Propagation of model and forcing uncertainty into hydrological drought characteristics in a multi-model century-long experiment in continental river basins

Abstract:

Droughts are creeping hydro-meteorological events that bring societies
and natural systems to their limits and inducing considerable
socio-economic losses. Currently it is hypothesized that climate change
will exacerbate current trends leading a more severe and extended
droughts, as well as, larger than normal recovery periods. Current
assessments, however, lack of a consistent framework to deal with
compatible initial conditions for the impact models and a set of
standardized historical and future forcings.

The ISI-MIP project provides an unique opportunity to understand the
propagation of model and forcing uncertainty into century-long time
series of drought characteristics using an ensemble of model predictions
across a broad range of climate scenarios and regions. In the present
study, we analyze this issue using the hydrologic simulations carried
out with HYPE, mHM, SWIM, VIC, and WaterGAP3 in seven large continental
river basins: Amazon, Blue Nile, Ganges, Niger, Mississippi, Rhine,
Yellow. All models are calibrated against observed streamflow during
the period 1971-2001 using the same forcings based on the WATCH data
sets. These constrained models were then forced with bias corrected
outputs of five CMIP-5 GCMs under four RCP scenarios (i.e. 2.6, 4.5,
6.0, and 8.5 W/m2) for the period 1971-2099.

A non-parametric kernel density approach is used to estimate the
temporal evolution of a monthly runoff index based on simulated
streamflow. Hydrologic simulations corresponding to each GCM during the
historic period of 1981-2010 serve as reference for the estimation of
the basin specific monthly probability distribution functions. GCM
specific reference pdfs are then used to recast the future hydrologic
model outputs from different RCP scenarios. Based on these results,
drought severity and duration are investigated during periods: 1)
2006-2035, 2) 2036-2065 and 3) 2070-2099. Two main hypothesis are
investigated: 1) model predictive uncertainty of drought indices among
different hydrologic models is negligible compared to the uncertainty
originated from different GCMs and 2) the projected drift of drought
characteristics is hydrologic model independent and it is only driven by
the GCM variability. The temporal evolution between drought severity and
duration is also analyzed.