Accuracy of PERSIANN-CDR Precipitation Satellite Database in Simulation Assessment of Runoff in SWAT Model on Maharlu Basin

Introduction
Rainfall is the most important meteorological factor driving the hydrology of river basins. For the development and management of water resources, it is required to have a reliable coverage of rain gauges, rainfall satellite data and weather radars. Well-maintained ground-based rainfall stations give the best rainfall estimation with high accuracy over time for a small area. The spatial sampling error becomes higher in estimating rainfall when using interpolation techniques. This issue becomes particularly critical in data scarce regions with unevenly distributed rain gauge stations. This is recognized as one of the principal sources of uncertainty in hydrological modeling. Currently, with more and more global precipitation datasets developed, a decline is seen in the application of reanalysis products in hydrological modelling. As a hot research field, many studies focus on the application of directly measured precipitation data on flood risk evaluation at basin scales and discuss their potential for hydrological prediction of ungauged/poorly gauged basins. In this study, we used PERSIANN-CDR gridded database precipitation for modeling runoff in SWAT model in Maharlu Lake basin.
Materials and methods
The PERSIANN-CDR was used as a gridded database of precipitation for modeling runoff in SWAT model in Maharlu Lake basin. The PERSIANN-CDR is initially compared with rain gauged data and after that it was used as input to SWAT model. SWAT model was calibrated by rain gauge data during 1983 to 2013. The Warmup period set to 3 years. Three discharge stations were used for calibration. Correlation coefficient of, Nash-Sutcliff, POD, CSI, FAR, RMSE, ME and BIAS had been assessed to determine the accuracy of PERSIANN-CDR. SWAT model uncertainty and sensitivity were calculated in SWAT-CUP by SUFI2 method.  
Results and discussion
Comparing the PERSIANN-CDR in monthly scales, we found that this satellite wheatear database less estimates the variables in all months. The results showed average of correlation coefficient is 0.6 and RMSE showed a high error in rainy seasons. In SWAT model, calibration period was set to 1983 to 2010 with validation from 2011 to 2013. Calibration with gauged data showed satisfactory Nash-Sutcliff and R2 statistical indices about 0.6 for the area. The best result was occurred in Chenar-Sokhte-khosh discharge station, R2 was about 72% in calibration and 81% in validation. Calibration with PERSIANN-CDR database showed that this database is not good enough to be used in this semi-distributed model. In Chenar-Sokhte-khosh discharge station, R2 is calculated about 0.59 and Nash-Sutcliff about 0.21. R-factor and P-factor was presented about 0.5 in all discharge stations. These factors show that uncertainty calculation was occurred in good form. The simulation of annual runoff showed that the average runoff simulated using observation database was 1.68 m3/s, the mean runoff simulated by PERSIANN-CDR is 0.84 m3/s, and mean runoff of discharge stations were 1.77 cubic meters per second. On monthly scale, PERSIANN-CDR estimated less runoff like rainfall over all months. Both databases simulate runoff values relative to those recorded in the autumn months less than actual values. The results of this study, which were conducted using the PERSIANN-CDR satellite product, unlike the other studies with global exploratory bases, displayed that in the simulation with the SWAT model, this base cannot be accurately high in simulation. The error of estimating precipitation has been entered directly into the model and caused an error.
Conclusion
In this study, with the accuracy of precipitation data, PERSIANN-CDR satellite data on rainfall estimation revealed that this database estimated precipitation values less than real values in all months of the year. Runoff simulation using this satellite product expresses the explanatory factor and the efficiency of Nash-Sutcliff about 0.59 and 0.21.
Despites the time series of precipitation values, this satellite database has a high correlation with the actual values observed on rain-fed stations, but as the findings show the estimated rainfall values are always lower than actual recorded values. Based on the findings from this study, the PERSIANN-CDR satellite is not very accurate on the area of the Maharlou Lake Basin, located in eastern Zagros. In the semi-distributed SWAT model, it cannot simulate runoff. Therefore, it is suggested that before applying estimated rainfall data, this satellite database will have its error and bias values compared with the observed data on rain gauge stations and, then, the estimated precipitation values are corrected based on the bias.