Quantification of model uncertainty in sub-daily extreme precipitation projections

Sub-daily precipitation is projected to intensify more rapidly with global warming, that in turn may lead to frequent disasters such as flash floods, landslides, etc. However, the precipitation projections from Global Climate Model (GCMs) are known to be highly uncertain, with an increasing magnitude at finer spatio-temporal scales. Hence, realistic quantification of the inherent uncertainty is vital to improve the confidence in these projections, for the mitigation/adaptation applications. In the present study, the model uncertainty in sub-daily (3-hourly) precipitation extremes is estimated, by considering 20 Coupled Model Intercomparison Project Phase 5 (CMIP5) models, during four different seasons for the near (2026-45) and far-future (2081-99) periods. An uncertainty quantification approach using "mutual information based independence weight" is proposed in this study, to account the linear and nonlinear dependence between GCMs along with the equitability property. The CMCC-CM model is found to be the most independent model having highest independence weight (0.064) in the ensemble. The uncertainty in northern hemisphere is found to be more during June to August period, whereas in southern hemispheres during December to February period. Overall, there is an increase in model uncertainty in far-future, especially over the equatorial regions. However, the behaviour of different indices is not in agreement with each other. For example, extreme characteristics such as monthly maximum 3hourly-precipitation has increasing uncertainty; whereas the mean characteristic like, simple hourly-precipitation intensity index pos-sesses unchanging uncertainty. Model uncertainty is most sensitive to the precipitation characteristics of the regions, followed by elevation, a significant geophysical factor. The study also highlighted the need for model improvement over the grids having elevation <2 km and above 4 km. The study identifies the possible regions, seasons, factors etc. where model improvement can be made. Moreover, it is anticipated that improvement of resolution, updation of convective parameterization schemes and land surface parameterization schemes in GCMs may enhance the skill of GCMs and can improve the confidence in the projection of precipitation on sub -daily scale.