Dynamical Downscaling Forecasts

Nested RCMs provide an essential component of the model hierarchy. They enable the predictability of regional climate processes to be studied in much greater spatial detail, and provide a means to make downscaled seasonal climate predictions for appli-

Fig. 2.7. Observed (solid line) and RSM simulated (dashed line) drought index for the season of February-March-April in the rainfed agriculture region of Ceara. The temporal correlation between them is 0.74 (Sun et al. 2007)

cations. Dynamical downscaling of GCM climate forecasts has been performed in several regions (Diez et al. 2005; Druyan et al. 2002; Fennessy and Shukla 2000; Murphy 1999; Sun et al. 2006; Syktus et al. 2003). Most of them are experimental forecasts. To our knowledge, the first and the only operational climate dynamical downscaling prediction system is the one developed for northeast Brazil (Sun et al. 2006). Operational downscaling forecasts have been issued for northeast Brazil since December 2001. The NCEP RSM with a resolution of 60 km and ECHAM4.5 GCM (T42) are the core of this prediction system. This is a two-tiered prediction system. SST forecasts are produced first, which then serve as the lower boundary condition forcing for the ECHAM4.5 GCM-NCEP RSM nested system.

Two SST scenarios are predicted. The first SST scenario is to persist the observed SST anomalies from the most recently completed calendar month and add them to the observed seasonal cycle. Dynamical predictions using persisted SST anomalies are run only one season into the future. The second SST scenario is the predicted SST anomalies for the upcoming six months. A mix of dynamical and statistical models has been used to construct the SST predictions, varying by tropical ocean basins, and damped persistence with 3 months e-folding time has been used for the extratropical oceans.

Dynamically downscaled forecasts during 2002-2004 have been validated using the ranked probability skill score (RPSS). The overall rainfall forecast skill is positive over a majority of the Nordeste. Forecast skill varies with seasons. The forecast skill is generally higher for March-April-May (MAM) and AMJ seasons than JFM and FMA seasons (Fig. 2.8).

To examine the added value of the RSM forecasts, skill comparison between the downscaled forecasts and the driving global model forecasts was performed. The ECHAM4.5 GCM probability forecasts were generated using the same methods as the

48" W 44° W 40" W 3i'W 32* W 48" W 44'W 40* W 36*'W 32" W

Fig. 2.8. Geographical distributions of RPSS (%) averaged for the one-month lead forecasts during 2002-2004; a all season mean; b JFM season; c FMA season; d MAM season; e AMJ season

48" W 44° W 40" W 3i'W 32* W 48" W 44'W 40* W 36*'W 32" W

Fig. 2.8. Geographical distributions of RPSS (%) averaged for the one-month lead forecasts during 2002-2004; a all season mean; b JFM season; c FMA season; d MAM season; e AMJ season

RSM forecasts. The ECHAM4.5 GCM probabilistic forecasts were linearly interpolated onto RSM grids in order to calculate the RPSS using the high-resolution observations. The GCM forecast scores were aggregated for the whole Nordeste. As shown in Table 2.2, the

Table 2.2. Skill comparison between one-month lead RSM forecasts and the driving ECHAM GCM forecasts. The RPSS (%) is aggregated for the Nordeste region

JFM FMA MAM AMJ

ECHAM RSM ECHAM RSM ECHAM RSM ECHAM RSM

Table 2.2. Skill comparison between one-month lead RSM forecasts and the driving ECHAM GCM forecasts. The RPSS (%) is aggregated for the Nordeste region

JFM FMA MAM AMJ

ECHAM RSM ECHAM RSM ECHAM RSM ECHAM RSM

2002

7.1

4.5

5.2

10.1

14.9

23.5

1.2

16.9

2003

-6.1

-3.2

-2.6

7.2

9.4

15.3

5.4

12.2

2004

25.7

-7.4

-0.8

0.4

-5.7

28.6

5.8

18.5

scores of the RSM forecasts are higher than those of the driving GCM forecasts for most seasons, implying that the smaller spatial scale rainfall generated by the RSM is skillful.

Skill scores are based on 12 forecasts. Thus, the results here may be subject to high sampling variability. More reliable skill should be obtained using large forecast samples.

The dynamical downscaling prediction system for northeast Brazil continuously evolves, reflecting continued improvement. A new forecast product, the weather index was issued in January 2005. The weather index uses the daily rainfall time series to measure the severity of drought and flooding conditions. It has been successfully demonstrated that crop yields in the rainfed agriculture region are highly related to the weather index, and the downscaling prediction system is skillful to predict this index (Sun et al. 2007). The NCAR CCM3 and the CSU Regional Atmospheric Modeling System (RAMS) will also be added to this downscaling forecast system, and multimodel ensembling methods will be implemented to consolidate the downscaling forecasts in 2006.

Future Directions 2.5.1

Improved Model Physics and Parameterizations

Parameterization schemes are based on a spectral gap between the scales being parameterized and those being resolved on the model grid. Therefore, all parameterization schemes are model resolution dependent. However, parameterizations in most regional models are the same as those used in GCMs. These may not be an adequately representation of physics processes in the regional models and may result in incorrect model climatologies and climate drift, which offset the effect of high resolution of the regional model. For instance, the assumption that convective response rapidly to changes in a large-scale, slowly evolving circulation is appropriate for convection parameterizations in GCMs, but it is probably inappropriate for simulations of most mesoscale convective systems in regional models, with 10-50 km grid spac-ings (Frank and Cohen 1987). Arriving at more general mixing schemes that can cope with the wide range of model resolution is a key problem of relevance to dynamical downscaling.

Land Initialization

Traditionally, the land conditions in regional models are initialized by the driving GCM land conditions using an interpolation scheme. However, the coarse resolution GCM and the fine resolution RCM "see" the land differently due to the heterogeneity of the land surface. This kind of initialization introduces erroneous land surface forcings to the region model. This is as important a limitation on dynamical downscaling as model flaws. The primary problem lies in the land because of the heterogeneity of the land and very limited observations. A major advancement now being proposed is in situ monitoring and generating fine resolution land reanalysis data. Because the land contains the "memory" for time scales longer than a few weeks, land initialization is of fundamental importance in improving dynamical downscaling at seasonal time scale.

Nesting Strategy

For the traditional one-way nesting method, the full large-scale circulation fields from the GCM are provided to the regional model with an interval of 3-12 hours. Errors in the large-scale circulations of the driving GCM are transmitted to the nested RCM, which often results in poor simulations or forecasts of the RCM. For instance, the ECHAM GCM produces a strong divergence bias in the lower troposphere over East Africa. When the traditional nesting method is used, this GCM bias suppresses the convection development in the RCM, and results in a dry bias for the RCM rainfall prediction. To reduce the errors in the driving large-scale fields, an anomaly nesting method has been introduced. This nesting method is based on the premise that systematic errors can be eliminated by replacing the driving GCM climatology with the observed climatology. A case study of dynamical downscaling of seasonal climate over South America reveals that the substantial gains are realized through anomaly nesting (Misra and Kanamitsu 2004). More tests on this method are needed. Reduction of errors in the driving GCM fields can significantly improve the regional model performance.

Downscaling Forecasts - Linking Prediction and Application

Current forecast products generally lack the spatial, temporal and element specificity that users seek for their particular decision-making needs - forecasts are generally made for 3-month seasons, large regions over 1000 km in width, and mean temperature and precipitation totals only. Dynamical downscaling shows potential to improve climate forecasts towards users' need. Development of downscaling forecasts system, particularly for developing regions, helps with the design of policies to reduce the climate vulnerability of the most needed populations.

Was this article helpful?

0 0

Post a comment