Fuzzy operators, satellite images pre-processed in previous stages were called by eCognation software, then they were defined as a project. Afterwards, two mentioned classes were defined to do the classification process, for each class, the desired Fuzzy operator was determined. For suitable classification, it was done in various scales and weight coefficients of shape and compactness. Scale 75, shape 6.0 and compactness 8.0 presented suitable classification. The training samples, parameters of lighting, mean and standard deviations were chosen as distinct features of classes for object-oriented classification. Using the nearest adjacent neighbor algorithm, object-oriented classification was done for each of the Fuzzy operators. After drawing the snow-covered areas through Cornell functions of support vector machine and object-oriented fuzzy operators, the accuracy of classification was computed.
Conclusion
The results indicate that AND algorithm showing the logic share and minimum return value out of Fuzzy values is the highest accuracy (98%) and to classify digital images,the object-oriented processing methods of satellite imagery enable more precision due to the data related to texture, shape, position, content and geometrical features as compared to Cornell functions of support vector machine.
Sara Attarchi; Najmeh Poorakbar
Abstract
Extended Abstract
Introduction
Free access to the Landsat dataset and Sentinel 2 images has provided a great opportunity for long-term monitoring of resources. Landsat 8 was launched in 2013 to continue the mission of the previous Earth observation satellites. Landsat 8multi-spectral sensor, Operational ...
Read More
Extended Abstract
Introduction
Free access to the Landsat dataset and Sentinel 2 images has provided a great opportunity for long-term monitoring of resources. Landsat 8 was launched in 2013 to continue the mission of the previous Earth observation satellites. Landsat 8multi-spectral sensor, Operational Land Imager (OLI), provides multi-spectral images with 30-meter resolution. Sentinel 2 was launched in 2015 with a multispectral sensor called MSI which captures images with different spatial resolutions (10m to 60m). The secret mission of Landsat satellites started in the 1970s and they have the longest archive of satellite images collected from the Earth. Sentinel 2 offers higher spatial, spectral and temporal resolutions and therefore it is important to compare the compatibility of Sentinel 2 and Landsat 8 images. OLI and MSI sensors both operate in the optical region, thus weather conditions can impose some limitations on their data acquisition. In such circumstances, data collected by a compatible and similar sensor can replace the cloud-covered images.
Generally, spectral features of new sensors are designed in such a way toconform to the corresponding bands of the previous sensors. The present study compares the corresponding bands of MSI and OLI sensors. The efficiency of both sensors in the classification of a heterogeneous and complex region has also been investigated.
Materials & Methods
Three near-simultaneous pairs of Landsat 8 and Sentinel-2 scenes were obtained to conduct a comparative study. Images were acquired in August 2017, November 2017, and July 2018.Minudasht - in northern Iran- was selected as the study area because of the presence of different land cover classes including rainfed agricultural lands, irrigated agricultural lands, forests, residential areas, and bare lands.Thescenes were processed for further analysis. First, the scenes were atmospherically corrected. In the next step, spatial resolution of MSI bands was resampled to 30 m, and each pair of mages were geometrically co-registered. To do so, 10 tie points were selected, and scenes were co-registered usingthe first-degree polynomial method. RMSE values were reported 2.5 m, 2.4 m, and 2.8 m for August 2017, November 2017, and July 2018, respectively. To investigate the similarities and differences of the sensors’ spectral content, the correlation between corresponding bands of the two sensors was estimated.
Then, images were classified using the support vector machine (SVM) algorithm. Five distinct land cover classes were found in the region including rainfed agricultural land, gardens and irrigated agricultural land, forests, residential areas, and bare lands. The training samples were selectedfromthe land use map and high-resolution Google Earth images. Approximately 300 training samples were selected for each land cover class. The accuracy of classification results was compared to verify the efficiency of two sensors in land cover mapping. Independent validation samples were selected for each class. Overall accuracy, commission error, and omission error were calculatedbased on the confusion matrices.
Results & Discussion
The reported correlation coefficientfor all corresponding bands was higher than 0.8. Results indicate a high level of similarity between the two sensors. Similar findings were reported by previous studies. Overall classification accuracy ofOLIimagescollected in August 2017, November 2017, and July 2018 was 91. 35 %, 89.60 %, and 93.12%, respectively. Overall classification accuracy ofMSI images collected inAugust 2017, November 2017, and July 2018 was 94.76 %, 95.55 %, and 94.07%, respectively. As it is obvious, Sentinel 2showed a higher performance in comparison to Landsat’s, because of its higher spatial resolution. A medium spatial resolution image collected from a complex landscape is often composed of mixed pixels, since different land cover types exist in one pixel. As the image’s spatial resolution improves, the dimensions of each pixeldecrease. Therefore, the number of mixed pixels will decrease and a higher classification accuracy will be expected.
Conclusion
Results confirm the similarity of two sensors in land cover classification. However, the findings could not be extended to other applications. MSI sensorslacka thermal bandand thus are not applicable when such a feature is needed (for an instance inthe retrieval of land surface temperature). In such applications, MSI cannot substitute OLI. For further studies, it is necessary to compare the performance of these sensors in different regions, since different land cover types may impactclassification results. Findings of the present study may raise attention to the differences between Landsat 8- OLI and Sentinel 2 MSI. Further studies can be conducted to investigate the differences between these two sensors. The possible similarities of othersimilar sensors can also be a topic for further investigations.
Ali Asghar Alesheikh; Saeed Mehri
Abstract
Extended Abstract
Introduction
Oak is a common species in Iran and the most important one in Zagros forests. Zagros forests play a crucial and effective role in water supply, soil conservation and climate modification in Iran. Unfortunately, a significant part of those forests suffer from oak decline. ...
Read More
Extended Abstract
Introduction
Oak is a common species in Iran and the most important one in Zagros forests. Zagros forests play a crucial and effective role in water supply, soil conservation and climate modification in Iran. Unfortunately, a significant part of those forests suffer from oak decline. Oak decline (or oak mortality) is a widespread phenomenon in oak forests around the world, which has gained the attention of many researchers in forestry over the past decade. In Iran, this phenomenon was first observed in Zagros forests in 2013. Factors affecting oak decline and their mutual interactions are not clearly identified, which makes understanding and modeling of these processes challenging. Only a few studies have been performed in relation to this phenomenon in Iran. Thus, we chose to determine the most effective parameters and find the best modeling method for oak decline in Iran and especially in Lorestan province.
Materials & Methods
In order to find effective environmental variables, related literature review was thoroughly investigated. Environmental parameters including temperature, precipitation, elevation, slope, direction, soil type, and amount of aerosols were selected as basic influencing parameters. All parameters were then interpolated to produce raster data with 30-meter cell resolution. To find the optimal combination of the parameters, four operators including multiplication, logarithm, hyperbolic transformations, and principal component analysis (PCA) were used. A total 385 different combinations of the influencing parameters were produced using the above mentioned operators. The relation and weight of each parameter are unknown, thus Artificial Neural Networks were used to model oak decline process. Three feed forward artificial neural network, including Back-propagation Neural Network (BP), Probabilistic neural network (PNN) and Support Vector Neural Network (SVNN) were selected as modeling methods. Then, 385 different combinations of the influencing parameters were used in the above mentioned models. To train and evaluate each neural network, a total number of 10000 samples were randomly selected from the study area. 70 percent of these random samples were used to train, 15 percent to evaluate and 15 percent to validate the models. Also, cross-validation method was used to avoid over fitting of neural networks. Finally, 1155 created NN models were compared using R parameter to find the best configuration for modeling oak decline and identifying the most influential environmental parameters in oak decline.
Results & Discussion
Evaluating 1155 different networks indicated that Probabilistic neural network (R=0.87) with 6 inputs, including 1) elevation, 2) slope, 3) direction, 4) aerosols, 5) soil type and 6) principal component of temperature and precipitation, performed better than SVNN and BP in modeling oak decline. Moreover, using different combinations of influencing factors improved the results and increased correlation coefficient (R) of optimal inputs by 0.05 as compared to initial inputs. Thus, it can be concluded that increased number of inputs does not necessarily guarantee a better performance. Furthermore, two principle parameters of temperature and perception have a more significant role in modelling drought stress as compared to other parameters.
Conclusion
Oak decline is a complicated phenomenon and different factors contribute to its occurrence. The present study investigates all environmental parameters affecting oak decline through a comprehensive literature review. Results indicate appropriate performance of probabilistic neural networks in modeling oak decline. Moreover, principal component analysis is considered to be a useful tool for modeling of drought stress in oak trees. Due to different accuracy and precision of these neural networks, it is necessary to evaluate different configurations. For further researches, it is suggested to use other parameters, such as distance from population centers, water table, age of oak trees, oak tree height and characteristics of other nearby trees.
Alireza Arofteh; Taher Reza Mohammed; Ali Hossingholizade; Ehsan Hoghoghi fard
Abstract
Increasing development of urban areas, the need for various information from the urban environment, and also technological advancements have increased the importance of automatic and semi-automatic classification and identification of this type of land cover. The diversity of remote sensing data have ...
Read More
Increasing development of urban areas, the need for various information from the urban environment, and also technological advancements have increased the importance of automatic and semi-automatic classification and identification of this type of land cover. The diversity of remote sensing data have created a wide scope for urban feature detection. Moreover, by launching satellite sensors with a spatial resolving power of less than 1 meter, a dramatic revolution has occurred in the tendency of remote sensing researchers toward classification of urban features. The existence of various features and different applications of spatial information in urban areas have made it possible to integrate various data sources with the aim of identifying different urban features. The present study seeks to integrate optimal properties extracted from optical and LiDAR data in order to identify urban features in the study area. In this regard, colored features, normal difference vegetative index (NDVI), first-order statistical texture in three windows of 5×5, 7×7 and 9×9, second-order statistical texture in three windows of 7×7, 11×11 and 15×15 extracted from the multispectral optical data were calculated along with features of normalized difference index (NDI), slope, slope direction, profile curve, surface curve, roughness, variance, laplacian, smoothness and normalized digital surface model (nDSM) extracted from the LiDAR data. Since increased amount of information has made the process of identifying features in the region time-consuming, the present study applies intelligent genetic algorithm to select optimal features from the calculated features. A total number of 361 features were produced from this data, including 9 colored features, a vegetation index, 144 first-order statistical texture, and 192 second-order statistical texture from multispectral optical data and 14 features from LiDAR data. Then, 17 features including seven features of the LiDAR data and 10 features of the multispectral optical data were determined using genetic algorithm as the optimal features for more appropriate identification of urban features. Finally, support vector machine (SVM) classification method was used to identify the desired features. Results indicate that compared to LiDAR data, multispectral optical data have a better performance in classifying vegetation features, while LiDAR data have been more suitable for the classification of road and building features. In other words, multispectral optical data work appropriately in identifying features with different radiometric information, while classification of features with similar radiometric information, such as roads and buildings is problematic. Thus, LiDAR elevation data help in identification of these features. Additionally, using optimal features along with the primary bands have increased the accuracy of urban features classification. Using optimal features and initial data, the accuracy of support vector machine algorithm classifier in the study area is calculated to be 88.734, which shows 25.438% improvement compared to the initial multispectral optical data classification, and 18. 236% improvement compared to the initial LiDAR data classification.
Faeze Eslamizade; Heidar Rastiveis
Abstract
Extended abstract Introduction Given the population growth and increasing urbanization, the occurrence of natural disasters like earthquake can cause heavy losses and damages and interrupt the development of cities and countries. Among these disasters, the earthquakeisof great importance due to its unpredictability ...
Read More
Extended abstract Introduction Given the population growth and increasing urbanization, the occurrence of natural disasters like earthquake can cause heavy losses and damages and interrupt the development of cities and countries. Among these disasters, the earthquakeisof great importance due to its unpredictability and high frequency in relation to other events, as well as its location on the earthquake belt. According to the last year's estimate, Iran has been one of the 6 countries with high mortality rates in earthquakes. Therefore, finding a way to minimize the losses can be critical. Crisis managers need quick information from the affected area after the earthquake to minimize the fatalities and financiallosses. The destruction map is one of the information that helps crisis managers. These maps show the destructed buildings or roads with their degree of destruction. With these maps, the destructed buildings and roads can be found quickly. Materials & Methods Many methods are used to prepare the destruction maps, such as aerial/satellite images, LiDAR data, etc. These information can be used to determine the destructed buildings automatically or by visual interpretation. Visual interpretation for determining the degree of destruction requires operator. Although this method has high accuracy, it is less considered because it is time consuming and needs specialists to interpret the data. Therefore, researchers have focused on automated processing techniques for the production of the destruction maps. Various automatic change detection techniques are used to evaluate the destruction resulting from earthquakeby comparing satellite images in two pre and post-earthquake periods based on satellite and aerial images. LiDARdata is one of the most important sources of information to determine destructed buildings with high accuracy and speed. LiDAR data provides the possibility of 3-D demonstration of the destructed region. This information is a great help in preparing the destruction map automatically. The recent expansion of the LIDAR technology is due to the high spatial power of these data. As a result, many researchers have focused on developing an automatic destruction map using Lidardata.Although considering the textural information from the Lidar data, like homogeneity in the destructed region can be effective in distinguishing between the destroyedand undestroyed buildings. In this paper, a new algorithm is proposed to prepare the destruction map after the earthquake by integratingthe post-event high resolution satellite images and post-event LiDAR data. In the proposed method, different textural descriptors of the LiDAR image and data are extractedafter the necessary preprocessing on the satellite image andLiDAR data after the earthquake. In the next step, using the layer of buildings extracted from the map,the areas of the buildings are extracted from the satellite image and LiDAR data, as well as the satellite image descriptors and LiDARdata.Then, the textural descriptorsextracted from the satellite image and LiDAR data are combinedtogether. After that, the points inside this area are categorized into two classes of "debris" and "intact" by the method of support vector machine. Finally, based on the area of the debris class of each building, destroyed and undestroyed buildings were identified by taking a threshold limit into consideration. This algorithm is executed on each building from the destruction part to produce the final destruction map Results&Discussion In order to evaluate the proposed method,the data set was selected from the city of Port-au-Prince, the capital of Haiti, after the 2010 earthquake. According to the USGS reports, 97,294 buildings were damaged and 188,383 were destroyed in Port-au-Prince and most of the southern parts of Haiti. Furthermore, reports show that 222,570 people were killed, 300,000 were injured, and 1.3 million people were displaced. The sample data set include post-event WorldView II satellite images as well as post-event LiDAR data. The WorldView II satellite took images on January, 16 2010, and the LiDAR date was also obtained from this topography website. Obtaining LiDAR data is from January, 21 2010 to January, 27 2010. The vector map of the selected test area was generated in ArcGIS environment. By evaluating the proposed method and using the existing data, the overall accuracy of 97% and the Kappa coefficient of 92% were obtained which proved the reliability of this technique. Conclusion In this paper, a new method for the generation of damage map based on the integration of high resolution satellite images and LiDAR data was proposed. The results show the ability of this method in generating destruction maps based on the satellite images with high resolution and LiDAR data. In comparing similar studies, the results are satisfactory. The selection of the appropriate descriptors, correct training data, the elimination of non-building areas from the sample data, the integration of satellite images and LiDARdate can be known as the reason behind obtaining these results.
Ali Shojaeeian; Sadegh Mokhtari Chelche; Leila Keshtkar; Esmaeil Soleymani rad
Abstract
Nowadays, remote sensing data is able to provide the latest information for the study of land cover and land uses. These images are of high importancedue to the presentation of timely information, diversity of forms, being digital and the possibility of processing in the preparation of user maps.Determining ...
Read More
Nowadays, remote sensing data is able to provide the latest information for the study of land cover and land uses. These images are of high importancedue to the presentation of timely information, diversity of forms, being digital and the possibility of processing in the preparation of user maps.Determining the land cover will be of great help to the area managers to make decisions. In this regard, the purpose of this researchis to compare the efficiency of parametric (least distant and box) and nonparametric (supporting vector machine) methods in land cover classification by using Landsat 8 satellite images in part of Dezful city. The nature of this research has been developmental-practical and its method has been descriptive-analytical. For this purpose, satellite data including Landsat 8 satellite images (13/8/2013) were prepared and analyzed using ENVI software. The efficiency of each classification method was investigated by calculating the two general accuracy and kappa coefficient. The results of the comparison of the methods used in the research showed that the SVM algorithm, especially the three linear, radial and polynomial kernels, had a better and more desirable accuracy than the parametric methods with 97.15%, 95.89% and 95.63% respectively. This study confirms the efficiency and more desirable capability of SVM algorithms in the classification of remote sensing images compared with parametric methods.
