Environ. Res. Lett. 10 (2015) 064011 I Takayabu et al weekly ensemble prediction model (WEP), which is the operational model used for weekly forecasts (Sakai, 2009). WEP, which is a global spectral model (GSM) with an equivalent grid point resolution of 60 km, was used to obtain the initial and boundary conditions, but the initial perturbations were calculated with a lower resolution GSM (equivalent horizontal grid resolution is 200 km). The ensemble had 51 members, prepared using the singular vector method (Buizza and Palmer 1995). Time integration was from 12 UTC on 4 November 2013 until 12 UTC on 11 November 2013. To avoid the occurrence of a shock at the lateral boundary, we used a cascade of nested models, downscaling finally to a WRF model with a 1 km grid resolution. The first downscaling step was to a regional model with a 20 km grid resolution (NHRCM20) (Sasaki et al 2011). To reproduce the TC track of the parent model in the downscaled model, we adopted spectral nudging above 15 000 m. The next downscaling step was to the regional model with a 5 km grid resolution (NHRCM05), and then two-way nesting from 3 km WRF to 1 km WRF models. In NHRCM05, the Kain-Fritsch convective parameterization scheme (Kain and Fritsch 1990, 1993) is used, and in the WRF models, cloud microphysical processes are explicitly handled. Dynamical downscaling using 3 km/1 km WRF models has been done, but not for all ensemble members of WEP calculations because of the limitation of the computer resources. To cover the wide range of ensemble members, the following three cases are selected. They are, Case m02: where the track is nearest to the observed track (best track) around Leyte and Samar Island (124.8E). Case m11: where the minimum central pressure (MCP) of the typhoon calculated in NHRCM05 is lowest among all. Case p18: where the MCP of NHRCM05 is the highest among all. Because the timing of starting of the integration also affects MCP, the following three times are selected as the initial conditions. They are, #1000: 12UTC, 5, November, #1001: 18UTC 5, November, and #1002: 00UTC 6, November. Thus we have 9 experiments following the above conditions. Furthermore, to estimate the storm surge around Tacloban caused by Typhoon Haiyan, we select all cases where the typhoon passed around Leyte Island (125E), within the range of 50 km to the best track data. Seven cases (05 m, 12 m, 15 m, 21 m, 25 m, 06 p, 09 p) are found other than Case m02. For these seven cases, only #1001 (initial condition at 18UTC 5, November) has been calculated. As a result, we calculate 16 cases by using 3 km/1 km WRF models. For the NAT ensemble simulations, we applied Pseudo Global Warming Downscaling (PGWD), in which the boundary conditions were assumed to be a linear coupling of the WEP data and the difference component of the climate change of air temperature 3 between the middle of the nineteenth century and 2013. To estimate the hypothetical natural SST, we removed linear trends in the monthly data from the HadISST (Hadley Centre Sea Ice and Sea Surface Temperature) data set (Rayner et al 2003) for 1870–2012 from the original SST data used in the ALL runs (Christidis and Stott 2014, Shiogama et al 2014). The difference between ALL-SST and NAT-SST in the target area (100°–180°E, 5°S–25°N) was between 0.2 K and 0.8 K (as shown in figure S2). To remove anthropogenic atmospheric warming, we subtracted the differences of atmospheric temperature between two 100-member ensemble simulations of the MIROC5 AGCM (Shiogama et al 2014, Watanabe et al 2010), performed with and without human influence, from the WEP data. The hypothetical anthropogenic SST signals of the MIROC5 ensemble are the same as in our regional climate model runs. We averaged the atmospheric temperature differences over the whole target area during November of 2010–2012 across 100 ensemble members. The averaged tropospheric and stratospheric temperature differences were +0.5 K and –1.5 K, respectively. We call this effect ‘ATM’. For the downscaling calculations, we used GHG (CO2, CH4 and N2O) levels in the 1850s reported by the Intergovernmental Panel on Climate Change Fifth Assessment Report (IPCC 2013). The Surge-Wave-Tide coupled model (SuWAT) (Kim et al 2008) was forced by the 1 km WRF model calculation result. SuWAT is a fully coupled model of storm surge and ocean waves based on the non-linear shallow-water equation and spectral wave model, and takes into account atmospheric pressure, wind stress, and wave radiation stress (wave effects on current). Input data are surface wind and sea level pressure. We applied SuWAT to three domains with spatial resolutions ranging from 0.1° (D1) to 740 m (D3), with twoway nesting. The astronomical tide was excluded for computation and survey for simplicity, although the maximum astronomical tidal range is 0.7 m and was close to mean water level at the landfall of Haiyan. 3. Results 3.1. Ensemble simulation of Typhoon Haiyan The evolution and intensity of Haiyan were well represented in all of the experiments. In WEP, the simulated typhoon tracks around Leyte and Samar Island (125°E), on which Tacloban is situated, were distributed in a band about 500 km wide in the north–south direction with its centre north of Tacloban, and the tracks in the downscaled results followed the WEP result (figure 2). The westwardpropagation speed of the typhoon was slower in the model simulations, compared with the observed track (best-track data), and as a result, the TC landfall on Leyte and Samar Island was delayed about a half