JWSS - Journal of Water and Soil Science

---

In this study the impact of climate change on temperature, rainfall and river flows of the Zayandeh Rud basin under two climate change scenarios for two periods (2010-2039 and 2070-2099) are investigated. For the evaluation of future climate change impact on stream flow to Chadegan reservoir, the global circulation model (GCM) outputs of the HadCM3 model (monthly temperature and precipitation) with two scenarios, A2 and B2, are obtained and downscaled to the local level for the selected time periods. The results indicate that the annual average of precipitation decreases and temperature increases for both periods that are more pronounced for the period 2079-2099. Such that 10% to 16% decrease in precipitation and 3.2 to 4.6ºC increases in temperature can be anticipated for scenarios A2 and B2, respectively. To predict future stream flow changes due to climate change, artificial neural networks (ANNs) have been applied and trained by the several input models and architectures for rainfall-runoff simulation. The results indicate that the maximum of 5.8% decrease in the annual flows. Comparison of the two scenarios indicates the more critical situation in scenario A2 for the basin.

---

In this study wholesale prices of selected crops, namely, tomato, onion and potatoes in Fars province were predicted for various time horizons by using common methods of forecasting and artificial neural networks (ANN). Monthly data from September 1998 to June 2005 period were obtained from Ministry of Jihad-e Agriculture. For comparing different methods data selected from September 1998 to December 2004 were utilized, and latest six - month data were mainly used to monitor the power of prediction. The MAE, MSE and MAPE criteria were used for comparing the ability of different forecasting methods. Results of this study showed that ANN had the lowest error in prediction of prices for one - to three - month periods, but for six - month prediction, all forecasting methods were not statistically different.

---

  Air temperature prediction models using satellite data are based on two variables of land surface temperature and vegetation cover index. These variables are obtained by atmospheric corrections in the values for the above data. Water vapor, ozone, and atmospheric aerosol optical depth are required for the atmospheric correction of visible bands. However, no measurements are available for these parameters in most locations of Iran. Using the common methods, land surface temperature can be measured accurately at 2 ° C. Given these limitations, efforts are made in this study to evaluate the accuracy of predicting maximum air temperature when uncorrected atmospheric data from the NOAA Satellite are used by a neural network. For this purpose, various neural network models were constructed from different combinations of data from 4 bands of NOAA satellite and 3 different geographical variables as inputs to the model in order to select the best model. The results showed that the best neural network was the one consisting of 6 neurons as the input layer (including 4 bands of NOAA satellite, day of the year, and altitude) and 19 neurons in the hidden layer. In this structure, about 91.4% of the results were found to be accurate at 3 ° C and the statistical criteria of R², RMSE, and MBE were found to be 0.62, 1.7 ° C, and -0.01 ° C, respectively.

---

Soil fertility measures such as cation exchange capacity (CEC) may be used in upgrading soil maps and improving their quality. Direct measurement of CEC is costly and laborious. Indirect estimation of CEC via pedotransfer functions, therefore, may be appropriate and effective. Several delineations of two consociation map units consisting of two soil families, Shahrak series and Chaharmahal series, located in Shahrekord plain were identified. Soil samples were taken from two depths of 0-20 and 30-50 cm and were analyzed for several physico-chemical properties. Clay and organic matter percentages as well as moisture content at -1500 kPa correlated best with CEC. Pedotransfer functions were successfully developed using regression and artificial neural networks. In this research, it seemed that one hidden layer with one node was sufficient for all neural networks models. The best regression model consisting of organic matter and clay variables showed R2=0.81 and RMSE=7.2 while best corresponding neural network with a learning coefficient of 0.3 and an epoch of 40 had R2=0.88 and RMSE=0.34. Data partitioning according to soil series and soil depths increased the accuracy and precision of the functions. Compared to regression, artificial neural network technique gave pedotransfer functions with greater R2 and smaller RMSE.

---

In mountainous basins, snow water equivalent is usually used to evaluate water resources related to snow. In this research, based on the observed data, the snow depth and its water equivalent was studied through application of non-linear regression, artificial neural network as well as optimization of network's parameters with genetic algorithm. To this end, the estimated values by artificial neural network, neural network-genetic algorithm combined method and regression method were compared with the observed data. The field measurement were carried out in the Samsami basin in February 2006. Correlation coefficient (r) mean square error (MSE) and mean absolute error (MAE) were used to evaluate efficiency of the various models of artificial neural networks and nonlinear regression models. The results showed that artificial neural network and genetic algorithm combined methods were suitable to estimate snow water equivalent. In general, among the methods used, neural network-genetic algorithm combined method presented the best result (r= 0.84, MSE= 0.041 and MAE= 0.051). Of the parameters considered, elevation from sea level is the most important and effective to estimate snow water equivalent.

---

Daily solar radiation intercepted at the earth’s surface is an input required for water resources, environmental and agricultural studies. However, the measurement of this parameter can only be done in a few places. This has led researchers to develop a number of methods for estimating solar radiation based on frequently available meteorological records such as hours of sunshine or air temperature. In this study two empirical Angestrom and Hargreaves- Samani models, which are respectively based on air temperature and sunshine duration were calibrated and evaluated for estimating solar radiation in southeast of Tehran, Iran. Also, two neural networks models were presented using similar inputs and above-mentioned empirical models. The results showed that the both empirical and neural network models provided closer agreement with the measured values, but the models based on sunshine hours gave better estimates than the models based on air temperature. The neural network model based on sunshine hours with a R2 of 0.97 and a RMSE of 1.34 MJ m-2 d-1 provided the best results

---

Estimating spatial distribution of precipitation is vital to execute water resources plans, drought, land-use plans environment, watershed management, and agricultural master plans. High variation in amount of precipitation in various parts, lack of measurement stations, and the complexity of relationship between precipitation and parameters affecting it have doubled the importance of developing efficient methods in estimating spatial distribution of precipitation. Artificial neural network has been proved to be efficient as a new way for modeling and predicting the processes for which no solution and explicit relationship has been available in accurately identifying and describing them. The purpose of this study is to investigate the efficiency of artificial neural network in estimating spatial monthly precipitation. To achieve this objective, neural network with multilayer perceptorn topology was employed for preparing model for spatial monthly precipitation in five synoptic and rain-gauge stations located in Kurdistan province. In order to design the topology of the model in each station, as the adjustable parameters (including transfer function, learning rule, amount of momentum, number of hidden layers, number of neurons of the hidden layers, and the number of epochs) changed, different neural networks were made and carried out. In each case, the topology with the minimum amount of root mean square error (RMSE) was selected as the optimal model. Owing to the fact that the selection of each of the variable parameters of neural network necessitated recurring trails and errors, and consequently teaching a large number of networks with various topologies, genetic algorithm method was utilized for finding the optimization of these parameters the efficiency of this method, too, was examined in terms of the optimization of neural network. The findings indicated that neural network enjoys a high degree of accuracy in modeling and estimating spatial distribution of monthly precipitation. In addition, combining it with genetic algorithm method was positively evaluated in optimizing the requirements for executing neural network. In most cases, mixed method proved its superiority over executing neural network without optimization. The most precise model in all of the stations under study was achieved by the use of transfer function, sigmoid, learning rule of Levenberg Marquardt in the selected models, the determination coefficient (R2) observed between the model output amounts and the data observed in station were found to be 0.86 0.89 0.94 0.77 and 0.94.

---

Drought is a natural feature of the climate condition, and its recurrence is inevitable. The main purpose of this research is to evaluate the effects of climatic factors on prediction of drought in different areas of Yazd based on artificial neural networks technique. In most of the meteorological stations located in Yazd area, precipitation is the only measured factor while generally in synoptic meteorological stations in addition to precipitation some other variables including maximum and mean temperature, relative humidity, wind speed, dominant wind direction and the amount of evaporation are also available. In this research it was tried to evaluate the role of the type and number of meteorological factor (as inputs of ANN model) on accuracy of ANN based drought prediction. Research area is a part of Yazd province containing only one synoptic and 13 non-synoptic meteorological stations. Three-year moving average of monthly precipitation was the main input of the models in all stations. The type of ANN used in this study was time lag recurrent network (TLRN), a dynamic architecture which was selected by evaluation of different types of ANN in this research. What was predicted is the three-year moving average of monthly precipitation of the next year, which is the main factor to evaluate drought condition one year before it occurs. For the Yazd synoptic meteorological station, several combinations of input variables was evaluated and tested to find the most relevant type of input variables for prediction of drought. However, for other 13 stations precipitation data was the only variable to use in ANN models for this purpose. Results in all stations were satisfactory, even where only one input (precipitation) was used to the models, although the level prediction accuracy was different from station to station. Result taken from this research, indicates high flexibility of ANN to cope with poor data condition where it is difficult to get acceptable results by most of the methods.

---

In this research, based on the observed data of Class A pan evaporation and application of non-linear regression (NLR), artificial neural network (ANN), neuro-fuzzy (NF) as well as Stephens-Stewart (SS) methods daily evaporation of Kerman region was evaluated. In the cases of NLR, ANN and NF methods, the input variables were air temperature (T), air pressure, relative humidity (RH), solar radiation (SR) and wind speed (U2) which were used in various combinations to estimate daily pan evaporation (Ep) defined as output variable. Performance of the methods was evaluated by comparing the observed and estimated data, using determination coefficient (R2), root mean square error (RMSE) and mean absolute error (MAE). Based on the observed data at Kerman meteorological station, the monthly and annual average evaporation values of the region were 272 and 3263 mm, respectively. The results of this study indicated that NF method is the most suitable method to estimate daily Class A pan evaporation. The statistics criteria of this model which is constituted based on the 5 input parameters were R2 = 0.85, RMSE=1.61 and MAE= 1.24 mm day-1. The sensitivity analysis of NF model revealed that the estimated EP is more sensitive to T and U2 (as the input variables), respectively. Due to weak accuracy of SS method, a new modification step of the model was also developed based on the SR and T in order to have a more exact daily evaporation estimation of the region. However, the result of the modified model was not acceptable

---

Application of drought early warning system is an important strategy for drought management. It is more pronounced in the arid regions where dams have vital role to overcome water shortages. This papers aims to develop and apply such a system that includes three main components, which are 1) drought monitoring, 2) forecasting inflows and water demands and 3) calculation of a warning index for decision about drought management. The system is presented for the Zayanderud Dam. For this, the future six months river inflows and demands are forecasted at different probabilistic levels using the artificial neural networks and considering respected uncertainties. Also, five drought levels are indicated based on the historical records of dam’s storage and the self organizing feature map technique. Furthermore, a drought alert index (DAI) is defined using current storage of dam and the forecasted flows and demands. Finally, the different alert levels are estimated, which vary from normal to sever water scarcity. The results showed that application of the designed warning system can have effective role in the dam’s operation, rationing policy and reducing drought losses.

---

Direct measurement of soil hydraulic characteristics is costly and time-consuming. Also, the method is partly unreliable due to soil heterogeneity and laboratory errors. Instead, soil hydraulic characteristics can be predicted using readily available data such as soil texture and bulk density using pedotransfer functions (PTFs). Artificial neural networks (ANNs) and statistical regression are two methods which are used to develop PTFs. In this study, the multi-layer perceptron (MLP) neural network and backward and stepwise regression models were used to estimate saturated hydraulic conductivity using some soil characteristics including the percentage of particle size distribution, porosity, and bulk density. Data of 125 soil profiles were collected from the reports of basic soil science and land reclamation studies conducted by Khuzestan Water and Power Organization. The results showed that MLP neural network having Bayesian training algorithm with the greater coefficient of determination (R2=0.65) and the lower error (RMSE =0.04) had better performance than multiple linear regression model in predicting saturated hydraulic conductivity.

---

Global solar radiation (Rs( on a horizontal surface in the estimation of evapotranspiration of plants and hydrology studies is an important factor. Average daily global solar radiation on a horizontal surface was estimated by artificial neural networks (ANNs) and five empirical models including FAO (No.56), Hargreaves-Samani, Mahmood-Hubard, Bahel and Annandale. The weather data was selected from Karaj, Shiraz, and Ramsar weather stations, which have arid, semi arid and very humid climates (based on De Martonne classification). Daily solar radiation was measured at the three sites selected. The ANN, with actual duration of sunshine and maximum possible duration of sunshine as input parameters, generated daily solar radiation estimates with highest level of accuracy among all models tested. Rs estimates by ANNs with only temperature indices as input and by Hargreaves-Samani, Annandale and Mahmood-Hubard, which are all temperature oriented models, had lower accuracy at all three sites. In contrast, ANNs with actual duration of sunshine and maximum possible sunshine hours as inputs in Karaj, Shiraz and Ramsar station with root mean square error (RMSE) of 2.08, 1.85 and 2.05 Mj m-2 day-1 respectively were the best models. After ANNs, FAO-56 model which is based on sunshine hours produced results closer to the measured values. Rs estimates by ANNs with only temperature indices as input and by Hargreaves-Samani, Annandale and Mahmood-Hubard which are all temperature oriented models, had lower accuracy at all the three sites. These models are not appropriate for estimating daily global solar radiation.

---

With the advent of advanced geographical informational systems (GIS) and remote sensing technologies in recent years, topographic (elevation, slope, and aspect) and vegetation attributes are routinely available from digital elevation models (DEMs) and normalized difference vegetation index (NDVI) at different spatial (watershed, regional) scales. This study explores the use of topographic and vegetation attributes in addition to soil attributes to develop pedotransfer functions (PTFs) for estimating soil saturated hydraulic conductivity in the rangeland of central Zagros. We investigated the use of artificial neural networks (ANNs) in estimating soil saturated hydraulic conductivity from measured particle size distribution, bulk density, topographic attributes, normalized difference vegetation index (NDVI), soil organic carbon (SOC), and CaCo3 in topsoil and subsoil horizon. Three neural networks structures were used and compared with conventional multiple linear regression analysis. The performances of the models were evaluated using spearman’s correlation coefficient (r) based on the observed and the estimated values and normalized mean square error (NMSE). Topographic and vegetation attributes were found to be the most sensitive variables to estimate soil saturated hydraulic conductivity in the rangeland of central Zagros. Improvements were achieved with neural network (r=0.87) models compared with the conventional multiple linear regression (MLR) model (r=0.69).

---

Land use classification and mapping mostly use remotely sensed data. During the past decades, several advanced classification methods such as neural network and support vector machine (SVM) have been developed. In the present study, Landsat TM images with 30m spatial resolution were used to classify land uses through two classification methods including support vector machine and neural network. The results showed that SVM and neural network with the total accuracy of 90.67 % and 91.67% are superior. SVM had a better performance in separating classes with similar spectral profiles. In addition, SVM showed a better performance in delineating class borders in comparison with neural network method. In summary, both SVM and neural network showed satisfactory results but the method of support vector machine proved better with a difference of 1% and 2% in overall accuracy and kappa coefficient, respectively. This was an expected outcome because SVMs are designed to locate an optimal separating hyperplane, while ANNs may not be able to locate this separating hyperplane.

---

Since the development of surface water control needs accurate access to flow behavior of sediment rates, the lack of sediment measurement stations, the novelty of most stations and the lack of statistics on the deposit make it difficult to properly evaluate and simulate the flow behavior and their sediments. In a watershed, the morphological characteristics and sediment load of flow affect each other. It is, thus, important to know about the extent of this relationship to manage and control the flow in downstream areas. In the present study, using artificial neural networks and sediment rating regression methods based on the data from 136 events and also morphological parameters, we have attempted to predict the sediment load of Bagh Abbas basin. In the first step, we used flow data to predict the sediment load of both methods, and then basin morphological characteristics such as the compactness factor and form factor were added to the models. The results of this study showed that by using neural networks of Multilayer Perceptron (MLP) type with Levenberg – Marquardt algorithm and the stimulation function of tangent Sigmoid with two hidden layers and four neurons in each layer, we can predict suspended sediment discharge rate with a sufficient accuracy. Accuracy of the results obtained from the ANN method was higher than the accuracy of rating curve method. In the evaluation of NGANN & GANN network methods and SRC & MARS regression methods, correlation coefficients were respectively calculated as 0.94, 0.93, 0.767, 0.766, and root mean square errors (RMSE), 0.45, 0.49, 2.3 and 2.3. Nash coefficient (NS) was calculated respectively as 0.71, 0.58, 0.27 and 0.23. Therefore, the most efficient method among the four models is artificial neural network combined with morphological data (GANN). Furthermore, the findings of the study show that adding geomorphological parameters to sediment rating has little effect on the model performance.

---

Selection of optimum interpolation technique to estimate water quality parameters in unmeasured points plays an important role in managing the quality and quantity of water resources. The aim of this study is to evaluate the accuracy of interpolation methods using GIS and artificial neural network (ANNs) model. To this end, a series of qualitative parameters of samples from water taken from Dehgolan aquifer located in Kurdistan, Iran including CL, EC and PH were evaluated by any of the models. In this study, qualitative data from 56 observation wells with good dispersion in the whole plain was used. The data of 46 observation wells were used for calibration and the data of other 10 wells were used for verification of models. The results showed ANNs, IDW, and Kriging excellence and accuracy over other models in estimation of quality parameters CL, PH and EC. However the ANNs model is more accurate than other models. In case of lack of time and the need for acceptable accuracy and less risk in the estimation of qualitative parameters, the use of ANNs model is superior to other statistical models used.

---

Due to water shortage in country, more accurate estimate of water reserve can be one of the most important guidelines on the optimal management of water resource and cycle for development of water productivity efficiency. Therefore, using artificial neural network techniques the water supply of 174 fallen trees from different species was simulated. From any part of each bole, components of constant volume were extracted and placed in 105ºC to be oven-dried to measure specific drought index and wood density. Three input layers of diameter at breast height, height and specific wood density were used to simulate the response variable. The method of trial and test were used for neural network topology architecture. The results showed that the use of only diameter as input layer based on the validation indices explained 65% of variance of test of data. Using the three layers in the neural network, optimized output including function of Tan-sigmoid in the designed architecture with the number of 15 neurons demonstrated the highest accuracy (R²=0/92, MSE= 0/001, RMSE=81/08). In order to save the costs and manpower and to avoid a destructive method, the optimized output in the form of black box has the wide applicability to predict the water reserve in the mixed-beech forests to manage water cycle in the studied ecosystem.

---

Soil erosion is undoubtedly one of the most important problems in natural areas of Iran and has destructive effects on different ecosystems. Considering that calculation of the sediment rate in sediment stations and direct measurements of erosion process is costly and difficult, it is critical to find ways to accurately estimate the amount of sediment yield in catchments especially in arid and hyper arid areas because of their high ecological sensitivity. One of the most commonly used methods in these areas is the sediment rating regression method. Therefore, in this study sediment observed data for 48 events (the corresponding discharge and sediment) in a 23-year period from Fkhrabad basin (Mehriz) were compared to the estimated data obtained from Multi-line rating method, extent middle class, middle class rating curve with correction factor QMLE, SMEARING correction coefficient FAO and Artificial Neural networks (ANNs). Finally, the accuracy of these methods were assessed using different evaluation criteria such as Root Mean Square Error (RMSE), coefficient of determination (R2) and the standard Nash (ME). Results showed that ANN outperformed the other methods with the RMSE, R2 and ME of 203.3, 0.86 and 0.66, respectively. The results suggest that these methods should be used cautiously in estimating the suspended sediment load in arid and hyper arid regions due to the nature of the observed data and temporal and seasonal flow systems in these regions. It was also indicated that the artificial neural network models have higher flexibility than other methods which makes them to be useful tools for modeling in poor data conditions.
 

---

A limited number of agricultural weather stations measure moisture in the soil surface. Furthermore, soil moisture information may be required in areas where there is no weather station. The aim of the present study was to use Landsat 8 satellite images to estimate soil surface moisture in an area without agricultural meteorological stations. Gravimetric soil moisture for a total of 14 samples was calculated in the cold season in depths of 0-10 cm when Landsat 8 satellite was overpassing poor rangeland of North of Sabzevar. Furthermore, the first four principal components were extracted from seven Landsat-derived vegetation indices and bio-physical factors affecting soil moisture. Afterwards, the first four components were used to estimate soil surface moisture at the moment of the satellite passing the region using a multivariate linear regression and neural networks. The obtained results of instantaneous soil surface moisture showed that the neural networks had mean absolute percentage error of while classical regression analysis had mean absolute percentage error of 40%. The results also showed the benefits of using both in-situ soil moisture data and Landsat 8 satellite images to model instantaneous soil surface moisture content for areas lacking meteorological networks.
 

---