Assessment of climate change impact on temperature extremes in a tropical region with the climate projections from CMIP6 model

Abstract

The global mean surface temperature is rising and under climate change it is expected to increase in the future. The changes are not uniform across the world and the impact assessment at a regional level is, thus, a necessity. Bangladesh, a tropical low-lying monsoon region, is greatly at risk under climate change. However, no attempt has been carried out to assess the changes in frequency of extreme temperatures (return period values) under the backdrop of climate change which is required in practical applications. The aim of this study is to investigate the changes in temperature extremes: maximum (Tx) and minimum (Tn) at the daily time scale under climate change in Bangladesh. The multi-model ensemble methodology comprised of five general circulation models and two emission scenarios from CMIP6 framework was used to evaluate the impact in two future time horizons: 2021–2060 and 2061–2100. The L-moment based frequency analysis with annual maxima (minima) data was used to quantify extreme temperatures in terms of return levels. With this approach, the identification of a probability distribution is one important aspect which is investigated in this study. We have found that the generalized normal (GNO) distribution is quite appropriate of describing the temperature extremes. There is a significant increase in location parameter which signifies a uniform shift for the extreme tail of the distribution. There are no appreciable overall changes in scale and shape parameter. The temperature extremes (both Tx and Tn) are expected to increase noticeably compared to the present observed condition. There is a tendency to have a greater estimate of extremes in the far future than the near future. The greater estimate is also perceived under the high emission scenario in comparison to the medium emission scenario. The country average change in 20-year return period value of Tx by the end of this century is about 3–4.7 °C compared to the corresponding changes of 2.4 to 3.7 °C for Tn. The findings are expected to assist in the evaluation of climate change impacts and adaptation strategies in Bangladesh.

Appendix

The GNO distribution with three parameter: location $(\xi )$, scale $(\alpha )$ and shape parameter $(\kappa )$ has the following expression (Hosking and Wallis 1997):

$$F\left(x\right)=\Phi \left(y\right)$$

(3)

where, $y=-{\kappa }^{-1}log\left\{1-\kappa \left(x-\xi \right)/\alpha \right\}$ and $\Phi \left(y\right)$ is the standard normal distribution function. The shape, $\kappa ,$ determines the tail of the distribution: when $\kappa <0,$ the distribution has positive skewness and a lower limit of $\xi +\alpha /\kappa$ ; when $\kappa >0,$ the distribution has negative skewness and an upper limit of $\xi +\alpha /\kappa$, and when $\kappa =0,$ the distribution becomes the normal distribution.

The parameter is estimated by L-moments. The estimation by Hosking and Wallis (1997) is given below:

$$\kappa \approx -{t}_{3}\frac{{E}_{0}+{E}_{1}{t}_{3}^{2}+{E}_{2}{t}_{3}^{4}+{E}_{3}{t}_{3}^{6}}{1+{F}_{1}{t}_{3}^{2}{+F}_{2}{t}_{3}^{4}+{F}_{3}{t}_{3}^{6}}$$

(4)

$$\alpha =\frac{{L}_{2}\times \kappa \times {e}^{-{k}^{2}/2}}{1-2\Phi \left(-\kappa /\sqrt{2}\right)}$$

(5)

$$\xi ={L}_{1}-\frac{\alpha }{\kappa }\left(1-{e}^{{\kappa }^{2}/2}\right)$$

(6)

where ${L}_{1}$ is the 1st L-moment, ${L}_{2}$ is 2nd L-moment and ${t}_{3}$ is the L-skewness; $\Phi \left(\right)$ is the standardized Normal variate; the coefficient values (Eq. 4) are as follows: ${E}_{0}$ = 2.0466534, ${E}_{1}$ = − 3.6544371, ${E}_{2}$ = 1.8396733, ${E}_{3}$ = − 0.20360244; ${F}_{1}$ = − 2.0182173, ${F}_{2}$ = 1.2420401, ${F}_{3}$ = − 0.21741801.

The quantile value (${\mathrm{\rm Z}}_{F}$) can be estimated as

$${\mathrm{\rm Z}}_{F}=\xi +\frac{\alpha }{\kappa }\left(1-{e}^{-k{\Phi }^{-1}\left(\mathrm{F}\right)}\right), \kappa \ne 0$$

(7)

$${\mathrm{\rm Z}}_{F}=\xi +\alpha \times {\Phi }^{-1}\left(\mathrm{F}\right),\upkappa =0$$

(8)

where, ${\Phi }^{-1}\left(\right)$ is the standardized inverse Normal variate.

For the estimation of $T$-year return value of extreme maxima, F being replaced by $(1-1/{\text{T}})$ while for the estimation of extreme minima, F being replaced by $(1/{\text{T}})$.