فهرست مطالب mehdi akhoondzadeh

A strong earthquake () (34.911° N, 45.959° E, ~19 km depth) occurred on November 12, 2017, at 18:18:17 UTC (LT=UTC+03:30) in Sarpol-e Zahab, Iran. Six different Neural Network (NN) algorithms including Multi-Layer Perceptron (MLP), Recurrent Neural Network (RNN), Gated Recurrent Unit (GRU), Convolutional Neural Network (CNN), Long-Short Term Memory (LSTM) and CNN-LSTM were implemented to survey the four months of GPS Total Electron Content (TEC) measurements during the period of August 01 to November 30, 2017 around the epicenter of the mentioned earthquake. By considering the quiet solar-geomagnetic conditions, every six methods detect anomalous TEC variations nine days prior to the earthquake. Since time-series of TEC variations follow a nonlinear and complex behavior, intelligent algorithms such as NN can be considered as an appropriate tool for modelling and prediction of TEC time-series. Moreover, multi-methods analyses beside the multi precursor’s analyses decrease uncertainty and false alarms and consequently lead to confident anomalies.

The Karun River is the biggest river basin in Iran, which supplies water demands of about 16 cities, several villages, thousands of hectares of agricultural. This river polluted because of domestic and urban sewerage, industrial sources, and irrigation of agricultural land, Hospital sewage and high tide level of Persian Gulf.
Therefore, because of the importance of this river, the water salinity of this river is determined in this study. The traditional methods of determining water salinity are costly in comparison with remote sensing methods.
In the present study, Landsat 8 (OLI) data was used to calculate the water salinity map for Karun River since not only it is free, but it also has an acceptable resolution.

Landsat 8 (OLI) images were used to calculate reflectance for a pixel and were attained from (US Geological Survey (USGS) 2019). First, radiometric correction was applied to normalize satellite images. This process convert Digital Number into radiance. Second, in order to attain the surface reflectance values, the process of atmospheric correction was applied using the Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes (FLAASH).
Water salinity was calculate by Iran Water and Power Recourses Development Company. Eight stations are located in the crucial point for EC measuring ALIKALE, GOTVAND, SHOOSHTAR, SHOTEYT, GARGAR, DEZ, AHVAZ, and ABADAN.
Iran Water and Power Recourses Development Company obtained 102 observed EC samples from June 2013 to July 2018 along the Karun River.
The Support Vector Machine was classically used for classification, Support Vector Classification, but extended for using along with regression issue, namely Support Vector Regression.
The results related to the quality of the SVR depend on some factors: the loss function Ɛ, the error penalty factor C and the kernel function parameters.
Usually, radial basis kernel function (RBF), k(x, x΄) = k(x,x΄)=exp⁡〖( -||x-x΄〗 2/σ^2), has been used in remote sensing studies, so, it is implemented in this study. Finally, the Genetic Algorithm (GA) is employed to optimize some parameters including C, Ɛ and σ.
GA is an optimization technique create by Holland (1975) and discussed the mechanism of GA in solving nonlinear optimization problems.
Besides, the Genetic Algorithm (GA) was applied to determine the best performer bands combination. Furthermore, we employ GA to optimize SVR parameters and number of layers and neurons of MLP neural network in order to maximize model accuracy.

Salinity intrusion is a complex issue in coastal, hot, and dry areas. Currently, remote sensing techniques have been widely used to monitor water salinity changes, ranging from inland river networks to deep oceans. The Karun River basin, with a basin area of 67,000 km^2, is located in the southern part of Iran. The salinity of Karun River has been increasing due to some critical factors, e.g. severe climate condition and regional physiography, industrial sources, domestic and urban sewerage, irrigation of agricultural land, fish hatchery, hospital sewage, and high tide level of Persian Gulf .
This study aimed at building Support Vector Regression (SVR) and Multilayer Perceptron (MLP) models to realize the salinity intrusion through the relationship between reflectance from the Landsat-8 Operational Land Imager images and salinity levels measured in situ. 102 observed samples were divided into 75% training and 25% test.
Besides, the Genetic Algorithm (GA) was applied to determine the best performer bands combination. Furthermore, we employ GA to optimize SVR parameters and number of layers and neurons of MLP neural network in order to maximize model accuracy.
The result showed that the MLP approach was the better model to estimate water salinity along the Karun River network in the study area, which coefficient of determination (R^2) and RMSE of test data is obtained as 0.73 and 390μs 〖cm〗^(-1).
GA analysis proved that bands 1, 2 and 3 are the best for modeling water salinity. In this study, the GA is used to determine the SVR meta-parameters including the loss function Ɛ, the error penalty factor C and σ parameters, which are obtained to be〖1×10〗^(-9), 1099 and 0.96, respectively, and number of layers and neurons of MLP neural network, which are obtained to be 5 and 35, respectively.
The result showed that the MLP approach was the better model to estimate water salinity along the Karun River network in the study area, which coefficient of determination (R^2) and RMSE of test data is obtained as 0.73 and 390μs 〖cm〗^(-1).

The present study calculated the relationship between reflectance retrieved from Landsat-8 OLI and water salinity in the Karun River. SVR and MLP models had acceptable operation by considering the large size, geographic complexity of the study domain and the wide range of field data that change between 385 and 4310μs cm^(-1). Augmentation field data is the critical priority work for future study to probe the relationship between water salinity and satellite images.In addition, the contribution of thermal bands can help to increase accuracy of models. Salinity intrusion is a complex issue in coastal and hot and dry areas. Currently, remote sensing techniques have been widely used to monitor water salinity changes, ranging from inland river networks to deep oceans. The Karun River basin, with a basin area of 67,000 km2, is located in the southern part of Iran. The salinity of Karun River has been increasing due to some critical factors, e.g. severe climate condition and regional physiography, industrial sources, domestic and urban sewerage, irrigation of agricultural land, fish hatchery, hospital sewage, and high tide level of Persian Gulf .This study aimed at building Support Vector Regression (SVR) and Multilayer Perceptron (MLP) models to realize the salinity intrusion through the relationship between reflectance from the Landsat-8 Operational Land Imager images and salinity levels measured in situ. A total of 102 observed samples were divided into 75% training and 25% test. Besides, the Genetic Algorithm (GA) was applied to determine the best performer bands combination. Furthermore, we employ GA to optimize SVR parameters and number of layers and neurons of MLP neural network in order to maximize model accuracy. The result showed that the MLP approach was the better model to estimate water salinity along the Karun River network in the study area, which coefficient of determination (R2) and RMSE of test data is obtained as 0.73 and 390μscm-1

Drought is one of the natural disasters in the world, which is associated with various global factors, most of which can be observed using remote sensing techniques. One of the factors affecting agricultural drought is the vegetation associated with other drought-related factors. These parameters have a complicated relationship with each other, so machine learning algorithms can be used to predict better and model this phenomenon. Factors considered in this study include vegetation as the most critical factor, Land Surface Temperature (LST), Evapo Transpiration (ET), snow cover, rainfall, soil moisture these are derived from the active and passive sensors of satellite sensors as the products of LST, snow cover and vegetation using images of products of the MODIS sensor, rainfall using the images of the TRMM satellite, and soil moisture using the images of the SMOS satellite during a period from June 2010 to the end of 2018 for the central region of Iran. After that, primary processing was performed on these images. The vegetation index (NDVI) is modelled and predicted using an Artificial Neural Network algorithm (ANN), Support Vector Regression (SVR), Decision Tree (DT), Random Forest (RF) for monthly periods. By using these methods we have been able to present a model with desirable accuracy. The ANN approach has provided higher accuracy than the other three algorithms. Also, an average accuracy with RMSE=0.0385 and =0.8740 was achieved.

Due to increase of water exploitation and drought, the need for water resources has risen in past decades. Numerous regions around the world are under threat of environmental crisis, as a result of climate change. Declination in the amount of precipitation can be led to various subsequences, such as significant reduction in the level of ground and surface water, e.g., lakes. Through the development of satellite imagery systems, it is possible to monitor and evaluate changes in rainfall, groundwater level, surface water area, and level.
Numerous studies have been conducted to observe and evaluate climate change after the launch of Gravity Recovery and Climate Experiment (GRACE) satellite mission. GRACE dataset has been used widely to determine water storage variations over the world as well as Iran. This satellite data has been used for various purposes including ground and surface water monitoring. Employing this dataset beside precipitation and satellite altimetry data have been used for observing changes in watersheds and lakes in numerous studies. Modelling and predicting environmental and climate changes are always an important task. Gathering several remote sensing data and predicting them would be helpful mostly for disaster management and also decision making.
Therefore, it is possible to observe and evaluate variation in rainfall, groundwater level, surface water area, and level. In this study, Urmia Lake and its watershed changes were monitored using various satellite data such as TRMM, GLDAS, GRACE, MODIS. Moreover, machine-learning based methods were developed to predict the lake surface changes.

To monitor Urmia lake changes, several data were used to survey variation in precipitation, ground and surface water storage, lake water level, and area in 183 months from April 2002 to June 2017. Sufficient temporal resolution of the data is an essential factor in monitoring of changes through the time. Accordingly, for monitoring the overall change of the Urmia lake, we prefer a satellite data with at least monthly temporal resolution. Therefore, overall variations of the lake and its corresponding basin were modeled using these data with adequate temporal resolution.
Tropical Rainfall Measuring Mission (TRMM) is an international collaboration which aims to observe rainfall for environmental studies. TRMM data provides precipitation in various temporal and spatial resolutions. In this study, TRMM-3b43 level 3 monthly data, with 0.25 degree spatial resolution estimates rainfall in Urmia lake basin, including 83 pixels in each time step.
The GLDAS hydrological model consists of various variables (e.g., soil temperature, soil moisture, precipitation, etc.). In this study, The GLDAS data with 1 degree spatial resolution provides terrestrial water storage (TWS) by integrating soil moisture (kg m-2), snow water equivalent (kg m-2), and canopy water storage (kg m-2). Three types of monthly GLDAS model data (MOS, VIC, and NOAH) were hired for this purpose.
GRACE is a joint missions between Germany and the USA, giving information about mass changes within Earth. The level 2 (RL05) data was of GRACE was used to monitor TWSA, which was computed from spherical harmonics using methods developed by Wahr and Swanson. In addition, a 300 km Gaussian filter was applied to reduce high frequency noises.
The investigated Global Reservoirs and Lakes Monitor (G-REALM) dataset including Jason-1, Jason-2/OSTM, and Jason-3 altimeters was employed to survey Water Level (WL) variation of Urmia lake.
In order to monitor lake extent changes during the 17 years, MODIS atmospheric corrected product MOD09Q1 version 6 data, with 250 meters spatial and 8-day temporal resolution was used through Google Earth Engine. The product provides surface spectral reflectance of bands 1 and 2, which is the composite of 8 products with the absence of clouds, cloud shadow, and aerosol loading. Although, the Normalized Difference Water Index (NDWI) is a common method to separate water from land and it also had the best result on Landsat data, Normalized Difference Vegetation Index (NDVI) performs transcendent distinguishing between water and land while using MODIS data and also in the specific case of Urmia lake. Therefore, in this study, the NDVI index was chosen as an appropriate index to separate water and non-water. To determine lake area, firstly, water region was detected. Then, area of water extent was computed as lake area.
For modeling the lake's area variation, machine learning based methods were investigated. As a time-series prediction problem, a Multilayer Perceptron (MLP) and a Long Short-Term Memory (LSTM) networks were constructed using TRMM rainfall, GLDAS, GRACE TWS, and altimeter WL as inputs (predictors) of the models, and lake's area as Target. About 80% of data was assigned to training, 10% to validation, and the same portion to test. A feedforward MLP including one hidden layer and 5 neurons and a Recurrent LSTM network with same hidden layer and 10 neurons, were obtained. In order to evaluate network's performance, Root Mean Square Error (RMSE) was used. In addition, the delay parameter of 12 months or one year was chosen for estimating future variations.

Except seasonal changes, amount of monthly rainfall during the mentioned period experienced a significant decrease from 2004 to 2008, and then it fluctuates to 2017. The changes in precipitation rate can affect other parameters considerably. As a result, water mass variation obtained from GLDAS data, falls from 2003 to 2008, and after that, similarly to rainfall variation, it fluctuates. However, TWSA computed by GRACE data, after reduction to 2008 and rise to 2010, behaved otherwise, and it went down steadily to 2017. Urmia lake WL declined during the whole period. This decrement was intensified from 2006 to 2010, after that it halted gradually to 2017 as consequence of increase in rainfall rate. Area of the lake decreased from 2004 to 2015, also it faced an extreme fall in 2008. Next, to 2017 the area increased slightly.
Due to a decade drought of Urmia lake, it was in critical circumstance. Consequently, estimating future variation of the lake is necessary. Instead of using physical models or assessing the impact of each parameter on the surface of the lake directly and indirectly, which are complicated tasks, a machine-learning based method is hired. Disregarding the exact relation between factors, this learning based method can determine and model changes. By using two of the most common ANN based methods including MLP and LSTM, variation of the lake during that period was modeled.
MLP and LSTM models reached overall RMSE (for normalized data) of 0.0586 and 0.0511, respectively, which indicates reliability of both models for predicting lake area changes, however LSTM network performed superior specially over test data (RMSE of 0.0487). In addition, to predict Urmia lake's further changes and assess LSTM model capabilities comprehensively, 4 networks were constructed to predict lake area of next 3, 6, 9, and 12 months. Accordingly, result demonstrates LSTM abilities for predicting upcoming year variation of the lake with RMSE of 0.0882 (better than prediction for 6 and 9 months).

Variation in each part of environment and climate (such as rainfall, TWS, WL and area of lakes) affects others. Therefore, it is possible to monitor and model these relations between the parameters. In this study, two ANN methods of MLP and LSTM were investigated to model Urmia lake surface area which the LSTM model performed transcendent. Moreover, LSTM method provides a model which is able to predict the lake area of next 12 months with a high accuracy.
In order to improve the network’s accuracy, it is suggested to increase the number of data and parameters, which are used as network input. It would help the network to implement the training stage with a higher capability to recognize diverse situations properly.

In recent years, thousands of people around the world are affected by earthquake. There are many prospects of doing research on earthquake, that the ultimate goal all the researchers want to achieve is the reduction effects caused by this phenomenon. Activities in recent decades in reducing the effects of natural disasters such as earthquake, cause attention on earthquake precursors. Since satellite data have global coverage, suitable temporal resolution and low cost, they are useful for monitoring earthquake precursors. By launching GRACE mission in 2002, the possibility of measuring gravity field variations in weekly temporal resolution is provided. In this paper, 8 years GRACE Level 2 weekly data (have been smoothed by DDK3 filter) have been analyzed in order to detect abnormal gravity field behavior before large earthquakes. We replaced the Earth’s oblateness values (C20) with those from Satellite Laser Ranging because of their poor accuracy. We know that GRACE stripe errors elongated in north-south direction, hence these strips generate fluctuations in east-west direction. Therefore by taking x-axis (north direction) derivative the of these, variations are dramatically suppressed. So independence of these components of gravitational gradient tensor to GRACE stripy errors, cause increase signal to noise ratio. By this consideration we used just components of gravitational gradient tensor for anomaly detection. However, we must note that horizontal derivative operator shifts the phase of the original anomaly distribution in spatial domain. So the positions of time series computation of two selected components are different. In addition second derivative of gravitational potential amplify high-frequency components of the earth gravity field and hence the gravitational gradient changes delineate more clearly in the rupture line, revealling refined mass redistribution features caused by the earthquake. In order to suppress seasonal variations and isolate seismic effects, we removed seasonal variations (annual and semiannual and S2 tidal wave) from time series using least squares analysis. The time of earthquakes are excluded in the least squares fit. Since a large part of the deformation is in the ocean, the hydrological model (e.g. GLDAS) cannot be used to remove seasonal variations. By considering fact that other preseismic anomaly (e.g. ionosphere precursors) does not occur in the vertical projection of earthquake epicenter, we test outskirt of each epicenter in order to detect the anomaly. In order to search for earthquake anomaly from time series a reasonable range of gravitational gradient variations must be determined. We used median and Inter-Quartile Range (IQR) of data as the first method for anomaly detection in time series. Afterwards, Bagging, Boosting and Random forest models has been proposed in the detection process of prominent gravity field anomalies prior the earthquakes. Gravity field depends of many parameters such as location, tidal force, oceanic variation, etc. So distribution of gravity field variation time series is not normally used. By consideration this fact we cannot use mean and standard division of data for anomaly detection. According to obtained results gravity field anomalies occur within time interval of 2-5 weeks before earthquakes. The results in this study indicate that in each case study, the unusual variations of gravity field have had different sign but the signs of two selected components of gravitational gradient tensor for each case study are the same.

The ionosphere as the upper part of Earth’s atmosphere consists of electrons and atoms affecting the signal propagation in the radio frequency domain. Nowadays, Global Navigation Satellite Systems (GNSS), like GPS, are widely used for various applications. The majority of navigation satellite receivers operate on a single frequency and experience an error due to the ionospheric delay. They compensate for the ionospheric delay using an ionospheric model which typically only corrects for 50% of the delay. An alternative approach is to map the ionosphere with a network of real-time measurements. Global Positioning System (GPS) networks prepare chance to study the dynamics and continuous variations in the ionosphere by complementary ionospheric measurements, which are usually obtained by different techniques such as ionosondes, incoherent scatter radars and satellites. The ionospheric delay is characterized by the Total Electron Content (TEC) along the signal path from the satellite to the receiver. Elimination (or reduction) of the ionosphere effects is possible using dual-frequency receivers by a very useful combination of dual-frequency data known as geometry free combination (L4) as follows: where the L4 signature is derived from L1(1.57542 GHz) and L2(1.22760 GHz) phase observables. Single-frequency users, however cannot take advantage of this combination. So, they have to use a proper ionospheric model to correct the ionospheric delay. Ionosondes (up to an altitude of 1000 km) can determine TEC while GPS measurements give completely information about the topside ionosphere. In this paper, the suitability of Neural Networks (NNs) in order to predict the Total Electron Content (TEC) obtained from Iranian Permanent GPS Network (IPGN) during the low-solar-activity period 2006 has been investigated. TEC has many non-linear variations while the neural network has a significant ability to model and approximate it (Williscroft and Poole, 1996; Hernandez-Pajares et al., 1997; Xenos et al., 2003; Sarma and Mahdu, 2005; Leandro and Santos, 2007). The input space included the day number (DN,seasonal variation), hour (HR,diurnal variation), sunspot number (SSN,measure of the solar activity) and magnetic index (measure of magnetic activity). To make the data continuous, the first two parameters were each split into sine and cosine components, two cyclic as follows: where DNS, DNC, HRS and HRC are the sine and cosine components of DN and HR, respectively. In this paper, the TEC values have been estimated using the PPP (Precise Point Positioning) module of the Bernese over Iranshahr (27˚N, 60˚E). Optimum situation of the neural network include of single hidden layer and eight neurons of inputs layer and fifty neurons of hidden layer and one neuron of output layer. To this end, the single hidden layer feed-forward network with a back propagation algorithm has been designed. An analysis was done by comparing predicted NN TEC(TEC values predicted by the NN model) with TEC values from the IRI2007 version of the International Reference Ionosphere, validating GPS TEC(TEC values calculated from the GPS measurements) with the maximum electron density obtained from ionosonde and calculating the performance of the NN model during equinoxes and solstices. The results show high correlation with GPS TEC and NN TEC. Their Root-Mean-Square Error(RMSE) and coefficient of determination (R2) are 1.5273 TECU and 0.9334 respectively. RMSE is defined as: where N is the number of data points. In table 3, the absolute error (Eabs) is defined as the magnitude of the difference between the NN predicted TEC and the GPS TEC, while the relative error(Erel) is the ratio of the absolute error to GPS TEC and can be represented as a percentage (Habarulema et al, 2007).These errors were calculated as follows: is the absolute error and is the relative error respectively. The difference (100- % gives the relative correction, which indicates the approximate TEC prediction accuracy for the NN model (Leandro and Santos, 2007). An average error of ~11.41% means that the NN can predict about 88.58% of the GPS TEC on average. Results show that the neural network works better rather than the IRI model for IRAN.

Remote sensing techniques made it possible to study thermal anomalies prior to major earthquakes regardless of complications in comprehending earthquake mechanisms. Thermal pre-cursors are one the main resources for earthquake prediction. In this article, land surface temperature, atmospheric temperature, surface latent heat flux and outgoing long-wave radiation have been studied to detect anomalies prior to Varzaghan (August 11, 2012), Boushehr (April 9, 2013) and Saravan (April 16, 2013) earthquakes.
To detect earthquake related anomalies, time series of each pre-cursor has been produced within the period of earthquake, land surface temperature and atmospheric temperature were acquired from MODIS products, surface latent heat flux from GLDAS library and outgoing long-wave radiation from AIRS products. These time series were predicted by an artificial neural network with ant colony optimization training method. The results of this study were compared with artificial neural network with Levenberg-Marquardt training algorithm. It has been shown that 10 to 13 days before Varzaghan earthquake, anomalies has appeared in all of the mentioned precursors, in case of Boushehr earthquake 6 to 9 days before the event, anomalies appeared in atmospheric temperature and outgoing long-wave radiation and also a strong anomaly appeared in surface latent heat flux 2 days prior to earthquake and in Saravan earthquakes anomalies have been detected 5 to 8 days before the earthquake in all of the studied thermal pre-cursors.

Earthquakes show unknown nonlinear behavior and given the magnitude of the earthquake, we would encounter certain changes in lithosphere, atmosphere and ionosphere. The ionospheric parameters have been found to be sorely susceptible to major earthquakes. In addition to the ionospheric variations generated by solar activity, there are remarkable temporary changes in the ionosphere that are generated by prompt changes in geomagnetic activity. Therefore, recognizing the ionospheric anomaly variations generated by seismic activity or geomagnetic activity is hard, exclusively when there is interposition from little geomagnetic storms. Processing the time series of total electron content (TEC), in order to ionospheric anomalies detection is a significant subject. Two wavelet methods were used to nonlinear and non-stationary time series of the TEC: the analytic wavelet transform (AWT) to detect variation in the TEC, and cross wavelet transform method (XWT) to analyze the mutual relationship between the variability of the ionospheric anomalies and the geophysical indices around the epicenter of the earthquake in the time-frequency domain. The Saravan (28.107˚N, 62.053˚E) earthquake happened on 16 April 2013 during the period of high solar activity in the 24th solar cycle. In this study, we utilized the CODE GIMs from 1 March 2013 to 31 April 2013 for the Saravan earthquake. Under quiet geomagnetic condition, the earthquake was considered the only reason of these changes and within 10 to 15 days before the earthquake and 7 days afterward, severe changes were observed. There was a powerful nonlinear context in the TEC data, generated by abnormal solar irradiance during the studied period. It is essential to eliminate the solar activity and geomagnetic activity traces from the ionospheric TEC to elude for representing error in the TEC time series. To recognize if the ionospheric perturbation detected by the AWT is connected to geomagnetic activity, we carried out the XWT for the TEC and AP time series from 1 March to 31 April 2013. It specifies that there is one common high energy region extract within the two time series. The common high energy region related to 17 March 2013. Accordingly, this increment was more probably caused by the geomagnetic storm effects. Within the dynamic range of earthquake, no energetic common point was observed which showed that geomagnetic activity had no role in ionospheric anomalies and another factor, very probably the earthquake was the root of the mentioned anomalies. Therefore, in order to reduce hazard, given TEC time series, the time and frequency of the earthquake could be predicated and defined by evaluating ionospheric parameters.