Results and Discussion:

The CMIP5 multi-model ensemble LOCA results for precipitation and temperature during baseline period show consistency with the observed values (Figure 3). Between the 3 models, GFDL-CM3 has the closest distribution to the observed precipitation with the 6.8% median difference and the closest number of outliers. The 25th percentile for the observation and GFDL-CM3 are 71.4℃ and 74.68℃ respectively. 75th percentiles are the same (148.3℃). Therefore, it indicates similar distribution. Figure 3b illustrates the temperature distribution for the observed and baseline period. It represents quiet similarity, especially for model GFDL-CM3, where the distance is between 25th and 75th percentile and the whiskers’ length are the same. The median difference of models (CCSM4, GFDL-CM3, GISS-E2-R) from the observed ones are 4.2%, 2.6%, and 4.7% respectively. We also compared the baseline ET from the climate data with the observation period (Figure 3c). Figure 3c also demonstrates similar distribution from 25th to 75th percentile from all dataset. However, upper whiskers for the observation is longer. This difference has no implication on the study, since here, we are not focused on extreme weather situation.
Average maximum and minimum temperature has important repercussions on hydrological implications. Figure 4 and 5 illustrate temperature behavior on daily and seasonal base, respectively. Figure 4 represents monthly average of basin-wide daily maximum temperature (Figure4b), monthly average of basin-wide daily minimum temperature (Figure 4c) and monthly average of basin-wide daily temperature (Figure 4a). For average temperature (Figure 4a) models match the observed average temperature, from mid-March to June (Spring) and from mid-September to mid-November (Fall). For Winter (DJF) and Summer (JJA), however, there are differences up to 1℃. This trend is the same of maximum temperature (Figure 4b) and minimum temperature (Figure 4c). However, the discrepancies for maximum temperature during Summer (the peak of the graph) and for minimum temperature during Winter (the legs of the graph) are more noticeable. This behavior indicates the more extreme the temperature, the more the difference between the models and the observed data. This also can be seen in daily temperature representation of the subbasin (Figure 5). On the figure, intensified wiggling behavior of the graphs at the troughs and sometimes at the peaks supports the idea. This specially is more obvious in months of December and January.
Figure 6 shows the monthly average of precipitation, ET, water yield, and surface runoff for 10 year for both the beasline and observed period. From mid-March to mid-June, and September and November, the average observed rainfall is the same as the models prediction with negligible differences. During Summer and Winter, however, there exist some discreppencies. These discrepencies could be attributed to the model baises. As SWAT model uses these model results as climate data, the baises can be projected to the simulated hydrological results such as ET, water yield, and surface runoff. Model predictions for ET (Figure 6b) match with the observed data except during summer with small differnces up to 8% in July. Highest level of water yield occurs during the month March (Figure 6c) where the differnces with the model predictions is around 13%. During Spring, Summer, and Fall models perdict the amount of average water yield close to the observed data. Since surface runoff amount and water yield are linked, the yearly pattern of surface runoff follows the water yield pattern. Surface runoff is understimated. For example in March when the highest amount of surface runoff happens through the year, the models predictions is 17% low for CCSM4 and GISS-E2-R and 11% low for GDFL-CM3. This difference is due to land use changes through the time period. Therefore, it indicates the biases of the land use map. In this study, we have not looked for extreme events that partially acount for these biases. Thus, considering the different source of inevitable baises it can be concluded that the results based on the models are reliable