Abstract

An ensemble of 1-month-lead seasonal retrospective forecasts generated by the Scale Interaction Experiment (SINTEX)–Frontier Research Center for Global Change (FRCGC), version 2 tuned for performance on a vector supercomputer (SINTEX-F2v), coupled global circulation model (CGCM) were downscaled using the Weather Research and Forecasting (WRF) Model to improve the forecast of the austral summer precipitation and 2-m air temperatures over Australia. A set of four experiments was carried out with the WRF Model to improve the forecasts. The first was to drive the WRF Model with the SINTEX-F2v output, and the second was to bias correct the mean component of the SINTEX-F2v forecast and drive the WRF Model with the corrected fields. The other experiments were to use the SINTEX-F2v forecasts and the mean bias-corrected SINTEX-F2v forecasts to drive the WRF Model coupled to a simple mixed layer ocean model. Evaluation of the forecasts revealed the WRF Model driven by bias-corrected SINTEX-F2v forecasts to have a better spatial and temporal representation of forecast precipitation and 2-m air temperature, compared to SINTEX-F2v forecasts. Using a regional coupled model with the bias-corrected SINTEX-F2v forecast as the driver further improved the skill of the precipitation forecasts. The improvement in the WRF Model forecasts is due to better representation of the variables in the bias-corrected SINTEX-F2v forecasts driving the WRF Model. The study brings out the importance of including air–sea interactions and correcting the global forecasts for systematic biases before downscaling them for societal applications over Australia. These results are important for potentially improving austral summer seasonal forecasts over Australia.

1. Introduction

Seasonal forecasting of precipitation in the austral summer months, from December to February (DJF), is beneficial for the agro-based regions of northern Australia (landmass to the north of 25°S). During DJF, northern parts of Australia experience monsoon climate (Wheeler and McBride 2005; Hendon et al. 2012) with reversal of low-level winter winds from easterlies to westerlies and increased precipitation. In fact, northern Australia receives most of its annual rainfall during the austral summer season. However, it is also a season with low predictability (Hendon et al. 2011, 2012). El Niño–Southern Oscillation (ENSO), the dominant climate mode influencing Australian climate (Allan 1988; McBride and Nicholls 1983; Risbey et al. 2009; Hendon et al. 2011; King et al. 2014), has less impact on the Australian summer precipitation compared to the premonsoon season (Nicholls et al. 1982; McBride and Nicholls 1983; King et al. 2014), leading to reduced forecast skill in the austral summer season. The composite of precipitation anomalies in the ENSO years indicates the impact of ENSO to be mostly confined to Cape York Peninsula during the austral summer season (Hendon et al. 2011; http://www.bom.gov.au/climate/enso/ninocomp.shtml), with significant increase (decrease) of precipitation in La Niña (El Niño) years over the region. Hendon et al. (2012) attribute the reduction in the forecast skill, between the austral summer season and the premonsoon season, to reduced air–sea interactions in the tropical oceans to the north of Australia. Analyzing the seasonal retrospective forecasts of the Predictive Ocean Atmosphere Model for Australia (POAMA), Hendon et al. (2012) found the easterlies in the premonsoon season to enhance the air–sea interaction feedbacks, resulting in stronger and persistent sea surface temperature (SST) anomalies to the north of Australia. However, the air–sea interactions are weaker in monsoon westerlies in the austral summer season, leading to weaker SST anomalies to the north of Australia, and thus reducing the predictability of the austral summer precipitation.

The purpose of this study is twofold: first, to evaluate the performance of the Scale Interaction Experiment (SINTEX)–Frontier Research Center for Global Change (FRCGC), version 2 tuned for performance on a vector supercomputer (SINTEX-F2v; Doi et al. 2014, 2016), global circulation model in forecasting the austral summer precipitation over Australia, and second, to investigate if dynamical downscaling of the SINTEX-F2v forecasts using the Weather Research and Forecasting (WRF) Model (Skamarock et al. 2005) can improve the skill of SINTEX-F2v forecasts over Australia during the difficult-to-predict summer season.

The technique of dynamical downscaling, using a regional climate model (RCM), is often used to improve the spatial and temporal representation of precipitation simulated by coarse-resolution general circulation models (GCMs) (Dickinson et al. 1989; Giorgi and Bates 1989; McGregor 1997). The improvement is because of better representation of orography and the land-use pattern in RCMs compared to GCMs. Regional climate models, driven by both reanalysis data and the output from GCMs, have been applied to study various aspects of the Australian climate (e.g., McGregor and Walsh 1993, 1995; Evans and McCabe 2010; Evans et al. 2012; Kala et al. 2015; and the references therein). McGregor and Walsh (1993) reported improvement in simulation of January climate over Australia by nesting a limited-area model into a GCM. The improvements were more evident near the regions of steep topography, which are better resolved in regional models compared to GCMs. Similarly, Walsh and McGregor (1995) reported improved results for the austral summer and winter months. Studies with the WRF regional model revealed added value to the driving reanalysis data over the Murray–Darling River basin in southeastern Australia (Evans and McCabe 2010). A few studies, for example Evans et al. (2012) and Kala et al. (2015), found the WRF Model simulation to be sensitive to physical parameterization schemes used in simulating extreme rainfall events and the seasonal climate of Australia. In addition, Kala et al. (2015) found that the WRF simulation is dependent on the type of reanalysis data used at the boundaries to drive the model. There have been no studies showing the importance of the dynamical downscaling in improving the seasonal forecasts over Australia during the austral summer season using the WRF Model. This study aims to fill that gap by downscaling seasonal forecasts and evaluating the forecasts.

Regional climate models, because of their binding domains, require either reanalysis data or an output from a GCM to drive them. As a result, the skill of an RCM is dependent on the quality of the driving data and is often reduced as a result of systematic biases embedded in the driver GCMs (Christensen et al. 1998). The biases in GCM forecasts are amplified in an RCM because of increased resolution, leading to deterioration of the RCM forecasts. Therefore, it is necessary to reduce at least the systematic biases in GCM forecasts before using the forecasts to drive RCMs. One of the techniques often used to reduce the systematic biases in a GCM forecast is to substitute GCM forecast climatology with observed climatology while superimposing the GCM forecast anomalies on it (Misra and Kanamitsu 2004; Holland et al. 2010; Xu and Yang 2012; White and Toumi 2013; Bruyère et al. 2014; Ratnam et al. 2016). Misra and Kanamitsu (2004) indicate this methodology significantly improves RCM-simulated precipitation over South America compared to RCM simulation without bias correction in the driver GCM simulation. Using a similar technique, Holland et al. (2010) and Bruyère et al. (2014) demonstrate an improvement in the simulation of tropical cyclones in a regional climate model. In this study we investigate if such a technique can improve the RCM forecast skill over Australia.

Recent studies (Ratnam et al. 2009, 2012, 2013, 2015; Kim and Hong 2010; Peng et al. 2012, and the references therein) reveal the importance of using atmosphere–ocean coupled regional models in improving the simulation of precipitation, especially over the tropical regions, compared to standalone atmospheric RCMs. The improvements are attributed to better representation of air–sea interaction feedbacks in the atmosphere–ocean coupled models. The two-tiered approach of specifying SST in atmosphere-only models in the regions where atmosphere is strongly forced by the SST often leads to spurious rainfall (Krishna Kumar et al. 2005; Hendon et al. 2012), thereby reducing skills of the forecasts. In this study we employ a coupled model, with WRF as the atmospheric component and a simple mixed layer ocean model as the oceanic component, to dynamically downscale SINTEX-F2v forecasts to investigate if the coupled model can improve the SINTEX-F2v precipitation forecasts.

In the following sections the model and methods used in the study are described, followed by the results.

2. Model and methodology

In this study, we use the WRF (version 3.6.1) Model to dynamically downscale SINTEX-F2v retrospective seasonal forecasts. The WRF Model, with a horizontal resolution of 27 km (Fig. 1a) and 30 vertical levels extending from surface to 50 hPa and with a domain covering the region 45°–10.0°S, 103°–164°E is used in the study. The WRF Model domain includes the Australian landmass, the eastern south Indian Ocean, and the western South Pacific Ocean (Fig. 1a). The domain is big enough to allow the model to develop regional features while restricting the model from deviating from the driving GCM output. The physics packages used in the WRF Model for the study are: (i) the Rapid Radiative Transfer Model (RRTM) for the longwave radiation (Mlawer et al. 1997), (ii) a simple cloud-interactive shortwave radiation scheme (Dudhia 1989), (iii) the Yonsei University (YSU) planetary boundary layer scheme (Hong et al. 2006), (iv) the unified Noah land surface scheme (Tewari et al. 2004) with 24 USGS land surface categories, and (v) the WRF single-moment 3-class (WSM3) microphysics scheme (Hong et al. 2004). To determine a suitable cumulus parameterization scheme for downscaling, we made model runs with two well-known convective schemes: 1) the Betts–Miller–Janjic (BMJ) cumulus parameterization scheme (Betts and Miller 1986; Janjić 1994), and 2) Kain–Fritsch (KF; Kain 2004) scheme with triggering function as implemented in the Japan Meteorological Agency nonhydrostatic model (Narita and Ohmori 2007).

Fig. 1.

(a) WRF Model domain used for downscaling experiments. Topography (m) of the region is shaded. The Cape York Peninsula region is shown by the red rectangle. (b) DJF mean (averaged over DJF 2000/01–DJF 2013/14) BOM_AWAP precipitation (mm day−1; shaded) and 850-hPa winds (m s−1; streamlines). (c) As in (b), but SINTEX-F2v forecast mean precipitation (mm day−1) and 850-hPa winds (m s−1). (d)–(h) As in (b), but forecast mean precipitation (mm day−1) and 850-hPa winds (m s−1) by forecast experiments WRF_BMJ_NBC, WRF_KF_NBC, WRF_KF_BC, WRF_KF_NBC_OC, and WRF_KF_BC_OC respectively.

Fig. 1.

(a) WRF Model domain used for downscaling experiments. Topography (m) of the region is shaded. The Cape York Peninsula region is shown by the red rectangle. (b) DJF mean (averaged over DJF 2000/01–DJF 2013/14) BOM_AWAP precipitation (mm day−1; shaded) and 850-hPa winds (m s−1; streamlines). (c) As in (b), but SINTEX-F2v forecast mean precipitation (mm day−1) and 850-hPa winds (m s−1). (d)–(h) As in (b), but forecast mean precipitation (mm day−1) and 850-hPa winds (m s−1) by forecast experiments WRF_BMJ_NBC, WRF_KF_NBC, WRF_KF_BC, WRF_KF_NBC_OC, and WRF_KF_BC_OC respectively.

The initial and the lateral as well as the lower boundary conditions for the WRF Model were derived from a six-member ensemble forecast system based on the SINTEX-F2v seasonal climate prediction system (Doi et al. 2014, 2016). SINTEX-F2v is a coupled ocean–atmosphere global model system with an atmospheric component, based on ECHAM5 (Roeckner et al. 2003), at a spectral resolution of T106 (~100 km) and with 31 vertical levels. The oceanic component, Nucleus for European Modelling of the Ocean (NEMO) modeling system (Madec et al. 2008), uses the ORCA-R05 grid at a resolution of 0.5° with 31 levels. The SINTEX-F2v system is similar to the SINTEX-F2 CGCM (Masson et al. 2012; Sasaki et al. 2012, 2013, 2014a,b) but is tuned for performance on the Earth Simulator, a vector supercomputer. The coupling between the atmospheric and oceanic components is performed every 2 h by exchanging sea surface temperature, sea ice fraction, freshwater, surface heat, surface current, and momentum fluxes using the Ocean Atmosphere Sea Ice Soil, version 3, coupler (OASIS 3; Valcke et al. 2004). The details of the model spinup and retrospective forecast experiments are given in Doi et al. (2016) and in Ratnam et al. (2016). The SINTEX-F1 system (Luo et al. 2005, 2007, 2008) is used for experimental forecasts (accessible at http://www.jamstec.go.jp/frsgc/research/d1/iod/sintex_f1_forecast.html.en). There is a plan to combine the forecasts of SINTEX-F1 system with this new SINTEX-F2v system in a multimodel ensemble experiment in the near future. SINTEX-F1 and SINTEX-F2v systems have demonstrated outstanding performance in predicting ENSO, the Indian Ocean dipole (IOD), and Ningaloo Niño (Doi et al. 2014, 2016), the primary drivers of the Australian climate. The SINTEX-F1 seasonal forecasts are used as guidance by Australian farmers (Doi et al. 2014) in agricultural practices. The potential utility of SINTEX-F2v forecasts to Australian society motivated us to investigate further if the SINTEX-F2v seasonal forecasts could be improved by dynamical downscaling.

We used 3 November initial conditions, 1-month-lead forecasts of SINTEX-F2v, to generate a 6-member ensemble of WRF forecasts targeted for the austral summer season (DJF) over Australia. The model runs were for the period from November–February (NDJF) 2000/01 to NDJF 2013/14. The SINTEX-F2v model does not generate the soil moisture and soil temperature fields required for the initialization of the unified Noah land surface scheme embedded in WRF. Instead, we used climatological (averaged over 1985–99) soil moisture and soil temperature derived from the ERA-Interim (Dee et al. 2011) dataset for the initialization of the land surface model. We used the ERA-Interim climatological soil moisture and soil temperature, as the ERA-Interim variables are not available in real time to the general public.

As mentioned in the previous section, the RCM (i.e., WRF) forecasts are affected by the systematic biases present in the GCM forecasts. In this study, we carried out two model experiments: first, driving the WRF Model with SINTEX-F2v forecasts (WRF_BMJ_NBC and WRF_KF_NBC) and, second, driving the WRF Model after correcting the systematic biases in the SINTEX-F2v forecasts (WRF_KF_BC). In this way, we try to bring out the importance of reducing the systematic biases in the GCM forecasts to improve the skill of RCM forecasts. The method followed for this bias correction is similar to the one used in the earlier studies of Misra and Kanamitsu (2004), Holland et al. (2010), White and Toumi (2013), Bruyère et al. (2014), and Ratnam et al. (2016). The procedures followed in the bias correction is as follows:

  • The 6-hourly SINTEX-F2v forecast data are broken down into a mean seasonally varying climatological component and a perturbation term: SINTEX-F2v = 〈SINTEX-F2v〉 + SINTEX-F2v′, where SINTEX-F2v is the model forecast, and 〈SINTEX-F2v〉 is the 6-hourly climatology of each member of the SINTEX-F2v ensemble; in this case, it is the 6-hourly mean of variables over the period NDJF 2000/01–2013/14, the period covering the SINTEX-F2v retrospective forecasts. SINTEX-F2v′ is the 6-hourly perturbation of each member of the ensemble derived from the 6-hourly climatology of each member.

  • Similarly, the ERA-Interim data, at a resolution of 0.75° × 0.75°, is broken down into a seasonally varying climatological component and a perturbation term: EraInt = 〈EraInt〉 + EraInt′, where EraInt is the ERA-Interim variable, and 〈EraInt〉 is the 6-hourly climatology of ERA-Interim variable, averaged over the period NDJF 1985/86–1999/2000. The period used for generating ERA-Interim 6-hourly climatology excludes the years of the WRF forecast. EraInt′ is the 6-hourly perturbation derived from the 6-hourly ERA-Interim climatology. The ERA-Interim climatology is interpolated to the SINTEX-F2v grids before merging both the fields.

  • Bias-corrected SINTEX-F2v forecasts are constructed by replacing SINTEX-F2v climatology with ERA-Interim climatology and superimposing the SINTEX-F2v perturbations on it: SINTEX-F2v corrected = 〈EraInt〉 + SINTEX-F2v′.

The above procedures are applied to the SINTEX-F2v forecasts at all levels and to all variables required for WRF Model forecasts (viz., geopotential height, air temperature, zonal and meridional wind, specific humidity, sea level pressure, and the sea surface temperature).

Further, we investigated if using an atmosphere–ocean coupled regional model would improve the skill of the RCM forecasts. The regional coupled model used in this study has the WRF Model as the atmospheric component and a simple mixed layer ocean model as the oceanic component. The mixed layer ocean model used in the WRF for the experiments follows that of Pollard et al. (1973), except that the implementation in WRF allows for nonzero mixed layer depth (Davis et al. 2008). In the scheme, the mixed layer is deepened and the water is cooled as a result of wind-driven mixing. We initialized the mixed layer model with a 50-m depth, which is approximately the observed mixed layer depth in November in the oceans surrounding Australia (de Boyer Montégut et al. 2004). Kim and Hong (2010) and Ratnam et al. (2013) used the WRF Model coupled to a simple mixed layer ocean model for studying the East Asian monsoon and the southern Africa climate, respectively. The mixed layer model includes the Coriolis effect on the model currents, and hence it influences the mixed layer heat budget. Both the previous experiments, WRF_KF_BC and WRF_KF_NBC, were repeated with the coupled regional model. The experiments are herein denoted as WRF_KF_BC_OC and WRF_KF_NBC_OC, respectively.

The model-generated precipitation is validated against the monthly Bureau of Meteorology (BoM) Australian Water Availability Project (AWAP) (BOM_AWAP) rainfall data (Jones et al. 2009), which is available at 5-km horizontal resolution. The forecast winds are compared with winds derived from ERA-Interim data at 1° × 1° resolution. The 2-m air temperatures forecast by the models are compared with the 0.5° × 0.5° resolution Global Historical Climatology Network (GHCN), version 2, and the Climate Anomaly Monitoring System (CAMS) dataset (GHCN+CAMS; Fan and Van den Dool 2008). The observation datasets and the SINTEX-F2v forecasts are linearly interpolated to WRF Model grids for uniformity in comparing the results.

The deterministic skill of the forecasts is calculated by the anomaly correlation coefficient (ACC) between the model forecast and the verification datasets using the following equation. The anomalies of both the forecasts and observations are calculated based on their respective climatologies from DJF 2000/01 to 2013/14:

 
formula

, and , where and are the forecast and observed anomalies with respect to the climatologies f and o. The value of m is from 1 to 14 to cover the period from DJF 2000/01 to DJF 2013/14.

3. Results

a. DJF mean precipitation forecast

Seasonal austral summer mean precipitation (Fig. 1b; shaded) of magnitude greater than 8 mm day−1 is observed over parts of northern Australia associated with seasonal mean cyclonic circulation in the low levels (at 850 hPa) (Fig. 1b; streamlines) over the region. The low-level cyclonic circulation transports moisture from the tropical warm oceans into northern Australia, thus resulting in an area of high precipitation. Precipitation of greater than 3 mm day−1 is also observed along the east coast of Australia during this season (Fig. 1b). The subtropical anticyclonic circulation over southern Australia at 850 hPa (Fig. 1b) inhibits precipitation, resulting in a region of low precipitation (<1 mm day−1) over southern and southwestern Australia during the season (Fig. 1b).

The north–south gradient in the austral summer precipitation over northern Australia (Fig. 1b) is realistically represented in the SINTEX-F2v seasonal forecast (Fig. 1c). However, the seasonal mean precipitation is underestimated over northern Australia, with most regions receiving precipitation of about 3–6 mm day−1. The low-level cyclonic circulation in the SINTEX-F2v seasonal forecast (Fig. 1c) has a westward shift over northern Australia, compared to observations (Fig. 1b). This systematic bias in the circulation results in a westward shift of the high-precipitation region observed over northern Australia (Fig. 1b), and precipitation of greater than 8 mm day−1 is confined to the northwestern parts of northern Australia. The root-mean-square error (RMSE) and bias between the SINTEX-F2v forecast and observed precipitation reveals that the SINTEX-F2v forecasts have an error of greater than 4 mm day−1 over the Cape York Peninsula and over regions to the west of the Cape York Peninsula (Fig. 2a). This is due to significant underestimation of precipitation (Fig. 2e) over those regions. The underestimation of forecast precipitation over northern Australia during the austral summer season is because of the biases in the SINTEX-F2v forecast low-level winds. The SINTEX-F2v model also underestimates the orographic precipitation along the east coast of Australia (Fig. 2e).

Fig. 2.

RMSE of the ensemble mean (a) SINTEX-F2v, (b) WRF_BMJ_NBC, (c) WRF_KF_NBC, and (d) WRF_KF_BC DJF forecast precipitation (mm day−1) with respect to BOM_AWAP estimated precipitation. The significant bias in the forecast precipitation of (e) SINTEX-F2v, (f) WRF_BMJ_NBC, (g) WRF_KF_NBC, and (h) WRF_KF_BC. (i),(j) The RMSE and (k),(l) the significant bias in the forecast precipitation of (i),(k) WRF_KF_NBC_OC and (j),(l) WRF_KF_BC_OC. (m) The RMSE and (o) significant difference between WRF_KF_NBC_OC and WRF_KF_NBC; similarly, (n) the RMSE and (p) significant difference between WRF_KF_BC_OC and WRF_KF_BC. Significance is at 95% using the two-tailed Student’s t test.

Fig. 2.

RMSE of the ensemble mean (a) SINTEX-F2v, (b) WRF_BMJ_NBC, (c) WRF_KF_NBC, and (d) WRF_KF_BC DJF forecast precipitation (mm day−1) with respect to BOM_AWAP estimated precipitation. The significant bias in the forecast precipitation of (e) SINTEX-F2v, (f) WRF_BMJ_NBC, (g) WRF_KF_NBC, and (h) WRF_KF_BC. (i),(j) The RMSE and (k),(l) the significant bias in the forecast precipitation of (i),(k) WRF_KF_NBC_OC and (j),(l) WRF_KF_BC_OC. (m) The RMSE and (o) significant difference between WRF_KF_NBC_OC and WRF_KF_NBC; similarly, (n) the RMSE and (p) significant difference between WRF_KF_BC_OC and WRF_KF_BC. Significance is at 95% using the two-tailed Student’s t test.

Not surprisingly, the WRF Model forecasts with different cumulus parameterization schemes and driven by SINTEX-F2v forecasts reveal differences in forecast precipitation (WRF_BMJ_NBC and WRF_KF_NBC in Figs. 1d and 1e, respectively) because of the convective nature of the austral summer precipitation over Australia. The precipitation forecast by WRF_BMJ_NBC over northern Australia is about 3–4 mm day−1, a clear underestimation of precipitation compared to the observed precipitation of greater than 8 mm day−1 over that region, whereas the WRF_KF_NBC forecast represents the precipitation realistically, with precipitation of magnitude greater than 8 mm day−1 over northern parts of Australia. The RMSE reveals large errors in the WRF_BMJ_NBC forecasts over northern Australia (Fig. 2b) resulting from underestimation of precipitation (Fig. 2f). Surprisingly, the biases in the austral summer WRF forecast with the BMJ scheme (WRF_BMJ_NBC; Fig. 2f) are larger compared to the biases in the SINTEX-F2v forecasts (Fig. 2e), indicating a deterioration of forecast skill from dynamical downscaling using the BMJ scheme. On the other hand, the WRF Model forecast with the KF scheme (WRF_KF_NBC) improves the seasonal mean precipitation of the SINTEX-F2v forecast over northern Australia (Fig. 1e) and has smaller RMSE values (Fig. 2c) and biases (Fig. 2g) over northern Australia, thereby demonstrating the importance of this particular cumulus parameterization scheme in the regional model while downscaling the precipitation over Australia during the austral summer season. Interestingly, the systematic bias of westward shift in the seasonal mean low-level cyclonic circulation over northern Australia in the SINTEX-F2v forecast (Fig. 1c) is corrected in both the WRF_BMJ_NBC and WRF_KF_NBC forecasts; however, the cyclonic circulation is shifted northward of the Australian landmass in the WRF_BMJ_NBC forecast (Fig. 1d) compared to the observed circulation (Fig. 1b). This results in less rainfall over the northern parts of Australia in the WRF_BMJ_NBC forecast (Fig. 1d). The seasonal mean cyclonic circulation over northern Australia is also shifted northward in the WRF_KF_NBC forecast (Fig. 1e), although less than in the WRF_BMJ_NBC forecast (Fig. 1d). Because of the northward shift of the cyclonic circulation, the precipitation has a large RMSE over some parts of northern Australia (Fig. 2c), with underestimation of precipitation over parts of northern Australia (Fig. 2g). In agreement with the studies of Kala et al. (2015), the southwest region of Western Australia shows little sensitivity to either of the prescribed cumulus parameterization schemes in the model: both WRF_BMJ_NBC and WRF_KF_NBC forecasts have comparable precipitation with small RMSE (Figs. 2b,c) and bias in the region (Figs. 2f,g). Further experimentation to improve the WRF seasonal forecasts was carried out with the KF scheme because of low RMSE values and bias in the forecast compared to the WRF forecast with the BMJ cumulus scheme.

Here we investigate if correcting the systematic biases in the SINTEX-F2v forecasts, such as the westward shift of cyclonic circulation over northern Australia (Fig. 1c), would improve the spatial and temporal representation of the WRF_KF_NBC forecast precipitation. We carried out an ensemble of WRF Model runs with the bias-corrected SINTEX-F2v forecast fields as the boundaries (WRF_KF_BC). The mean austral summer WRF_KF_BC precipitation forecast is presented in Fig. 1f. There is a clear improvement in the spatial representation of precipitation over Australia in the WRF_KF_BC forecast (Fig. 1f), compared to the WRF_KF_NBC forecast precipitation (Fig. 1e). In the WRF_KF_BC experiment, because of the removal of the systematic biases from the SINTEX-F2v forecasts, the location of the mean cyclonic circulation (Fig. 1f) is closer to the observation (Fig. 1b), resulting in an improved representation of the precipitation over Australia. The RMSE values are smaller in the WRF_KF_BC forecast precipitation (Fig. 2d) compared to the WRF_KF_NBC forecasts. Also, there is a significant reduction in the biases in the forecast precipitation over northern Australia in WRF_KF_BC (Fig. 2h) compared to the WRF_KF_NBC forecast (Fig. 2g). The reduction of bias in precipitation in WRF_KF_BC is prominent in the east and central parts of northern Australia (Fig. 2h) compared to the WRF_KF_NBC forecast precipitation. Compared to SINTEX-F2v forecast precipitation, WRF_KF_NBC and WRF_KF_BC forecasts have a significant improvement in the precipitation over northern Australia, showing the added value of dynamical downscaling to the SINTEX-F2v forecast precipitation. Nevertheless, the systematic correction of the SINTEX-F2v forecasts did not reduce the biases along the southeast coast of Australia in the WRF_KF_BC forecast (Fig. 2h). Both WRF_KF_NBC (Fig. 2g) and WRF_KF_BC (Fig. 2h) overestimated precipitation in that region. Also, both the WRF_KF_NBC (Fig. 2g) and WRF_KF_BC (Fig. 2h) forecasts have less precipitation over western parts of northern Australia compared to observed precipitation. In comparison to the SINTEX-F2v forecast, both the WRF_KF_NBC and WRF_KF_BC forecasts have significant positive biases in precipitation over southeastern Australia with larger biases in WRF_KF_BC forecast.

The improvements to the SINTEX-F2v forecasts by dynamical downscaling, as mentioned above, encouraged us to investigate if the forecasts could be further improved by using an atmosphere–ocean coupled regional model. We performed two additional experiments using an atmosphere–ocean coupled regional model with the WRF Model as the atmospheric component and a simple mixed layer ocean model as the oceanic component. The coupled model was driven by both the SINTEX-F2v forecast (WRF_KF_NBC_OC) and the bias-corrected SINTEX-F2v forecasts (WRF_KF_BC_OC). The spatial representations of the mean austral summer precipitation in both WRF_KF_NBC_OC (Fig. 1f) and WRF_KF_BC_OC (Fig. 1h) are similar to WRF_KF_NBC (Fig. 1e) and WRF_KF_BC (Fig. 1f). However, the RMSE (Fig. 2i) and model bias (Fig. 2k) are higher in the WRF_KF_NBC_OC compared to the WRF_KF_NBC (Figs. 2c,g) forecasts. The RMSE (Fig. 2j) and bias (Fig. 2l) forecast by WRF_KF_BC_OC are comparable to those forecast by WRF_KF_BC (Figs. 2d,h). The root-mean-square difference and the difference between the WRF_KF_NBC_OC and WRF_KF_NBC forecasts (Figs. 2m,o) reveal a significant reduction in precipitation along the eastern parts of northern Australia, indicating enhancement of the negative biases in the WRF_KF_NBC forecast by the coupled model. However, the differences between the WRF_KF_BC_OC and WRF_KF_BC forecasts (Figs. 2n,p) are small, and the coupled model improves the precipitation forecasts over the western parts of northern Australia.

b. Spatial distribution of seasonal precipitation anomalies

In this section, we analyze the anomalies forecast by the models. By generating the anomalies from the model climatologies, the systematic biases in the forecasts are reduced.

The deterministic skill of forecasts is calculated by the ACC [Eq. (1)] between the model forecast anomalies and observed anomalies. The spatial distribution of ACC between the SINTEX-F2v forecasts and the BOM_AWAP precipitation indicates that the SINTEX-F2v forecasts have significant correlation along the east coast of Australia and also significant correlation is over scattered regions of southern Australia (Fig. 3a), the regions that receive significantly less rainfall during the austral summer season compared to other regions of Australia. The low skill of SINTEX-F2v in forecasting the precipitation over northern Australia can be attributed to the large biases in the forecast precipitation (Fig. 2e) over that region. Similarly, the WRF_BMJ_NBC forecast also has low skill in forecasting the precipitation over northern Australia (Fig. 3b). This also is due to large biases in the WRF_BMJ_NBC forecast precipitation (Fig. 2f) over northern Australia. On the other hand, the WRF_KF_NBC forecast has significant positive correlation coefficients along the east coast of northern Australia (Fig. 3c), the region with less bias in the forecast precipitation (Fig. 2g), indicating improved skill over the region compared to SINTEX-F2v forecasts. But, most importantly, the WRF_KF_BC forecast, in which bias-corrected SINTEX-F2v forecasts drive the WRF Model, further improves the ACC (Fig. 3d); significantly higher ACC scores, compared to that in SINTEX-F2v, are present over parts of east coast of Australia, the Cape York Peninsula, and the northern parts of northern Australia (Fig. 3d).

Fig. 3.

Significant ACC between (a) SINTEXF2v, (b) WRF_BMJ_NBC, (c) WRF_KF_NBC, (d) WRF_KF_BC, (e) WRF_KF_NBC_OC, (f) WRF_KF_BC_OC, (g) WRF_KF_NBC_CSST, and (h) WRF_KF_BC_CSST forecast precipitation anomalies and BOM_AWAP estimated anomalies. Shaded values are significant at 95% using the one-tailed Student’s t test.

Fig. 3.

Significant ACC between (a) SINTEXF2v, (b) WRF_BMJ_NBC, (c) WRF_KF_NBC, (d) WRF_KF_BC, (e) WRF_KF_NBC_OC, (f) WRF_KF_BC_OC, (g) WRF_KF_NBC_CSST, and (h) WRF_KF_BC_CSST forecast precipitation anomalies and BOM_AWAP estimated anomalies. Shaded values are significant at 95% using the one-tailed Student’s t test.

Further, it is interesting that the inclusion of air–sea interaction improves the ACC (although confined to only the eastern and northern parts of northern Australia). The WRF_KF_NBC_OC (Fig. 3e) forecast has reduced skill over the northwestern part of northern Australia compared to the WRF_KF_NBC forecast because of larger biases in the forecast precipitation (Fig. 2o). The WRF_KF_NBC_OC forecast has significant skill over the eastern part of northern Australia and over scattered regions of Australia (Fig. 3e). On the other hand, the WRF_KF_BC_OC forecast has larger ACC values over the eastern and northern parts of northern Australia (Fig. 3f) compared to WRF_KF_BC (Fig. 3d) ACC values. The ACC over the Cape York Peninsula and northern parts of northern Australia is above 0.7 in the WRF_KF_BC_OC forecast (Fig. 3f). These results suggest that it is essential to correct the systematic biases in the driving GCM forecast to improve the austral summer forecast over Australia and also that using a regional coupled model with the bias-corrected GCM forecasts can further improve the skill of the forecasts. To highlight the role of air–sea interactions more clearly, we carried out experiments in which the SST simulated by the WRF_KF_NBC_OC and WRF_KF_BC_OC experiments were specified in the WRF Model (WRF_KF_NBC_CSST and WRF_KF_BC_CSST). The anomaly correlation coefficient between the WRF_KF_NBC_CSST forecast precipitation (Fig. 3g) and BOM_AWAP anomalies has significant values similar to those shown by the WRF_KF_NBC_OC experiment (Fig. 3c), although smaller in magnitude. Comparing Figs. 3e and 3g, it is clear that the inclusion of air–sea interaction in the WRF_KF_NBC_OC experiment was responsible for the improvement in the ACC values seen over the eastern parts of northern Australia. Similar conclusions can be reached by comparing WRF_KF_BC_CSST (Fig. 3h) and WRF_KF_BC_OC (Fig. 3f) forecast ACC scores.

We now analyze the spatial distribution of the anomalies forecast by the models for certain events to bring out the importance of dynamical downscaling of SINTEX-F2v forecasts in improving the forecast anomalies. The climate of Australia is affected by various climate modes, such as ENSO, the IOD, and Ningaloo Niño. However, it may be noted here that the impact of ENSO on Australian climate reduces during the austral summer season (Hendon et al. 2012). The IOD peaks in the austral spring season and is terminated by austral summer season and hence has no impact on the austral summer precipitation. Ningaloo Niño (Feng et al. 2013, Kataoka et al. 2014; Doi et al. 2013), a recently discovered climate mode, is a coastal phenomenon and refers to anomalous warming along the west coast of Australia. The ocean–atmosphere coupled feedback associated with this phenomenon can enhance seasonal rainfall prediction skill over west Australia (Doi et al. 2015). An unprecedented strong Ningaloo Niño event was observed in the austral summer season of 2010/11 off the west coast of Australia associated with a La Niña event in the equatorial Pacific. The Ningaloo Niño event of 2010/11 was demonstrated by Feng et al. (2013) to be mostly driven by the oceanic heat transport of the poleward-flowing Leeuwin Current during the season. The event impacted the marine ecosystem and coral reef there. The event peaked in February 2011. The SINTEX-F1 forecasting system was able to forecast the event successfully (Doi et al. 2013, 2014, 2016). To analyze if the WRF Model could improve the SINTEX-F2v forecast precipitation during the event, we plotted the DJF 2010/11 precipitation anomalies forecast by both the SINTEX-F2v and WRF in Fig. 4.

Fig. 4.

(a) DJF 2010/11–averaged observed OISSTv2 anomalies (°C). (b)–(d) As in (a) but forecast by SINTEX-F2v, WRF_KF_NBC_OC, and WRF_KF_BC_OC, respectively. (e) BOM_AWAP estimated precipitation (mm day−1) anomalies averaged over DJF 2010/11. (f)–(j) As in (e), but for SINTEX-F2v, WRF_KF_NBC, WRF_KF_BC, WRF_KF_NBC_OC, and WRF_KF_BC_OC forecast precipitation (mm day−1) anomalies, respectively. (k) Difference between SINTEX-F2v and BOM_AWAP DJF 2010/11 precipitation (mm day−1) anomalies. (l) As in (k), but for the difference between WRF_KF_BC and BOM_AWAP. (m) Difference between WRF_KF_BC and SINTEX-F2v forecast precipitation anomalies (mm day−1).

Fig. 4.

(a) DJF 2010/11–averaged observed OISSTv2 anomalies (°C). (b)–(d) As in (a) but forecast by SINTEX-F2v, WRF_KF_NBC_OC, and WRF_KF_BC_OC, respectively. (e) BOM_AWAP estimated precipitation (mm day−1) anomalies averaged over DJF 2010/11. (f)–(j) As in (e), but for SINTEX-F2v, WRF_KF_NBC, WRF_KF_BC, WRF_KF_NBC_OC, and WRF_KF_BC_OC forecast precipitation (mm day−1) anomalies, respectively. (k) Difference between SINTEX-F2v and BOM_AWAP DJF 2010/11 precipitation (mm day−1) anomalies. (l) As in (k), but for the difference between WRF_KF_BC and BOM_AWAP. (m) Difference between WRF_KF_BC and SINTEX-F2v forecast precipitation anomalies (mm day−1).

During the Ningaloo Niño event of DJF 2010/11, the National Oceanic and Atmospheric Administration (NOAA) Optimum Interpolation SST, version 2 (OISSTv2; Reynolds et al. 2002), anomalies indicate anomalous warming along the west coast of Australia (Fig. 4a). Anomalies exceeding 3°C are observed along the west coast of Australia during the event (Fig. 4a). The 1-month-lead forecast of the SINTEX-F2v model could capture realistically both the magnitude and spatial extent of the warm SST anomalies along the west coast of Australia (Fig. 4b). Interestingly, the WRF_KF_NBC_OC (Fig. 4c) and WRF_KF_BC_OC (Fig. 4d) forecasts could also reproduce the warm SST anomalies along the west coast of Australia during DJF 2010/11, although weaker in magnitude compared to OISSTv2 anomalies. The WRF_KF_NBC_OC (Fig. 4c) and WRF_KF_BC_OC (Fig. 4d) forecasts could capture warm SST anomalies along the west coast of Australia because the Ningaloo Niño event signal present in the SINTEX-F2v forecast was used to drive the WRF Model. The poleward transport of the heat by the Leeuwin Current, which was suggested to be the main driver of the Ningaloo Niño event of DJF 2010/11 (Feng et al. 2013), is not taken into account in the regional coupled model forecasts because of the absence of the advection term in the mixed layer formulation. The absence of the advection term resulted in weak SST anomalies along the west coast of Australia in the WRF_KF_NBC_OC and WRF_KF_BC_OC forecasts.

Australia received above-normal rainfall in the austral summer season of DJF 2010/11, associated with the Ningaloo Niño and La Niña events (Fig. 4e). Cyclonic anomalies are observed over Australia during DJF 2010/11 (Fig. 4e; streamlines). The SINTEX-F2v forecast (Fig. 4f), even though it could capture the SST anomalies along the west coast of Australia during DJF 2010/11, has bias in forecasting the precipitation anomalies along the west coast of Australia. The SINTEX-F2v forecast underestimated the precipitation by more than 4 mm day−1 along the west coast of Australia (Fig. 4k) and overestimated the precipitation over northern Australia (Fig. 4k). The northerly wind anomalies (Fig. 4e) during DJF 2010/11 transport moisture from the warm tropical regions (Fig. 4a) into the western regions of Australia resulting in positive precipitation anomalies (Fig. 4e) observed over the region. Unlike in the observations, the 850-hPa wind anomalies in the SINTEX-F2v forecast are northwesterly (Fig. 4f) and hence transport moisture into northern Australia, resulting in increase of precipitation over northern Australia (Fig. 4f). Thus, the biases in the forecast wind anomalies resulted in the biases in the SINTEX-F2v forecast precipitation during the DJF 2010/11 season.

The WRF_KF_NBC (Fig. 4g), WRF_KF_BC (Fig. 4h), WRF_KF_NBC_OC (Fig. 4i), and WRF_KF_BC_OC (Fig. 4j) forecasts also have positive precipitation biases over northern Australia associated with the low-level cyclonic wind anomalies. For example, the difference between the WRF_KF_BC forecast and BOM_AWAP observed precipitation anomalies (Fig. 4l) indicates positive biases over northern Australia. Comparing the WRF_KF_BC and SINTEX-F2v forecast precipitation anomalies (Fig. 4m), it is evident that the WRF Model forecast has smaller biases over the west coast of Australia compared to the SINTEX-F2v forecast. This brings out the importance of dynamically downscaling the SINTEX-F2v forecasts. However, the spatial representations of precipitation anomalies in the WRF_KF_BC_OC and WRF_KF_BC forecasts of the DJF 2010/11 season are similar, suggesting that using a regional coupled model does not always lead to improvements in the precipitation forecasts.

c. 2-m air temperature

The 2-m air temperature is one of the parameters, other than precipitation, often evaluated to determine the skill of model forecasts. In this section, we present the analysis of the SINTEX-F2v and WRF forecast 2-m air temperature. GHCN+CAMS-estimated mean 2-m air temperature over Australia, averaged over the period DJF 2000/01–DJF 2014/14, has values greater than 32°C over western parts of Australia, and values ranging from 28° to 32°C are observed over most parts of northern Australia (Fig. 5a). The SINTEX-F2v model forecast has significant bias, in excess of 4°C, over most parts of Australia (Fig. 5b; shaded). Interestingly, the 2-m temperature forecast by the WRF Model (Figs. 5c–g) has improved 2-m air temperatures compared to the SINTEX-F2v forecasts (Fig. 5b). WRF_BMJ_NBC (Fig. 5c), WRF_KF_NBC (Fig. 5d), and WRF_KF_NBC_OC (Fig. 5f) forecasts have a significant positive bias of around 1°–2°C over northern Australia and a negative bias of 1°–2°C over central and the east coast of Australia. Large improvement is evident in the WRF_KF_BC (Fig. 5e) and WRF_KF_BC_OC (Fig. 5g) 2-m air temperature forecasts compared to SINTEX-F2v (Fig. 5b) and other WRF forecasts (Figs. 5c,d,f). WRF_KF_BC (Fig. 5e) and WRF_KF_BC_OC (Fig. 5g) forecasts have a positive bias of 1°C over northern Australia and a negative bias over other parts of Australia. These results clearly demonstrate the importance of bias correcting the driving SINTEX-F2v austral summer forecasts before dynamical downscaling using the WRF Model over Australia to improve the skill of the 2-m air temperature forecasts.

Fig. 5.

(a) GHCN+CAMS-estimated DJF mean (averaged over DJF 2010/11–DJF 2013/14) 2-m air temperature (°C). Significant bias (at 95% using the two-tailed Student’s t test; shaded) and mean 2-m air temperatures (contours) forecast by (b) SINTEX-F2v, (c) WRF_BMJ_NBC, (d) WRF_KF_NBC, (e) WRF_KF_BC, (f) WRF_KF_NBC_OC, and (g) WRF_KF_BC_OC. The significant ACC (at 95% using the one-tailed Student’s t test) between (h) SINTEX-F2v, (i) WRF_BMJ_NBC, (j) WRF_KF_NBC, (k) WRF_KF_BC, (l) WRF_KF_NBC_OC, and (m) WRF_KF_NBC_OC forecast anomalies and GHCN+CAMS-estimated 2-m air temperature anomalies.

Fig. 5.

(a) GHCN+CAMS-estimated DJF mean (averaged over DJF 2010/11–DJF 2013/14) 2-m air temperature (°C). Significant bias (at 95% using the two-tailed Student’s t test; shaded) and mean 2-m air temperatures (contours) forecast by (b) SINTEX-F2v, (c) WRF_BMJ_NBC, (d) WRF_KF_NBC, (e) WRF_KF_BC, (f) WRF_KF_NBC_OC, and (g) WRF_KF_BC_OC. The significant ACC (at 95% using the one-tailed Student’s t test) between (h) SINTEX-F2v, (i) WRF_BMJ_NBC, (j) WRF_KF_NBC, (k) WRF_KF_BC, (l) WRF_KF_NBC_OC, and (m) WRF_KF_NBC_OC forecast anomalies and GHCN+CAMS-estimated 2-m air temperature anomalies.

The deterministic skill of the 2-m air temperature forecast is determined by calculating the ACC as given in Eq. (1). The ACC between the austral summer SINTEX-F2v forecast and GHCN+CAMS DJF mean estimate shows the SINTEX-F2v forecast to have significant skill confined to a small region over northern Australia (Fig. 5h). In fact, even the WRF_BMJ_NBC (Fig. 5i) and WRF_KF_NBC (Fig. 5j) forecasts have no significant skill in forecasting the 2-m air temperature anomalies. WRF_KF_NBC_OC (Fig. 5l) has skill in forecasting 2-m air temperatures over a small region of east northern Australia. However, the WRF_KF_BC (Fig. 5k) and WRF_KF_BC_OC (Fig. 5m) forecasts, in which the WRF Model was driven by the bias-corrected SINTEX-F2v forecasts, have significant skill in forecasting 2-m air temperatures over scattered regions of northern Australia, whereas SINTEX-F2v and other WRF forecasts have no significant ACC scores over northern Australia.

d. Discussion

The above analysis highlights the importance of choosing a suitable convective scheme and also bias correcting the driving GCM output prior to driving the regional model for improving the austral summer forecasts over Australia by an RCM. The RCM driven by bias-corrected GCM output adds value to the GCM precipitation forecast that can be further improved by coupling the RCM to an ocean model. In this section, we will discuss sensitivity of the forecasts to the initialization of the land surface model and the mixed layer ocean model.

Australian climate variability is strongly related to soil moisture variability (Jones and Trewin 2000; Timbal et al. 2002), and proper initialization of the land surface model is found to be beneficial for improving the subseasonal forecast skills (Hirsch et al. 2014) over Australia. As mentioned previously, the unified Noah land surface scheme in the WRF Model was initialized using climatological ERA-Interim soil moisture and soil temperature because of the lack of availability of these parameters in the SINTEX-F2v forecast. Here, we investigate if the WRF Model forecast precipitation and 2-m air temperatures would improve if the model is initialized with the ERA-Interim estimated soil moisture and soil temperatures (WRF_KF_BC_RSMST) corresponding to the WRF Model start date instead of using climatological ERA-Interim soil moisture and temperature. The WRF_KF_BC_RSMST forecast precipitation has biases (Fig. 6a), similar to the biases in the WRF_KF_BC (Fig. 2h) forecast. The WRF_KF_BC_RSMST (Fig. 6b) precipitation forecast also has an ACC score comparable to the WRF_KF_BC forecast (Fig. 3d), with significant skill near the east coast of northern Australia and near the west coast of Australia. Similarly, comparing the biases in the forecast 2-m air temperatures of WRF_KF_BC_RSMST (Fig. 6c) and WRF_KF_BC (Fig. 4e) and the anomaly correction coefficients of the WRF_KF_BC_RSMST (Fig. 6d) and WRF_KF_BC (Fig. 4k) forecast 2-m air temperature anomalies, we find that the model results are not significantly different between the two experiments at seasonal time scales. Therefore, the results from the initialization of real-time observed ERA-Interim soil moisture do not add significant value to the predictability of precipitation and 2-m air temperature forecasts under the present model setup at seasonal time scales. Further studies are needed to understand the role of soil moisture under different time and space scales, which is beyond the scope of the present research.

Fig. 6.

(a) Significant bias (at 95% using the two-tailed Student’s t test) in WRF_KF_BC_RSMST forecast precipitation (mm day−1). (b) Significant ACC (at 95% using the one-tailed Student’s t test) between WRF_KF_BC_RSMST forecast precipitation and BOM-AWAP precipitation. (c) As in (a), but for bias in the forecast 2-m air temperature (°C); (d) as in (b), but for ACC between WRF_KF_BC_RSMST forecast and GHCN+CAMS 2-m air temperature. (e)–(h) As in (a)–(d), but for the WRF_KF_BC_OC100 experiment.

Fig. 6.

(a) Significant bias (at 95% using the two-tailed Student’s t test) in WRF_KF_BC_RSMST forecast precipitation (mm day−1). (b) Significant ACC (at 95% using the one-tailed Student’s t test) between WRF_KF_BC_RSMST forecast precipitation and BOM-AWAP precipitation. (c) As in (a), but for bias in the forecast 2-m air temperature (°C); (d) as in (b), but for ACC between WRF_KF_BC_RSMST forecast and GHCN+CAMS 2-m air temperature. (e)–(h) As in (a)–(d), but for the WRF_KF_BC_OC100 experiment.

The ocean mixed layer model used in the coupled regional model was initialized by specifying a mixed layer depth of 50 m. We carried out experiments to verify the sensitivity of the model forecasts to the specification of initial depth by initializing the model with a depth of 100 m (WRF_KF_BC_OC100) and driving the model with similar bias-corrected SINTEX-F2v forecasts. The results of the experiment are shown in Figs. 6a–d. Comparing the WRF_KF_BC_OC100 forecasts of precipitation and 2-m air temperature (Figs. 6a–d) with WRF_KF_BC_OC forecasts (Figs. 2l, 3f, and 4g,m), it is evident that the specification of initial depth has no large impact on the forecasts of precipitation and 2-m air temperatures at seasonal time scales. The mixed layer depth in the mixed layer ocean model in WRF is a prognostic variable, depending on heat fluxes and winds for its prediction. During the model integration, the initially specified mixed layer depth adjusts over the seasonal time scale and hence has no large impact on the seasonal forecasts of precipitation and 2-m air temperature.

4. Conclusions

In this study, we made an effort to improve the austral summer seasonal forecasts over Australia by dynamically downscaling 1-month-lead SINTEX-F2v CGCM forecasts using the WRF Model. The SINTEX-F2v forecasts underestimated the precipitation over northern Australia (Fig. 1c) because of a systematic bias concerning westward displacement of 850-hPa winds (Fig. 1c). The SINTEX-F2v forecasts were downscaled using a regional model based on WRF. Two model configurations (WRF_BMJ_NBC and WRF_KF_NBC) using two different convective schemes (viz., BMJ and KF) were tested. Analysis of the WRF Model forecasts revealed that the BMJ scheme underestimated the mean rainfall and also showed large root-mean-square errors compared to WRF forecasts with the KF scheme over northern Australia.

The systematic biases present in the SINTEX-F2v are transmitted to the downscaled WRF_KF_NBC precipitation. An attempt was made to correct this bias in experiment WRF_KF_BC, in which the mean biases present in the SINTEX-F2v forecast were corrected. This is done by replacing the SINTEX-F2v 6-hourly climatology with the corresponding climatology derived from ERA-Interim while allowing the anomaly part of SINTEX-F2v to freely evolve in the downscaling experiments. The results exhibited a dramatic improvement in the pattern and skill of precipitation forecasts.

As the oceans around tropical northern Australia drive the atmosphere, we used a regional coupled model, with the WRF Model as the atmospheric component and a simple mixed layer model as the oceanic component. We carried out two experiments: one by driving the coupled WRF Model with SINTEX-F2v forecasts and the other with mean bias-corrected SINTEX-F2v forecasts. Anomaly correlation coefficients showed improvements over northeastern Australia due to the coupling. The WRF_KF_BC_OC experiment, the second type of coupled downscaled experiment, is seen to have an improved skill compared to SINTEX-F2v and other WRF experiments. The 2-m air temperature forecasts also have similar results. The large biases in the SINTEX-F2v 2-m air temperatures are reduced in the WRF downscaled model forecasts. The reduction of biases is seen to be highest in the WRF_KF_BC and WRF_KF_BC_OC experiments. To see if the improvements in the ACC of the coupled model were due to inclusion of air–sea interaction, we conducted one experiment (WRF_KF_BC_CSST), where we specified the SST forecast from the WRF_KF_BC_OC experiment. The results showed that the WRF_KF_BC_CSST experiment has lower ACC compared to the WRF_KF_BC_OC experiment. This indicates the importance of air–sea interactions in forecasting the precipitation and temperature, particularly over coastal parts of Australia during the austral summer season. The results in this study highlight the importance of correcting biases in the driving GCM forecasts before using those in a regional climate model at seasonal time scales.

The model experiments with the WRF Model showed that the dynamical downscaling adds value to the SINTEX-F2v forecasts, which can be beneficial for agriculture and other social activities of the regions. This study showed the importance of including air–sea interactions in improving the austral summer forecasts over Australia. This encourages us to use a fully coupled regional model (Ratnam et al. 2015) to develop high-resolution regional forecasts over Australia in the future.

Acknowledgments

The authors thank the three anonymous reviewers, whose comments improved the manuscript substantially. This research is supported by Environment Research and Technology Development fund (2-1405) of the Ministry of Environment, Japan, and JSPS KAKENHI Grants 16H04047 and 16K17810. The SINTEX-F2 seasonal climate prediction systems were run by the Earth Simulator at JAMSTEC. We are grateful to Drs. Wataru Sasaki, Jing-Jia Luo, Sebastian Masson, and our European colleagues at INGV–CMCC, LOCEAN, and MPI for their contribution to developing the prototype of the systems.

REFERENCES

REFERENCES
Allan
,
R.
,
1988
:
El Niño–Southern Oscillation influences in the Australasian region
.
Prog. Phys. Geogr.
,
12
,
313
348
, doi:.
Betts
,
A. K.
, and
M. J.
Miller
,
1986
:
A new convective adjustment scheme. Part II: Single column tests using GATE wave, BOMEX, and arctic air-mass data sets
.
Quart. J. Roy. Meteor. Soc.
,
112
,
693
709
, doi:.
Bruyère
,
C. L.
,
J. M.
Done
,
G. J.
Holland
, and
S.
Fredrick
,
2014
:
Bias corrections of global models for regional climate simulations of high-impact weather
.
Climate Dyn.
,
43
,
1847
1856
, doi:.
Christensen
,
O. B.
,
J. H.
Christensen
,
B.
Machenhauer
, and
M.
Botzet
,
1998
:
Very high-resolution regional climate simulations over Scandinavia—Present climate
.
J. Climate
,
11
,
3204
3229
, doi:.
Davis
,
C.
, and Coauthors
,
2008
:
Prediction of landfalling hurricanes with the advanced hurricane WRF Model
.
Mon. Wea. Rev.
,
136
,
1990
2005
, doi:.
de Boyer Montégut
,
C.
,
G.
Madec
,
A. S.
Fischer
,
A.
Lazar
, and
D.
Iudicone
,
2004
:
Mixed layer depth over the global ocean: An examination of profile data and a profile-based climatology
.
J. Geophys. Res.
,
109
,
C12003
, doi:.
Dee
,
D. P.
, and Coauthors
,
2011
:
The ERA-Interim reanalysis: Configuration and performance of the data assimilation system
.
Quart. J. Roy. Meteor. Soc.
,
137
,
553
597
, doi:.
Dickinson
,
R. E.
,
R. M.
Errico
,
F.
Giorgi
, and
G. T.
Bates
,
1989
:
A regional climate model for the western United States
.
Climatic Change
,
15
,
383
422
, doi:.
Doi
,
T.
,
S. K.
Behera
, and
T.
Yamagata
,
2013
:
Predictability of the Ningaloo Niño/Niña
.
Sci. Rep.
,
3
,
2892
, doi:.
Doi
,
T.
, and Coauthors
,
2014
:
Study on the predictability of climate variations and their mechanisms. Annual Rep. of the Earth Simulator Center April 2013–March 2014, Center for Earth Information Science and Technology, 49–54. [Available online at http://www.jamstec.go.jp/ceist/j/publication/annual/annual2013/pdf/2project/chapter1/49masumoto.pdf.]
Doi
,
T.
,
S. K.
Behera
, and
T.
Yamagata
,
2015
:
An interdecadal regime shift in rainfall predictability related to the Ningaloo Niño in the late 1990s
.
J. Geophys. Res. Oceans
,
120
,
1388
1396
, doi:.
Doi
,
T.
,
S. K.
Behera
, and
T.
Yamagata
,
2016
:
Improved seasonal prediction using the SINTEX-F2 coupled model
.
J. Adv. Model. Earth Syst.
,
8
,
1847
1867
, doi:.
Dudhia
,
J.
,
1989
:
Numerical study of convection observed during the winter monsoon experiment using a mesoscale two-dimensional model
.
J. Atmos. Sci.
,
46
,
3077
3107
, doi:.
Evans
,
J. P.
, and
M. F.
McCabe
,
2010
:
Regional climate simulation over Australia’s Murray-Darling basin: A multitemporal assessment
.
J. Geophys. Res.
,
115
,
D14114
, doi:.
Evans
,
J. P.
,
M.
Ekström
, and
F.
Ji
,
2012
:
Evaluating the performance of a WRF physics ensemble over south-east Australia
.
Climate Dyn.
,
39
,
1241
1258
, doi:.
Fan
,
Y.
, and
H.
van den Dool
,
2008
:
A global monthly land surface air temperature analysis for 1948–present
.
J. Geophys. Res.
,
113
,
D01103
, doi:.
Feng
,
M.
,
M. J.
McPhaden
,
S.
Xie
, and
J.
Hafner
,
2013
:
La Niña forces unprecedented Leeuwin Current warming in 2011
.
Sci. Rep.
,
3
,
1277
, doi:.
Giorgi
,
F.
, and
G. T.
Bates
,
1989
:
The climatological skill of a regional model over complex terrain
.
Mon. Wea. Rev.
,
117
,
2325
2347
, doi:.
Hendon
,
H. H.
,
E.-P.
Lim
, and
M. C.
Wheeler
,
2011
:
Seasonal prediction of Australian summer monsoon rainfall. The Global Monsoon System, 2nd ed. C. P. Chang et al., Eds., World Scientific, 73–83
.
Hendon
,
H. H.
,
E.-P.
Lim
, and
G.
Liu
,
2012
:
The role of air–sea interaction for prediction of Australian summer monsoon rainfall
.
J. Climate
,
25
,
1278
1290
, doi:.
Hirsch
,
A. L.
,
J.
Kala
,
A. J.
Pitman
,
C.
Carouge
,
J. P.
Evans
,
V.
Haverd
, and
D.
Mocko
,
2014
:
Impact of land surface initialization approach on subseasonal forecast skill: A regional analysis in the southern hemisphere
.
J. Hydrometeor.
,
15
,
300
319
, doi:.
Holland
,
G. J.
,
J. M.
Done
,
C. L.
Bruyere
,
C.
Cooper
, and
A.
Suzuki
,
2010
:
Model investigations of the effects of climate variability and change on future Gulf of Mexico tropical cyclone activity. Proc. Offshore Technology Conf., Houston, TX, Offshore Technology Conf., OTC 20690
, doi:.
Hong
,
S. Y.
,
J.
Dudhia
, and
S. H.
Chen
,
2004
:
A revised approach to ice microphysical processes for bulk parameterization of cloud and precipitation
.
Mon. Wea. Rev.
,
132
,
103
120
, doi:.
Hong
,
S. Y.
,
Y.
Noh
, and
J.
Dudhia
,
2006
:
A new vertical diffusion package with an explicit treatment of entrainment processes
.
Mon. Wea. Rev.
,
134
,
2318
2341
, doi:.
Janjić
,
Z. I.
,
1994
:
The step-mountain eta coordinate model: Further developments of the convection, viscous sublayer, and turbulence closure schemes
.
Mon. Wea. Rev.
,
122
,
927
945
, doi:.
Jones
,
D. A.
, and
B. C.
Trewin
,
2000
:
On the relationships between the El Niño–Southern Oscillation and Australian land surface temperature
.
Int. J. Climatol.
,
20
,
697
719
, doi:.
Jones
,
D. A.
,
W.
Wang
, and
R.
Faucett
,
2009
:
High-quality spatial climate data-sets for Australia
.
Aust. Meteor. Oceanogr. J.
,
58
,
233
248
.
Kain
,
J. S.
,
2004
:
The Kain–Fritsch convective parameterization: An update
.
J. Appl. Meteor.
,
43
,
170
181
, doi:.
Kala
,
J.
,
J.
Andrys
,
T. J.
Lyons
,
I. J.
Foster
, and
B. J.
Evans
,
2015
:
Sensitivity of WRF to driving data and physics options on a seasonal time-scale for the southwest of Western Australia
.
Climate Dyn.
,
44
,
633
659
, doi:.
Kataoka
,
T.
,
T.
Tozuka
,
S. K.
Behera
, and
T.
Yamagata
,
2014
:
On the Ningaloo Niño/Niña
.
Climate Dyn.
,
43
,
1463
1482
, doi:.
Kim
,
E. J.
, and
S. Y.
Hong
,
2010
:
Impact of air–sea interaction on East Asian summer monsoon climate in WRF
.
J. Geophys. Res.
,
115
,
D19118
, doi:.
King
,
A. D.
,
N. P.
Klingaman
,
L. V.
Alexander
,
M. G.
Donat
,
N. C.
Jourdain
, and
P.
Maher
,
2014
:
Extreme rainfall variability in Australia: Patterns, drivers, and predictability
.
J. Climate
,
27
,
6035
6050
, doi:.
Krishna Kumar
,
K.
,
M.
Hoerling
, and
B.
Rajagopalan
,
2005
:
Advancing dynamical prediction of Indian monsoon rainfall
.
Geophys. Res. Lett.
,
32
,
L08704
, doi:.
Luo
,
J.-J.
,
S.
Masson
,
E.
Roeckner
,
G.
Madec
, and
T.
Yamagata
,
2005
:
Reducing climatology bias in an ocean–atmosphere CGCM with improved coupling physics
.
J. Climate
,
18
,
2344
2360
, doi:.
Luo
,
J.-J.
,
S.
Masson
,
S.
Behera
, and
T.
Yamagata
,
2007
:
Experimental forecasts of the Indian Ocean dipole using a coupled OAGCM
.
J. Climate
,
20
,
2178
2190
, doi:.
Luo
,
J.-J.
,
S.
Masson
,
S.
Behera
, and
T.
Yamagata
,
2008
:
Extended ENSO predictions using a fully coupled ocean–atmosphere model
.
J. Climate
,
21
,
84
93
, doi:.
Madec
,
G.
, and Coauthors
,
2008
:
NEMO ocean engine, version 3.0. Note du Pôle de Modélisation de l’Institut Pierre-Simon Laplace 27, 217 pp
.
Masson
,
S.
,
P.
Terray
,
G.
Madec
,
J. J.
Luo
,
T.
Yamagata
, and
K.
Takahashi
,
2012
:
Impact of intra-daily SST variability on ENSO characteristics in a coupled model
.
Climate Dyn.
,
39
,
681
707
, doi:.
McBride
,
J. L.
, and
N.
Nicholls
,
1983
:
Seasonal relationships between Australian rainfall and Southern Oscillation
.
Mon. Wea. Rev.
,
111
,
1998
2004
, doi:.
McGregor
,
J. L.
,
1997
:
Regional climate modeling
.
Meteor. Atmos. Phys.
,
63
,
105
117
, doi:.
McGregor
,
J. L.
, and
K.
Walsh
,
1993
:
Nested simulations of perpetual January climate over the Australian region
.
J. Geophys. Res.
,
98
,
23 283
23 290
, doi:.
Misra
,
V.
, and
M.
Kanamitsu
,
2004
:
Anomaly nesting: A methodology to downscale seasonal climate simulations from AGCMs
.
J. Climate
,
17
,
3249
3262
, doi:.
Mlawer
,
E. J.
,
S. J.
Taubman
,
P. D.
Brown
,
M. J.
Iacono
, and
S. A.
Clough
,
1997
:
Radiative transfer for inhomogeneous atmospheres: RRTM, a validated correlated-k model for the long wave
.
J. Geophys. Res.
,
102
,
16 663
16 682
, doi:.
Narita
,
M.
, and
S.
Ohmori
,
2007
:
Improving precipitation forecasts by the operational nonhydrostatic mesoscale model with the Kain–Fritsch convective parameterization and cloud microphysics. 12th Conf. on Mesoscale Processes, Waterville Valley, NH, Amer. Metor. Soc., 3.7. [Available online at http://ams.confex.com/ams/pdfpapers/126017.pdf.]
Nicholls
,
N.
,
J. L.
McBride
, and
R. J.
Ormerod
,
1982
:
On predicting the onset of the Australian wet season at Darwin
.
Mon. Wea. Rev.
,
110
,
14
17
, doi:.
Peng
,
S. Q.
,
D. L.
Liu
,
Z. B.
Sun
, and
Y.
Li
,
2012
:
Recent advances in regional air–sea coupled models
.
Sci. China Earth Sci.
,
55
,
1391
1405
, doi:.
Pollard
,
R. T.
,
P. B.
Rhines
, and
R. O. R. Y.
Thompson
,
1973
:
The deepening of the wind-mixed layer
.
Geophys. Fluid Dyn.
,
3
,
381
404
.
Ratnam
,
J. V.
,
F.
Giorgi
,
A.
Kaginalkar
, and
S.
Cozzini
,
2009
:
Simulation of the Indian monsoon using the RegCM3–ROMS regional coupled model
.
Climate Dyn.
,
33
,
119
139
, doi:.
Ratnam
,
J. V.
,
S. K.
Behera
,
Y.
Masumoto
,
K.
Takahashi
, and
T.
Yamagata
,
2012
:
A simple regional coupled model experiment for summer-time climate simulation over southern Africa
.
Climate Dyn.
,
39
,
2207
2217
, doi:.
Ratnam
,
J. V.
, and Coauthors
,
2013
:
Dynamical downscaling of austral summer climate forecasts over southern Africa using a regional coupled model
.
J. Climate
,
26
,
6015
6032
, doi:.
Ratnam
,
J. V.
,
Y.
Morioka
,
S. K.
Behera
, and
T.
Yamgata
,
2015
:
A model study of regional air–sea interaction in the austral summer precipitation over southern Africa
.
J. Geophys. Res. Atmos.
,
120
,
2342
2357
, doi:.
Ratnam
,
J. V.
,
S. K.
Behera
,
T.
Doi
, and
S. B.
Ratna
,
2016
:
Improvements to the WRF seasonal hindcasts over South Africa by bias correcting the driving SINTEX-F2v CGCM fields
.
J. Climate
,
29
,
2815
2829
, doi:.
Reynolds
,
R. W.
,
N. A.
Rayner
,
T. M.
Smith
,
D. C.
Stokes
, and
W.
Wang
,
2002
:
An improved in situ and satellite SST analysis for climate
.
J. Climate
,
15
,
1609
1625
, doi:.
Risbey
,
J. S.
,
M. J.
Pook
,
P. C.
McIntosh
,
M. C.
Wheeler
, and
H. H.
Hendon
,
2009
:
On the remote drivers of rainfall variability in Australia
.
Mon. Wea. Rev.
,
137
,
3233
3253
, doi:.
Roeckner
,
E.
, and Coauthors
,
2003
:
The atmospheric general circulation model ECHAM5. Part I: Model description
.
Max-Planck-Institut für Meteorologie Tech. Rep. 349, 140 pp. [Available online at http://mms.dkrz.de/pdf/klimadaten/service_support/documents/mpi_report_349.pdf.]
Sasaki
,
W.
,
K. J.
Richards
, and
J. J.
Luo
,
2012
:
Role of vertical mixing originating from small vertical scale structures above and within the equatorial thermocline in an OGCM
.
Ocean Modell.
,
57–58
,
29
42
, doi:.
Sasaki
,
W.
,
K. J.
Richards
, and
J. J.
Luo
,
2013
:
Impact of vertical mixing induced by small vertical scale structures above and within the equatorial thermocline on the tropical Pacific in a CGCM
.
Climate Dyn.
,
41
,
443
453
, doi:.
Sasaki
,
W.
,
T.
Doi
,
K. J.
Richards
, and
Y.
Masumoto
,
2014a
:
The influence of ENSO on the equatorial Atlantic precipitation through the Walker circulation in a CGCM
.
Climate Dyn.
,
44
,
191
202
, doi:.
Sasaki
,
W.
,
T.
Doi
,
K. J.
Richards
, and
Y.
Masumoto
,
2014b
:
Impact of the equatorial Atlantic sea surface temperature on the tropical Pacific in a CGCM
.
Climate Dyn.
,
43
,
2539
2552
, doi:.
Skamarock
,
W. C.
, and Coauthors
,
2005
:
A description of the Advanced Research WRF version 2. NCAR Tech. Note NCAR/TN-468+STR, 88 pp.
, doi:.
Tewari
,
M.
, and Coauthors
,
2004
:
Implementation and verification of the unified NOAH land surface model in the WRF Model. 20th Conf. on Weather Analysis and Forecasting/16th Conf. on Numerical Weather Prediction, Seattle, WA, Amer. Meteor. Soc., 14.2a. [Available online at https://ams.confex.com/ams/84Annual/techprogram/paper_69061.htm.]
Timbal
,
B.
,
S.
Power
,
R.
Colman
,
J.
Viviand
, and
S.
Lirola
,
2002
:
Does soil moisture influence climate variability and predictability over Australia?
J. Climate
,
15
,
1230
1238
, doi:.
Valcke
,
S.
,
A.
Caubel
,
R.
Vogelsang
, and
D.
Declat
,
2004
:
OASIS3 Ocean–Atmosphere Sea Ice Soil user’s guide. CERFACS Tech. Rep. TR/CMGC/04/68, 70 pp
.
Walsh
,
K.
, and
J. L.
McGregor
,
1995
:
January and July climate simulations over the Australian region using a limited-area model
.
J. Climate
,
8
,
2387
2403
, doi:.
Wheeler
,
M. C.
, and
J. L.
McBride
,
2005
:
Australian–Indonesian monsoon
.
Intraseasonal Variability in the Atmosphere–Ocean Climate System, W. K. M. Lau and D. E. Waliser, Eds., Springer Praxis Books, 125–173
, doi:.
White
,
R. H.
, and
R.
Toumi
,
2013
:
The limitations of bias correcting regional climate model inputs
.
Geophys. Res. Lett.
,
40
,
2907
2912
, doi:.
Xu
,
Z.
, and
Z.-L.
Yang
,
2012
:
An improved dynamical downscaling method with GCM bias corrections and its validation with 30 years of climate simulations
.
J. Climate
,
25
,
6271
6286
, doi:.

Footnotes

© 2017 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).