For initial errors that has minute amplitude (10% to that of the current-day operational IC error), the growth of the errors in short-term shows very good consistency with the upscale error growth model proposed in Zhang et al. (2007). The errors grow through convective instability first, confined to the precipitation region, and then expand outward and propagate to large-scale balanced fields through adjustment process. After ~3 days, the large-scale error induced by this cascade-like mechanism is comparable to our current-day operational IC uncertainty. Given the existence of a mesoscale -5/3 energy-cascading range and the decreasing eddy turn-over time with decreasing scale, further reducing the initial amplitude of the error does not give us much gain on our prediction (Sun and Zhang 2016).
For EDA experiments with initial errors that have amplitude and spatial structure similar to our current-day operational analysis, the exponentially growth of the errors due to baroclinic instability dominates in short term. This is also the case for the minute perturbation experiments after 3 days, when the error in these experiments reach our current-day IC uncertainty. At regions where strong baroclinic wave packets exist, error could propagate at the group velocity of the baroclinic waves and signature of downstream development of the errors could be identified. After ~10 days in the EDA experiments, the synoptic-scale errors (~3000km) start to saturate and baroclinic growth of the errors becomes much slower, even quasi-stagnant, which means that we are reaching our current practical limit of the synoptic-scale systems. The errors in the 0.1 EDA experiment also start to catch up with that in the EDA experiments after 10 days.
Error growth at later stage are strongly modulated by the stationary waves, with strong zonal asymmetries during winter season. At the end of our 20-day integration, the error still grows at the largest scale, albeit at a really slow rate, implying possible interactions between the synoptic scale eddies and the stationary waves. These interactions might have strong impact on our sub-seasonal and seasonal forecast, yet it’s beyond the scope and the time scale of this study. Further investigation is needed.
Experiments in the summer season give us similar picture with the winter cases. Difference shows up mainly at the later stage. The background flow in the summer has weaker baroclinicity and less stationary wave signals. Correspondingly, the baroclinic growth of the error is slower and the error distribution at later stage is more zonally symmetric, compared to its winter counterparts. Nonetheless, the error saturation time of the synoptic-scale systems in summer is comparable with that of the winter cases.