Influence of snow and soil moisture initialization on sub-seasonal predictability and forecast skill in boreal spring

• Published in: Climate dynamics 2016-07, Vol.47 (1-2), p.49-65
• Main Authors: Thomas, Jaison Ambadan, Berg, Aaron A., Merryfield, William J.
• Format: Article
• Language: English
• Subjects: Air temperature; Analysis; Climate models; Climatology; Earth and Environmental Science; Earth Sciences; Geophysics/Geodesy; Influence; Moisture content; Oceanography; Snow; Snow-water equivalent; Soil moisture; Soils; Spring; Surface temperature; Weather forecasting
• ISSN: 0930-7575; 1432-0894
• DOI: 10.1007/s00382-015-2821-9

This study examines the influence of snow and soil moisture initialization on sub-seasonal potential and actual prediction skill of Canadian Climate Model version 3 (CanCM3) predictions of springtime (April–May) near surface air temperature. Four series of ten-member ensemble forecasts, initialized on 1st April where each series use different land surface initialization, were performed for the 20 year period 1986–2005. Potential predictability of temperature for extratropical Northern Hemisphere land is assessed using synthetic truth and signal-to-noise methods, and compared with actual prediction skills determined through validation against an ensemble mean of six reanalysis products. These metrics are computed for the forecasted 15 days averaged values of temperature at 15, 30 and 45 days lead times. Three of the four land surface initializations considered are intended to be realistic. These are obtained from the Canadian LAand Surface Scheme (CLASS) land surface component of the climate model driven off line with bias-corrected meteorological fields, with and without rescaling to the climate model’s land climatology, and from climate model runs where the atmospheric component is constrained by reanalysis fields. A fourth land surface initialization that is intended to be unrealistic consists of a “scrambled” version of that obtained from rescaled offline-driven CLASS, in which each ensemble member is assigned values from a year other than the one being forecasted. Comparisons of forecasts using the scrambled and corresponding realistic land initializations indicate that the latter show higher potential predictability overall especially over North America and parts of Eurasia at all lead times. The higher potential predictability is primarily attributed to correct initialization of land surface variables, in particular the snow water equivalent, and the frozen and liquid components of soil moisture. Our results also indicate that predictability is governed mainly by forecast signals, with high forecast noise also playing a role. Actual skills, though lower than potential predictability, likewise show a positive influence of realistic land initialization.