Abstract:
Climate of Bangladesh is expected to be changed under the influence of Global Warming
conditions. However, it is essential to quantify it irrespective of time and space. Empirical Statistical
Downscaling (ESD) is an approach through which the magnitudes of the projected variable at the
local level with time can be estimated based on the projected output of Global Climate Models
(GCMs). In this study, two main climate variables of mean temperature and rainfall are considered
for projection as well as to calculate their projected magnitudes at different future lengths. In this
way, 10 most suitable GCMs are considered for temperature projection. But for rainfall projection
5 most appropriate GCMs are selected for the pre-monsoon season (March-May); 3 most
appropriate GCMs for monsoon (June-September) and post-monsoon seasons (October-November)
separately. The mean of the selected models for each group is also calculated to determine the
ensemble prediction. To understand the current climate and its behaviour and to justify the model
performance on historical records, daily mean temperature (average of the maximum and minimum
temperature records) and rainfall of 34 stations (which are widely distributed over Bangladesh) are
collected from the archive of Bangladesh Meteorological Department (BMD) during the period of
1981-2010. There is a missing value within daily records, which is essential to extract, and the
missing value of temperature and rainfall are calculated using the R-package of multiple regression,
and consequently, an updated climate record has been prepared for this study. To proceed for the
study three emission scenarios of RCP2.6, RCP4.5 and RCP8.5 under CMIP5 are selected and the
relevant GCM model outputs are commonly selected from all of these scenarios. Then the GCM
outputs are downscaled through ESD Package (an R-package specially developed for statistical
downscaling), adopting the Principal Component Analysis (PCA) method to produce the projected
magnitudes of mean temperature and rainfall at each of the selected locations on month basis for
the period of 2021-2100. The simulated results are evaluated with observation at station location
basis as well as at the national average (considering the average of all BMD stations location). The
evaluation of the projection result has been conducted using the technique of the five-fold crossvalidation
method.
To
evaluate
how
well
the
downscaled
GCM
ensembles
represent
the
past
trends
and
interannual
variability
for
each
station,
the
observed
seasonal
data
in
the
period
1981–2010
is
compared
with
the
statistical
characteristics
of
the
downscaled
ensembles.
The
trend
in
the
period
of
1981–2010 is calculated for the observed value and each downscaled ensemble member at all
stations. Then the result has been analyzed. Analysis reveals that the seasonal and annual mean temperatures are projected to increase in the near future and far future for each emission scenario.
The increment rates during pre-monsoon season are +0.62, +0.5 and +0.54°C in the near future and
0.78, 1.19 and 2.04°C in the far future, respectively for RCP2.6, RCP4.5 and RCP8.5. In the
monsoon season, the projected rates are +0.25, +0.13 and +0.37°C in the near future and +0.33,
+0.45 and +1.27°C in the far future, respectively. The projection of post-monsoon season is higher
than other seasons; in the near future, the magnitudes are +0.6, +0.72 and +0.94°C and are +0.76,
+1.42 and +2.81°C in the far future. Winter season is also projected to be warming at high rate, and
the projection rates are +0.45, +0.44 and +0.86°C for the near future and are +0.65, +1.06 and
+2.23°C in the far future. The annual mean temperature is likely to be higher with the projection
rate of +0.48, +0.45 and +0.68°C in the near future and it is of +0.71, +1.16 and +2.83°C in far
future, respectively for RCP2.6, RCP4.5 and RCP8.5. Analysis also depicts that the frequency of
the wet-day (with rainfall ≥ 1mm/day) is projected to increase in all seasons in the near and far
future for RCP2.6. It is projected to increase in pre-monsoon and post-monsoon seasons as well as
annually both in the near and far future, but it is likely to decrease in both of the time slabs of
monsoon season for RCP4.5. For the case of a very high emission scenario of RCP8.5, wet-day
frequency is projected to increase in pre-monsoon but decrease in monsoon season both in the near
and far future. As a whole, the annual wet day frequency indicates an increasing trend.
Description:
A dissertation submitted to the Institute of Statistical Research and Training (ISRT), University of Dhaka in fulfillment of the requirements for the degree of Doctor of Philosophy in Applied
Statistics.