Comparison of statistical downscaling techniques for development of future climate change scenarios for the coastal areas of Bangladesh

Abstract

Coastal region of Bangladesh is vulnerable to climate change due to its flat topography and dense population and poverty. Generation of plausible climate change information for this region is important for various impact studies by considering uncertainty of projections. However, Global Climate Model (GCM) provides climate change predictions in coarse resolution (>100km) which often not adequate for the need of high resolution information for the impact studies. Downscaling provides way to generate high resolution climate change predictions from the coarse resolution GCM output. The dynamic downscaling is more computationally demanding but physically based whereas statistical downscaling are less computationally intensive and robust. To capture uncertainty of various climate predictions, statistical methods are desirable to derive fine scale climate information from the GCM output. On the other hand, a number of statistical downscaling techniques are available ranging from Simple Linear Regression to more sophisticated techniques like, artificial neural networks or weather generators. It is also essential to compare performances of different statistical techniques available for generating climate change information. In this study, the comparison between two Statistical downscaling models SDSM and LARS-WG has been examined and their performance to project the future precipitation data has been observed. Three coastal districts, namely, Shatkhira, Khulna and Patuakhali have been selected as a case study to evaluate performance of these two downscaling methods. Daily Rainfall data from Bangladesh Meteorological Department (BMD) has been collected for the last twenty years from 1986 to 2005. Global climate modeling data from HadCM3 developed by Met office, UK for the moderate emission scenarios SRES A1B (which is balanced emission scenario). The performance has been evaluated through statistical indicators. The monthly variation of rainfall during the calibration period (1986-1995) and validation period (1996-2005) was observed. In monthly variability comparison, LARS-WG performed better than SDSM. LARS-WG performs better for almost all the four seasons during the calibration period. However, for Khulna and Patuakhali district, SDSM performed well in pre-monsoon and monsoon season. The statistical indicators also exhibit less error for LARS-WG. For percentage change of precipitation from baseline to 2050s the maximum change is observed in Satkhira district in both increasing and decreasing percentage change. To capture the variability of precipitation LARS-WG is more capable than SDSM. However, further study can be carried out using other districts of coastal zones of Bangladesh and other statistical downscaling techniques.