Severe lack of rainfall and its unfavorable distribution, very high evapotranspiration rate and the low water-holding capacity in some of the soils are the most important challenges for the establishment of plants in arid areas. This research was aimed at comparing efficiency of different water harvesting methods for establishment of six native plant species in Chahdar rangeland of Mashhad. It was designed in a completely randomized design block, splitted in time and space. Experiments contained two major treatments that were assessed during two consecutive growth seasons. First major treatment, included different water harvesting methods i.e. contour furrow, pits and small basins. For the second treatment, six perennial range plants of Agropyron elongatum, Secal montanum, Sanguisorba minor, Medicago sativa, Onobrychis khorasanicum and Krashninkova ceratoides were sown and compared. Seeding emergence and survival were recorded three times (spring 2014, autumn 2014 and spring 2015) and data were analyzed by using SPSS software. Results indicated significant differences between water harvesting treatments and between plant species. Highest survival rate, at the end of the second growth season, was respectively found for Onobrychis khorasanicum and Krashninkova ceratoides. Highest establishment rate was found in pits, mainly because of higher water soil accumulation as compared to that of contour furrow and small basins. According to the results of this research, best restoration practice for Chahdar rangeland is sowing seeds of Onobrychis khorasanicum and Krashninkova ceratoides within the small pits.

Choosing the appropriate inputs for intelligent models is important. Because it can reduce costs and save time and increase accuracy and efficiency of the models. This work aims at  the use of gamma test to select the optimum combination of input variables including delayed records of in time series modeling precipitation. Monthly time series of rainfall for the period 1383 to 1393 was used for Rasht station Rainfall data as with different lags were employed as input to gamma test. Results showed that time series with three delays (lags), provides better results. The simulation was performed using Bayesian network and multivariate linear regression. The performance of models was assessed using three criteria, i.e. coefficient of determination (R²), root mean square error (RMSE), and dispersion index (SI). Bayesian neural network using a three-month delay the coefficient of determination of 0.82, root mean square error of 17.84 and a diffusion index of 0.17 showed better performance as compared with multivariate regression. The results established the significant role of the gamma test integrated with intelligent models in the appropriate selection of input variables..