Can climate models represent the precipitation associated with extratropical cyclones?

Abstract

Extratropical cyclones produce the majority of precipitation in many regions of the extratropics. This study evaluates the ability of a climate model, HiGEM, to reproduce the precipitation associated with extratropical cyclones. The model is evaluated using the ERA-Interim reanalysis and GPCP dataset. The analysis employs a cyclone centred compositing technique, evaluates composites across a range of geographical areas and cyclone intensities and also investigates the ability of the model to reproduce the climatological distribution of cyclone associated precipitation across the Northern Hemisphere. Using this phenomena centred approach provides an ability to identify the processes which are responsible for climatological biases in the model. Composite precipitation intensities are found to be comparable when all cyclones across the Northern Hemisphere are included. When the cyclones are filtered by region or intensity, differences are found, in particular, HiGEM produces too much precipitation in its most intense cyclones relative to ERA-Interim and GPCP. Biases in the climatological distribution of cyclone associated precipitation are also found, with biases around the storm track regions associated with both the number of cyclones in HiGEM and also their average precipitation intensity. These results have implications for the reliability of future projections of extratropical precipitation from the model.

Appendix

1.1 Decadal variability in HiGEM

The data used in this study is taken from a single decade of an 80 year integration of the model forced with late twentieth-century radiative forcings. Figure 8 shows the percentage of the total precipitation contributed by ETCs in each of these decades. It can be seen that though there are small variations in the locations of the highest precipitation accumulations, associated with variability in the storm track locations, the biases in the model climatology relative to the observations remain consistent across the decades.

Fig. 8

Decadal variability in the percentage of the total precipitation contributed by ETCs (%) in HiGEM. Each panel represents a single decade from the model integration. The masked and stippled areas in the percentage plots are where the total precipitation for that decade is less than 1 mm day⁻¹

1.2 Spin-up in ERAI

The ERAI reanalysis (Dee et al. 2011; Simmons et al. 2007) uses a 4D-Var data assimilation system to incorporate observations over a 12-h analysis period, with forecasts commencing at 00:00 and 12:00 UTC. Precipitation is not an analysed field and must therefore be taken from short-range forecast accumulations. During the first several hours of the forecast simulation, the precipitation field (and many other fluxes and tendencies) is affected by “spin-up” as the model adjusts from the initialised fields, such that the estimates at the start of the forecast period are less robust than at later lead times (see Kållberg 2011). Moving too far from the start of the forecast also leads to degradation of the precipitation estimates. Given these issues, a suitable window must be chosen for the accumulations to be used in this study. The primary constraints on the selection of an appropriate period is the desire to use the GPCP dataset in this work. The GPCP dataset provides daily total precipitation estimates from 00:00 to 24:00 UTC. Further, the ERAI forecasts are initialised every 12 h so any accumulations must be selected from continuous 12 or 24 h periods in the forecast.

Given the requirement to have daily accumulations centred on 12:00 UTC, it would be possible to use either (1) a continuous 24-h accumulation from a single forecast for each day or (2) two 12 h accumulations from successive forecasts. To assess this, evaluating any spin-up/adjustment (as the model is initialised) or model drift (as the forecast increases in length and the model is no longer closely constrained by observations) is required. The adjustment/drift effect can be more readily demonstrated using shorter accumulation periods. In this analysis, given the 12 h or greater accumulation periods required for comparison to GPCP, 0–12, 12–24 or 24–36 h are likely candidate lead times for analysis. Longer forecasts are likely to degrade the quality of the estimates as the forecast model relaxes into a state which is less constrained by observations.

Fig. 9

The effect of spin-up on composite precipitation. The composites are taken from the 200 Atlantic storms shown in Fig. 3. Accumulations are shown for leadtimes 0–3, 12–15, 24–27 and 36–39 h. All figures are expressed in mm day⁻¹. Accumulations are from each 3-h window

In Fig. 9 composite precipitation taken from the 200 Atlantic storms in Fig. 3 is shown. The results for other regions are not materially different. The composites show accumulated precipitation at variable lead times from 0–3, 12–15, 24–27 and 36–39 h where the accumulation is from each 3 h window.

The storm position is centred on the analysed location at the time the precipitation fields are extracted, since to sample the forecast location would be laborious and add little value to the analysis. As such, there is some spatial offset in the location of the maxima in the longer lead times. This is because ETCs tend to propagate too slowly in the forecast model, giving the impression that the precipitation maxima move closer to the storm centre as lead time increases (see Froude et al. 2007a, b). It is clear that the 0–3 h accumulation has lower accumulated precipitation than the longer lead times. It is also apparent that the intensity of the precipitation maxima steadily degrades with lead times beyond 12 h.

Fig. 10

Composite precipitation intensities. The left panels show a composite at 12-h lead time, with the locations of the cross-sections (A–F) in the corresponding right hand panels overlaid. Lead times in the right hand panels are 0–3 (dotted line), 12–15 (solid), 24–27 (long dashes) and 36–29 h (short dashes). All figures in mm day⁻¹. Accumulations are from each 3-h window

Fig. 11

ERAI precipitation accumulations used for the daily periods in this analysis. Precipitation is extracted from 12 to 24 h in two forecasts which are initialised at 12 UTC on the previous day (FC1) and 00:00 UTC on the day of interest (FC2). The two accumulations are then combined

This is further evident in the precipitation cross sections shown in Fig. 10. Given the location of the precipitation maximum changes due to the storm centring, the cross-sections are shown for a number of locations, though the differences between the lead times remain clear. As a result of the adjustment effect in the 0–3 h period and the steady decay observed in the precipitation maxima and structure in the 24 and 36 h lead times, the forecast periods utilised in this study for comparison to GPCP are accumulations from 12 to 24 h from forecasts starting at 12:00 UTC the previous day and 00:00 UTC on the day of interest, as shown in Fig. 11. The two forecast accumulations are combined to provide daily precipitation estimates for comparison to GPCP. The 12- to 24-h forecast estimates have previously been found to compare well to gridded gauge data (Simmons et al. 2010), with longer lead times degrading the quality of the estimates (Kobold and Sušelj 2005). The results of de Leeuw et al. (2014) also indicate that this lead time offers the best estimates available from ERAI given the daily accumulations required in this study.