---

In recent years, groundwater fluctuations and rising of Parsabad plain in recent years has resulted in the formation of some problems such as water logging and salinity of agricultural lands as well as in the destruction of rural and urban's building foundations and in the subsidence of roads and building. In this research, considering geomorphology features and the style of their effects on groundwater fluctuations by means of Geographical Information System (GIS), the trend of groundwater was modeled so that the effect of groundwater fluctuation can be minimized by change in irrigation system. The monthly data of 30 observational well, geological, topographical, climatic data, soil and territorial recourses of study area was used for surveying the level of groundwater fluctuations. Using the 2002 to 2008 period data, trend of groundwater fluctuation and hydrograph level were drawn. Then the relationship between different geographic factors and groundwater fluctuation were evaluated. Pearson correlation and linear multivariate regression model between effective parameters in groundwater fluctuation and were prepared, in which operation criteria in test course shows adequate performance of mentioned model. The obtained results imply the increase in groundwater level of plain and that its flow is from south to north especially northeast. It was found that, among factors, elevation, slope, earth material (porosity and transition capacity) are highly effective, permeability, precipitation and evaporation are less effective in groundwater fluctuation whereas the depth of groundwater doesn’t have any effect on groundwater fluctuation. Operating continuous draining in most part of plain and modifying irrigation system from traditional to mechanized system can prevent the salinity of cultivated soils and the subsidence of some homes.

---

Introduction
Annual precipitation forecasting usually guarantees success in sustainable management of water resources and watersheds in cases like determination of rain-fed and water cultivation area and water resources consumption planning. Although precipitation does not follow a specified pattern; however, it has a correlation with some climatic parameters. If these parameters were easy to find, simple applicable models could be developed to predict annual precipitation. One of the simplest methods for predicting annual precipitation is regression models. The block box models such as Artificial Neural Network (ANN) has also been used for precipitation forecasting.  This study was an attempt to predict annual precipitation using some climatic parameters by multiple linear regression and ANN models.  
Materials and Methods
Chaharmahal-Bakhtiari and Yazd provinces are located in the southwestern and central part of Iran with semi-arid and arid climate, respectively. Mean annual precipitation of these two provinces are 560 and 110 mm, respectively. Meteorological data of the weather stations belong to 2001-2011 in Chaharmahal-Bakhtiari and 2003-2013 in Yazd province; respectively, were analyzed. Various parameters were calculated for predicting annual precipitation using long-term daily precipitation and temperature data of the meteorological stations. Among the parameters, total precipitation in the first half of the water year (R_6m, mm), time to 47.5 mm cumulative precipitation since the beginning of autumn (t_47.5, day), long-term mean annual precipitation (R_m, mm), average summer temperature of the preceding water year (T_su, °C) and average temperature of preceding summer and current autumn of water year (T_su.au, °C) that had a high correlation with annual precipitation, were used in multiple linear regression (MLR) models and artificial neural network (ANN) techniques. normalize mean square error (NRMSE) and degree of agreement (d) were used to evaluate accuracy of the models for predicting annual precipitation.
Results and Discussion
Results showed that the obtained MLR models were significant at a probability level of less than 0.01. Results also showed that both methods (MLR and ANNN) could accurately estimate the annual precipitation. Evaluation and verification of the models with NRMSE values less than 0.3 and d values greater than 0.8 confirmed the performance of the models. The best topology of the ANN network in the study was a multiple layer perceptron network with one hidden layer and two neurons and sigmoidal activation function. The findings of the study support the fact that higher temperature in summer and autumn was a sign of higher and lower annual precipitation in Chaharmahal-Bakhtiari and Yazd Provinces, respectively. Besides, higher time period to 47.5 mm cumulative precipitation from the beginning of autumn implies fewer amount of annual precipitation.  
Conclusion
This study showed that annual precipitation could be predicted by MLR and ANN methods in both arid and semi-arid regions with acceptable accuracy. According to the results, a rainy water year will be expected continuing a warmer summer and autumn in Chaharmahal-Bakhtiari Province; however, there will be a dry water year in Yazd Province followed by these conditions. Furthermore, if 47.5 mm cumulative precipitation takes a longer time since the beginning of autumn annual precipitation will decrease.