Downscaling of selected global climate model projections over Bangladesh using weather generators

Abstract

Bangladesh is prone to extreme weather-related disasters, especially related to floods, cyclones and droughts. Climate change has the potential to impact the global hydrological cycle as well as regional hydrology across the world and increase the magnitude and frequency of these disasters. Generally, the outputs of global climate models (GCMs) are used for regional climate change impact assessment. GCMs simulate time series of global climate variables considering emissions of different green house gases (GHGs) but they often fail to capture the non-smooth fields such as precipitation precisely. Spatial downscaling is therefore used for better understanding and assessment of future climate conditions at local/ watershed scales under different climate change scenarios.

To simulate future climate at local scale, weather generators (WGs) are appealing statistical downscaling technique. This study has statistically downscaled the climate data from two GCMs (CanESM2 and MIROC5) at 30 climate stations of Bangladesh using two recently developed improved weather generators (K-nearest neighbor weather generator (KNNCAD version 4) and Maximum Entropy based weather generators (MEBWG)). Eight future climate datasets, from two downscaling methods and each of them driven by two GCM datasets (MIROC5 and CANESM2) under two RCPs (RCP 4.5 and 8.5), were produced for three time slices 2020, 2050 and 2080. This study has also investigated the primary sources of uncertainty attributed to the selection of GCM, emission scenario, and downscaling model on future projections of selected climate variables.

This study reveals that the average maximum and minimum temperature will increase in future gradually. The multi model ensemble average of maximum temperature anomaly has been found to be 0.86, 1.40 and 1.87oC under RCP 4.5 and 0.94, 2.02 and 3.34oC under RCP 8.5 for 2020, 2050 and 2080, respectively. The multi model ensemble average of minimum temperature anomaly is 1.29, 1.98 and 3.27oC for 2020, 2050 and 2080, respectively for RCP 8.5. Spatial analysis of projected maximum temperature shows that it may increase up to 20% in some regions.

In case of precipitation, the results obtained from the two downscaling models were found reasonably comparable in terms of direction of change, but they disagree in terms of magnitude. The multi-model ensemble average of precipitation anomaly (averaged over the whole country) has been found to be 2.48, 3.30 and 5.19% under RCP 4.5 and 2.94, 5.86 and 7.76% under RCP 8.5 for 2020, 2050 and 2080, respectively. Analysis of rainfall indices such as 1-day maximum precipitation, contribution from heavy precipitation events exceeding 99% precipitation and total precipitation show increasing trends for all three future time slices.

The projected change in precipitation may be non-uniform over the country. This study found that the precipitation may change on annual scale from -5 to 10% under RCP 4.5 and -5 to 15% under RCP 8.5 in different regions. The spatial patterns of projections may also vary under different GCMs and downscaling model, suggesting uncertainty in the projection of precipitation changes. The increase of precipitation is higher for MIROC5 GCMs using KNNCAD downscaling model. The overall uncertainties in the projections of temperature and precipitation can be due to model formulation, scenario uncertainty and the difference in parameterization schemes employed for the representation of convection in GCMs. However, the overall precipitation and temperature are likely to be higher in the future over the region and such changes will likely exacerbate the environmental impacts of extreme climate in the future. Because of the variability in the future projections obtained with the four projected datasets under each RCP scenario considered herein, further work in the same region should consider a larger ensemble of global climate models to evaluate if a larger ensemble provides more conclusive results.