A31I-04
The impact of quantified land surface uncertainties on seasonal forecast skill
Abstract:
The land surface is a key component in seasonal forecasting, and well-represented soil moisture is particularly important for the simulation of heatwaves.Methods to represent uncertainties in the atmosphere of climate models have been shown to improve forecasts. However these methods have not yet been applied to the land surface component of climate models.
We consider three methods of incorporating uncertainties into CHTESSEL, the land surface model of the ECMWF forecasting system. These methods are: stochastic perturbation of soil moisture tendencies, static and then stochastic perturbation of key soil parameters. We present analysis of the results of fully coupled seasonal hindcasts with each method applied.
We find significant improvement for extreme events, particularly in terms of forecast reliability of upper and lower quintile soil moisture. These improvements also propagate into the atmosphere, impacting the reliability of seasonal-average predictions of latent and sensible heat flux anomalies and air temperature. This improvement is consistent over the hindcast, and also for particular cases such as the 2003 European summer (MacLeod et al 2015).
We also present work with an uncoupled version of CHTESSEL. Extending the method of Wood & Lettenmaier (2008), we quantify the global evolution over forecast lead-time of the relative magnitudes of initial condition, forcing and parameter uncertainty in the land surface. Among other things this gives some indication of where predictability from initial conditions is more persistent, and where uncertainty in land surface parameters has the largest impact on simulated soil moisture.
MacLeod DA, CLoke, HL, Pappenberger F and Weisheimer AF (2015), Improved seasonal prediction of the hot summer of 2003 through better representation of uncertainty in the land surface, QJRMS
Wood, AW, and Lettenmaier DP (2008), An ensemble approach for attribution of hydrologic prediction uncertainty, GRL