-
The numerical weather prediction model used in this study is the mesoscale version of the Global/Regional Assimilation and PrEdiction System (GRAPES_Meso), which is a new generation of numerical weather forecasting systems for mesoscale weather prediction developed by the China Meteorological Administration (CMA) (Chen and Shen, 2006). The model adopts a height-based terrain-following coordinate, a semi-implicit and semi-Lagrangian (SI-SL) time difference scheme, a fully compressible non-hydrostatic balance dynamic framework, and a physical parameterization package (Wu et al., 2005; Chen et al., 2008). The physical parameterization schemes selected in this study include the rapid radiative transfer model (RRTM) for the long-wave scheme (Rosenkranz, 2003), the Dudhia shortwave radiation scheme (Dudhia, 1996), the WRF single-moment six-class (WSM6) microphysics scheme (Hong and Lim, 2006), the Noah land surface scheme, the Monin–Obukhov surface layer scheme (Johansson et al., 2001), and the medium-range forecast (MRF) planetary boundary layer scheme (Hong and Pan, 1996).
The model uses a single domain, which covers the area 15°–30°N, 104°–122.9°E, with a horizontal grid spacing of 3 km (
$ 631\times 501 $ grid points) and 51 vertical layers reaching up to 33 km. The time step of the model is 30 s. The initial and lateral boundary conditions are obtained from NCEP GFS analyses at 0.25° resolution. -
Some numerical experiments were designed to evaluate the performance of the multi-scale IAU scheme. The blending experiment, illustrated in red in Fig. 2, was first designed to ensure that the regional model derived a physically valid state after initialization (Ulmer and Balss, 2016). The background field was integrated from the initial condition of the regional model and downscaled from the global analysis at
${(t}_{0}-6)\;\mathrm{h}$ , with a 6 h warm-up time. Then, the background field$ {G}_{\mathrm{x}\mathrm{b}} $ at the analysis time$ {t}_{0} $ was generated. The GRAPES_Meso 3DVAR (Xue et al., 2008) and the blending method were successively applied at$ {t}_{0} $ . The Gaussian correlation model was used in the GRAPES_Meso 3DVAR system. The horizontal correlation length of specific humidity was shorter at 200 km, while that of the other variances was 500 km. The 3DVAR analysis field$ {G}_{\mathrm{x}\mathrm{a}} $ at$ {t}_{0} $ was obtained by using the 3DVAR system. After that, the 2-D DCT blending method mentioned in section 2.1 was immediately performed to obtain the blended analysis field$ {G}_{\mathrm{b}\mathrm{l}\mathrm{n}\mathrm{d}} $ at$ {t}_{0} $ . Specifically, the large-scale analysis increment from NCEP GFS$ {G}_{\mathrm{g}\mathrm{l}\mathrm{o}\mathrm{b}\mathrm{a}\mathrm{l}} $ and the small-scale analysis increment from the 3DVAR analysis field$ {G}_{\mathrm{x}\mathrm{a}} $ were blended to create the blended analysis increment. After the background field$ {G}_{\mathrm{x}\mathrm{b}} $ was replaced by the blended analysis field$ {G}_{\mathrm{b}\mathrm{l}\mathrm{n}\mathrm{d}} $ , the model performed a 36-h continuous forecast.Figure 2. Illustration of two categories of numerical experiments, which included the blending experiment (red lines) and the IAU experiment (blue lines). In the blending experiment, the background field
$ {\mathrm{G}}_{\mathrm{x}\mathrm{b}} $ was integrated from the global analysis field at (t0−6) h, and the 3DVAR and blending methods were successively applied at the analysis time t0 to obtain the blended analysis field$ {G}_{\mathrm{b}\mathrm{l}\mathrm{n}\mathrm{d}} $ . In the IAU experiment, the background field was also integrated at (t0−6) h, and the increment, which was calculated in the blending experiment, was added to the integration at every time step during the time window. Both the blending experiment and the IAU experiment performed a 36-h continuous forecast after t0.In addition to the blending experiment, the IAU experiment was also designed and presented in blue in Fig. 2. The background field was integrated from the global analysis field at
$ ({t}_{0}-6)\;\mathrm{h} $ , and the increment obtained from the blending experiment was treated as constant forcing in a model’s prognostic equation through the IAU scheme over the time window$ \left({t}_{0}-\tau /2,{t}_{0}+\tau /2\right) $ . Here, the center of the time windows was at$ {t}_{0} $ , and the relaxation time was$ \tau $ . After$ {t}_{0} $ , the model performed a 36-h continuous forecast.For the IAU scheme, the relaxation time,
$ \tau , $ is a crucial parameter that determines the filtering properties of the analysis increments. IAU experiments with different configurations were designed to test initialization sensitivity to relaxation time and find the optimal relaxation time (ORT) for each scale increment. In the seven large-scale IAU experiments, the large-scale increment was added into the integration at every time step over different relaxation times in Typhoon Mangkhut (2018). The CMA tropical cyclone (TC) database was chosen to verify the TC track and intensity forecasts (Ying et al., 2014; Lu et al., 2021). Based on the verified results, the ORT of the large-scale increment could be determined. Similarly, seven small-scale IAU experiments were also designed to obtain the ORT of the small-scale increment. For the multi-scale IAU experiment, the relaxation times of the large- and small-scale increments were derived from the ORTs of the large-scale and small-scale IAU experiments, respectively. Finally, three all-scale IAU (traditional IAU) experiments and a control experiment (without IAU) were also conducted to evaluate the performance of the multi-scale IAU scheme. Table 1 shows the configuration of all numerical experiments in detail.Category Analysis increment Relaxation time Control experiment Without increment Without IAU Blending experiment Large-scale + small-scale Without IAU Large-scale IAU experiment Large-scale Without IAU 0.5 h, 1 h, 1.5 h, 3 h, 6 h, and 9 h Small-scale IAU experiment Small-scale Without IAU 0.5 h, 1 h, 1.5 h, 3 h, 6 h, and 9 h All-scale IAU experiment Large-scale + small-scale 1 h, 3 h, and 6 h Multi-scale IAU experiment Large-scale ORT* for large-scale increment Small-scale ORT* for small-scale increment * ORT represents the optimal relaxation time in the IAU scheme. Table 1. Configuration of 20 numerical experiments.
Category | Analysis increment | Relaxation time |
Control experiment | Without increment | Without IAU |
Blending experiment | Large-scale + small-scale | Without IAU |
Large-scale IAU experiment | Large-scale | Without IAU |
0.5 h, 1 h, 1.5 h, 3 h, 6 h, and 9 h | ||
Small-scale IAU experiment | Small-scale | Without IAU |
0.5 h, 1 h, 1.5 h, 3 h, 6 h, and 9 h | ||
All-scale IAU experiment | Large-scale + small-scale | 1 h, 3 h, and 6 h |
Multi-scale IAU experiment | Large-scale | ORT* for large-scale increment |
Small-scale | ORT* for small-scale increment | |
* ORT represents the optimal relaxation time in the IAU scheme. |