Mohammad Hossein Rezaei Moghaddam; Keyvan Mohammadzade; Majid Pishnamaz Ahmadi
Abstract
Extended Abstract
Introduction
With their dynamic nature, water resources are essential fortheenvironment and play a vital role in human life, development of communities, and climate change. Water bodies have been declining over time due tothe rapid growth of urbanization, excessive abstraction of ...
Read More
Extended Abstract
Introduction
With their dynamic nature, water resources are essential fortheenvironment and play a vital role in human life, development of communities, and climate change. Water bodies have been declining over time due tothe rapid growth of urbanization, excessive abstraction of water, damming, increasing demand for agricultural products, pollution anddegradationofthe environment. Therefore, monitoring water bodies and retrievingrelated information are essential for management of environmental issues and decision making in this field. Accurate recognitionof water bodiesiscrucialin many applied fields, such as environmental monitoring, production of land cover and land use maps, flood risk assessing and monitoring, and drought monitoring.Modern methods such as object-oriented processing take advantage of remote sensing capabilities to make accurate and precise recognition of water bodies possible. Classical methods on the other hand, cannot accurately classify satellite imagery with similar spectral information merging into each other. This reduces the accuracy of pixel-based classification methods. Therefore, object-oriented processing of satellite images is used in the present study to obtain precise maps for the identification of waterbodies.
Materials and methods
A part of Aji Chai River, near the city of Khajeh in Harris County, has been selected as the study area. The total study area included 28 square kilometers. Based on the aim of the present study, the study area was selected in a way to contain linear features, arable lands, and other topographical and human-madefeatures (shading factor) which interfere with the extraction of water bodies and reduce the classification accuracy. Object oriented methods (the closest neighbor and fuzzy object-oriented methods) were used in the present study to identify and extract water bodies from high resolution images (Sentinel 2A imagery).
Discussion and results
Different functions used in OBIA techniques,such as GLCMtextual features, average number of bands in the image, geometric information (shape, compression and asymmetry), and normalized difference vegetation index(NDVI) were used in the present studyto precisely extract land cover. Moreover, algorithms with the highest membership degree in the class of water bodies were considered as effective factors in classification. Usual methods of extracting and monitoring water bodies use spectral information of pixels, and therefore, have limited ability in distinguishing water bodies from linear features, such as roads, clouds, shaded regions, and residential areas. These methods also have limited capabilities in mountainous areas, especially when they are required to separate water from snow. In other words, these methods cannot separate water bodies from regions with lower albedo. Therefore, the present study takes advantage of object-oriented methods (the nearest neighbor and fuzzy methods) and evaluate their effectiveness in the extraction of water bodies.
Conclusion
In this study, the nearest neighbor and fuzzy object-oriented methods were used to extract water bodies and their efficiencies were compared. To improve the results in the nearest neighbor method, the separation space between the samples was optimized using the FSO algorithm, then the water bodies were extracted with 95% accuracy and a Kappa coefficient of 93%. Findings of the present studyindicated that this method cannot distinguish water bodies from shaded regions, and linear featuressuch as roads, and residential areas, and categorizes these features as water bodies, which reduces the accuracy of the final results. In the next step, water bodies were once more extracted using object-oriented fuzzy model. In this method, membership degrees were first calculated for each sampleand then applied in the classification procedure. High accuracy of the results of this method (overall accuracy of 98% and a kappa coefficient of 96%) indicated the superiority of this method over the previous one (nearest neighbor). In this method, water bodies are completely distinguished from linear features such as roads, as well as shaded regions, clouds and residential areas. The results of this study can be generalized to other rivers and water bodies. Compared to classical methods, object-oriented methods are more time efficient and accurate.
yousef ebadi; Javad Javdan; Mohammad Hossein Rezaei Moghaddam
Abstract
Extended Abstract Introduction Groundwater, its quantitative/qualitative variables, and variability directly affect human life, and thus has always been one of the major topics in scientific and academic research. Due to geographical, climatic and hydrological conditions, and specific patterns of surface ...
Read More
Extended Abstract Introduction Groundwater, its quantitative/qualitative variables, and variability directly affect human life, and thus has always been one of the major topics in scientific and academic research. Due to geographical, climatic and hydrological conditions, and specific patterns of surface water and subsurface water resources exploitation, this country has always faced water scarcity. As a result of global and regional changes in temporal and spatial patterns of rainfall, this has intensified in recent years. Therefore, exploitation of groundwater resources has been considered as an option for supplying agricultural, industrial and drinking water. However, excessive exploitation of these resources will result in their destruction. In recent years, excessive removal of groundwater and reduction of groundwater levels have resulted in some problems like subsidence in some plains. This makes it necessary to study the quantitative and qualitative changes of these resources more clearly. Due to the complex nature of aquifers’ hydrogeological systems, accurate investigation of these resources seems costly and even impossible. Thus in order to achieve a better understanding, it is necessary to use different methods for estimation and evaluation of such variables. Material & Methods Most environmental features are completely continuous in nature, which makes it impossible to measure these features in every part of these environments. Thus, we can generalize measured samples to other areas lacking accurate measurements, and in this way estimate these variables in unmeasured areas. This is also true about quantitative and qualitative variables of groundwater, i.e. by collecting samples from some sections, we can measure different characteristics in these samples. This surface modelling -or in other words, generalization of points to surface- can be achieved with mathematical and statistical relationships and rules. Due to the spatial structure of the measured specimens, geo statistics is used in this regard. In recent years, artificial intelligence models, inspired by the natural nervous system and simulating its function, have yielded a very satisfactory result in groundwater estimation and studies. In order to evaluate the accuracy of geo statistical methods and artificial neural networks, the present study takes advantage of statistics and measurements collected from groundwater level of 46 wells in Shabestar-Sufiyan plain in 2014. Kriging method (geo statistics) and multilayer perceptron neural network method (MLP) were used along with error propagation pattern (BP) to estimate unmeasured features in the study area. MATLAB 2016B was used to perform the neural network modeling and ARCGIS10.5 was used to perform Kriging method and prepare the final maps. In both neural network and kriging models, geographical coordinates of observed wells was used as input and measured water table was introduced as the study goal. Primary data reduces the accuracy of models. Thus, data was normalized before being introduced to the neural network model. After the initial analysis of data dispersion and normalization, logarithmic transfer function was used due to the relative improvement of data in Kriging estimator model. Results & Discussion Results indicate that at the training and testing stage (with Sigmoid tangent activation function (Tansig) and 9 neurons in the middle layer), neural network method (MLP) with a high correlation coefficient (0.96) and root mean square error of 13.18 is more accurate than Kriging method with J-shaped Variogram model, a correlation coefficient of 0.90 and root mean square error of 20.10. Due to realistic results provided by neural network method, it is considered to be a more efficient method in estimation of water table in Shabestar-Sufiyan Plain. This is also consistent with earlier hydrogeological studies (regarding aquifers) performed on the ability and flexibility of Artificial Intelligence models. Conclusion Results obtained from the current research, and previous studies conducted in this field indicate that most artificial intelligence computing models are capable of evaluating and estimating continuous environmental variables. On the other hand, understanding groundwater resources’ conditions is considered to be crucial. Thus, new methods, such as artificial neural networks (ANN) and adaptive neuro-fuzzy inference methods (ANFIS) and fuzzy inference systems (FIS), which provide greater accuracy can help decision makers and researchers in maintenance and improvement of the groundwater status.