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