Evaluation of High-Resolution WRF Model Simulations of Surface ...

22 downloads 0 Views 885KB Size Report
Sep 16, 2014 - dicting the cyclone's track, and the intensity of cyclones in terms of central .... The Taylor diagram for the meridional wind (Fig. 3b) shows that ...
ATMOSPHERIC AND OCEANIC SCIENCE LETTERS, 2014, VOL. 7, NO. 5, 458463

Evaluation of High-Resolution WRF Model Simulations of Surface Wind over the West Coast of India S. VISHNU and P. A. FRANCIS Indian National Centre for Ocean Information Services, Ministry of Earth Sciences, Hyderabad-500090, India Received 6 February 2014; revised 4 March 2014; accepted 4 March 2014; published 16 September 2014

Abstract This paper presents results from a statistical validation of the hindcasts of surface wind by a high-resolution-mesoscale atmospheric numerical model Advanced Research WRF (ARW3.3), which is set up to force the operational coastal ocean forecast system at Indian National Centre for Ocean Information Services (INCOIS). Evaluation is carried out based on comparisons of day-3 forecasts of surface wind with in situ and remote-sensing data. The results show that the model predicts the surface wind fields fairly accurately over the west coast of India, with high skill in predicting the surface wind during the pre-monsoon season. The model predicts the diurnal variability of the surface wind with reasonable accuracy. The model simulates the land-sea breeze cycle in the coastal region realistically, which is very clearly observed during the northeast monsoon and pre-monsoon season and is less prominent during the southwest monsoon season.  Keywords: WRF, Arabian sea, surface wind field, validation, land-sea breeze Citation: Vishnu, S., and P. A. Francis, 2014: Evaluation of high-resolution WRF model simulations of surface wind over the west coast of India, Atmos. Oceanic Sci. Lett., 7, 458–463, doi:10.3878/j.issn.1674-2834.14.0009.

1

Introduction

Accurate prediction of surface meteorological parameters is essential for the successful implementation of an operational oceanographic service, as these forecasts generated by atmospheric general circulation models are routinely used for forcing ocean general circulation models to make short-term predictions of the state of the ocean. Despite being one of the key factors influencing the quality of ocean predictions, achieving accurate forecasts of surface meteorological variables remains one of the main challenges in numerical weather prediction. Recent studies by Colle et al. (1999) and Davis et al. (1999) show that high-resolution mesoscale models exhibit considerable skill in predicting surface meteorological processes, which are often missed or not resolved by coarse-resolution atmospheric models. The ability of high-resolution models to resolve mesoscale features can be one of the reasons for the better prediction of meteorological parameters. Mesoscale models are widely used for the simulation and prediction of atmospheric parameters including surface wind. Wang et al. (2013) studied the Weather ReCorresponding author: S. VISHNU, [email protected]

search and Forecasting (WRF) model’s skill in predicting the wind speed 24 hours in advance. The study showed that the predicted wind speed is greater than the actual wind speed and the WRF model forecasting differs significantly from location to location and season to season. Das et al. (2008) studied the skill of various mesoscale models in predicting the surface meteorological fields during the monsoon season over India. The study showed that, even though the mesoscale models have considerable skill in predicting the surface meteorological fields, their ability to predict the amplitude of the wind field is especially sensitive with respect to the region as well as the model used. A comparative study by Sousounis et al. (2004) on the performance of the WRF, Fifth-Generation Penn State/NCAR Mesoscale Model (MM5), Rapid Update Cycle (RUC), and ETA (ETA derives from the Greek letter η (eta) which denotes the vertical coordinate) models for a heavy precipitation event suggested that the WRF model is better at predicting intense rainfall events compared to other mesoscale models. Deb et al. (2008) evaluated the WRF model prediction of high rainfall events over Ahmadabad, India, and found that the model is reasonably good at capturing the large-scale circulation and moisture fields, but the simulated precipitation is underestimated. An assessment of the WRF model to skillfully predict several severe cyclones over the Bay of Bengal showed that the model is reasonably good at predicting the cyclone’s track, and the intensity of cyclones in terms of central pressure, maximum sustained winds, and precipitation (Raju et al., 2012). Extreme events, such as intense rainfall and tropical cyclones are also accompanied by potentially destructive extreme wind gusts. The models that can predict these extreme weather conditions may also simulate the wind with good accuracy. All these studies highlight that the WRF model is relatively good at predicting the atmospheric variables over India, even during extreme weather conditions. Being a country with a very large population living along the coastal region and dependent on the surrounding ocean in several ways, operational oceanographic services are critical for the socioeconomic development of India. Recognizing this, India set up an operational ocean forecast system called the Indian Ocean Forecast System (INDOFOS) in early 2010 (Francis et al., 2013), which can provides basin-wide ocean predictions with a lead time of up to five days. This system is now being further enhanced to incorporate high-resolution coastal prediction systems, which are expected to provide operational ocean

NO. 5

VISHNU AND FRANCIS: HIGH-RESOLUTION WRF MODEL SIMULATIONS OF SURFACE WIND

459

predictions at a resolution of approximately 2.5 km × 2.5 km. As one of the most important requirements for providing high-resolution operational ocean forecasts is high-resolution atmospheric forcing, an attempt is now being made to provide these forecasts using an appropriate mesoscale atmospheric model. In this study, we assess the performance of a high-resolution setup of the WRF model by validating the surface wind simulations over the west coast of India and the eastern Arabian Sea with observed data.

2

Materials and methods

2.1

Model configuration

In this study, we use a high-resolution mesoscale numerical model WRF ARW3.3.1 (Skamarock et al., 2008), which has several options for the physical parameterizations. The selection of a suitable set of physical parameterizations needs to be based on the region, scale, and application of interest, and hence it requires several experiments to finalize this. We carried out several such experiments by varying the parameterization schemes for clouds and radiation. The present setup of the model has the parameterization scheme which we found to be most suitable for the region of interest. The physical parameterization options chosen for the present setup of the model are given in Table 1. 2.2

Study area

The study area comprises the region (14–26°N, 66– 75°E), in which the model has a spatial resolution of 3 km × 3 km (Fig. 1). The boundary conditions for this domain are taken from the lower resolution (9 km × 9 km) model that covers the region (8–30°N, 60–80°E), through a two-way online nesting. This setup is further nested (again, online two-way nesting) in another WRF setup with a spatial resolution of 27 km × 27 km, with a spatial extend of (0–35°N, 30–120°E). The special features of the study region are that it has a complicated topography structure known as the Western Ghats and a strong seasonal variation in the circulation owing to the Indian monsoon. The cumulus parameterization scheme is switched off for the model domain with the highest spatial resolution, because the horizontal resolution is fine enough to explicitly Table 1 Different parameterization options chosen in the configuration of the model. Parameterization scheme

Description

Cumulus paramererization

Kain-Fritsch (Kain, 2004)

Microphysics

WSM3 (Hong et al., 2004)

Planetary boundary layer

YSU (Hong et al., 2006)

Longwave radiation

RRTM (Mlawer et al., 1997)

Shortwave radiation

Dudhia (Dhudia, 1989)

Surface layer

MM5 (Hong et al., 2006)

Land surface model

Noah (Chen and Dhuadia, 2001)

Note: WSM3 is WRF Single-Moment 3 class Microphysics scheme, YSU is Yonsei University scheme. RRTM is rapid and radiative transfer model, and MM5 is fifth-generation penn state/NCAR mesoscale model.

Figure 1 Domain and topography (units: m) of the WRF model setup. Location of buoys considered in this study are marked as DS1 and SW1.

resolve cumulus convection (Skamarock et al., 2008). 2.3

Data and methodology

The model is integrated in the hindcast mode with initial and boundary conditions from the Global Forecast System (GFS) of the National Centers for Environmental Prediction (NCEP) (Sela, 2009) for the period 1 February 2012 to 31 January 2013. The initial condition of the model is updated daily and run for the next five days. As the representation of coastlines and orography greatly depends on the spatial resolution of topography data, US Geological Survey (USGS) topography and land use category data (downloaded from www.mmm.ucar.edu/ wrf/src/wps_files), which has a high spatial resolution of 30” (~ 925 m) is used to set up the model. The hindcast wind field at a height of 10 m is validated using observations from Advanced Scatterometer (ASCAT) (Verspeek et al., 2013) as well as two moored buoys at (15°N, 69°E) (DS1) and (20.28°N, 71.88°E) (SW1) in the eastern Arabian sea (Premkumar et al., 2000), deployed by the National Institute of Ocean Technology (NIOT). The data from DS1 are available for a continuous period of 11 months (1 February 2012 to 31 December 2012) and the data from SW1 are available for a continuous period of 75 days (1 April 2012 to 16 June 2012). The wind measured by the buoys is at a height of 3 m, and hence it is computed to a 10 m height using the formula U (10) = U  h 

ln(10 / Z 0 ) , ln(h / Z 0 )

where U(h) denotes the observed wind speed (m s−1) at a height of h (m), U(10) is the estimated wind speed (m s−1) at the 10 m height, and Z0 is the roughness length (m). A relatively skillful model should be able to accurately simulate both the amplitude (standard deviation) and pat-

460

ATMOSPHERIC AND OCEANIC SCIENCE LETTERS

tern of variability (correlation). A Taylor diagram (Taylor, 2001) is a very good tool to illustrate the model prediction ability by representing the statistical parameters, such as model correlation, normalized root-mean-square error (RMSE), ratio of variances between model, and observations, simultaneously. Hence in this study, a Taylor diagram is used for statistical analysis of the model-predicted fields.

3

Result and discussion

3.1 Validation of daily averaged surface wind simulated by the model Since the forecasts with one to two days of lead time are highly influenced by the initial condition, we considered the third-day forecasts for this validation exercise to assess the performance of the model predictions. As a preliminary assessment of the quality of the surface wind prediction, the time series of the zonal and meridional components of the surface wind predicted by the model is shown along with that observed by the DS1 buoy and derived from scatterometer (ASCAT) at (15°N, 69°E) in Fig. 2. The model-simulated wind shows better agreement with the ASCAT wind than with the buoy wind. This could be due to the fact that the model and ASCAT winds are gridded averages over a region, whereas the buoy data are point observations. It is seen that the model prediction for zonal wind is not in good agreement with the observations during the first week of the southwest monsoon (June–September). Observations show a rapid strengthening of the southwesterly from the first week of June. The model does not capture these sudden changes. During the northeast monsoon (October–December), the model zonal wind shows a clear negative bias (easterly bias) compared to the observations. The model simulations for

VOL. 7

the meridional wind fields during the pre-monsoon (March–May) and northeast monsoon seasons have good agreement with both of the observations. In the southwest monsoon season, the buoy wind and ASCAT wind are significantly different to each other. Statistical analysis of the model-simulated wind is represented using a Taylor diagram in Fig. 3 for the prominent Indian seasons such as pre-monsoon, southwest monsoon, and northeast monsoon in a calendar year. A Taylor diagram for the zonal wind (Fig. 3a) shows that the model has high correlation, low RMSE, and a good agreement of standard deviation with the ASCAT observations for the whole year and for all seasons except the southwest monsoon. Correlation between the model zonal wind and the buoy observation shows a very high value (0.94) if we consider data for the entire year and a high value (greater than 0.75) for all individual seasons. Even though the zonal wind forecast shows a high correlation with the buoy observation for all the seasons, it has a high RMSE (higher than 75% of standard deviation of the buoy wind) and failed to simulate the amplitude of the wind accurately. This could be due to the large easterly bias in the zonal wind forecasts in all the seasons with a maximum of −3.109 m s−1 during the northeast monsoon season. The Taylor diagram for the meridional wind (Fig. 3b) shows that the model has good skill in predicting both the amplitude and pattern variability of the meridional wind compared to both the observations if the data for the full period is considered. However, the model shows only moderate skill in predicting both the amplitude and the pattern variability of the meridional winds during the southwest and northeast monsoon seasons. It is also noted that the model-predicted meridional wind also shows a southerly bias for all seasons with a maximum of −2.32 m s−1 for the southwest monsoon.

Figure 2 Time series of daily averaged (a) zonal and (b) meridional wind simulated by the model (black) is compared with buoy (red) and ASCAT (blue) observations at (15°N, 69°E).

NO. 5

VISHNU AND FRANCIS: HIGH-RESOLUTION WRF MODEL SIMULATIONS OF SURFACE WIND

461

Figure 3 Taylor diagrams of model-simulated wind with the observations from ASCAT (blue) and the buoy (red): (a) zonal wind; (b) meridional wind. Dotted semicircles over the horizontal axis represent the centered RMSE between model simulation and observations.

Figure 4 shows the RMSE (normalized with the standard deviation) between the model prediction and the observation by the ASCAT as well as the correlation between the observation and the prediction, for the zonal and meridional components of surface wind. The model wind field is spatially interpolated onto the ASCAT grids to make the horizontal resolution the same for the comparison. Most of the places over the model domain show very high correlation (> 0.90) and low RMSE (< 40%) for both the components of wind, especially on the western parts of the domain. The model-simulated zonal and meridional winds show a clear decreasing (increasing) tendency in the correlation (RMSE) towards the coast. This tendency is

higher for the meridional wind component. This may be either due to the error in the wind speed estimates by the ASCAT, because of high backscatter near the coast, or due to the inability of the model to represent the highly complex topography of region accurately. 3.2 Validation of diurnal variability of the model simulated surface wind The diurnal variability in the surface wind field is also studied using the model hindcasts. The model prediction shows the presence of a land-sea breeze during the northeast monsoon and pre-monsoon seasons; however, it is not prominent during the southwest monsoon season (figures

Figure 4 Correlation (left panels) and normalized RMSE (right panels) between model-simulated daily averaged wind and ASCAT observations from February 2012 to January 2013; zonal wind (top panels); meridional wind (bottom panels).

462

ATMOSPHERIC AND OCEANIC SCIENCE LETTERS

VOL. 7

Figure 5 Monthly averaged synoptic hourly model predictions (blue) of the zonal (left panels) and meridional wind (right panels) compared with SW1 buoy (red) observations: April 2012 (top panels); May 2012 (bottom panels).

not shown). It is seen that during the southwest monsoon season, even though the wind speed shows diurnal variation, the wind direction remains west-southwesterly throughout. This may be due to the presence of strong mean southwesterlies that completely mask the sea breeze (Nair and Narayanan, 1980). Figure 5 shows the comparison of the monthly averaged synoptic hour model simulated wind components with the SW1 buoy observation for April and May 2012. The model-predicted wind fields and the buoy observations show a similar diurnal variability. Although the model-simulated wind fields show a negative bias for both wind components, with a higher bias for the meridional wind component.

Acknowledgments. The lead author is grateful to INCOIS, Ministry of Earth Sciences (MoES) for providing the necessary facilities and to University Grants Commission (UGC) for funding to pursue this work. We would also like to acknowledge the two anonymous reviewers, whose comments and suggestions greatly improved the final manuscript. This is INCOIS publication 181.

4

References

Conclusions

This study analyzed the performance of the high-resolution setup of the WRF mesoscale model with regard to the surface wind conditions over the west coast of India and eastern Arabian Sea by validating the predicted surface wind fields using the buoy observations and ASCAT measurements. On the whole, the correlation between the predicted zonal wind and ASCAT observations is very high (~ 0.85–0.90) and the RMSE is low (~ 30%–70% of standard deviation of the observation), but the model predictions have relatively lower correlation (~ 0.75–0.80) and higher RMSE (~ 60%–99%) with the buoy observations. A comparison with observations suggests that the model possesses reasonably good skill in predicting the surface meridional wind (correlation is ~ 0.85–0.95 and RMSE is ~ 50%) during the pre-monsoon season, but the model has only moderate skill (correlation is ~ 0.60–0.85 and RMSE is above 75%) in predicting the surface wind fields during the southwest monsoon and northeast monsoon seasons. A comparison with the observations also suggests that the model can predict the diurnal variability of wind components accurately, but a significant negative bias is seen for both components of wind during all seasons, with a

wind during all seasons, with a maximum for the zonal wind during the northeast monsoon season. This negative bias could be due to the terrain-related model error, such as inaccurate representations of elevation, ruggedness, and surface roughness. Therefore, more analyses are needed to study the influence of the surface winds with the complex terrain.

Chen, F., and J. Dudhia, 2001: Coupling an advanced land surface-hydrology model with the Penn State-NCAR MM5 modeling system. Part I: Model implementation and sensitivity, Mon. Wea. Rev., 129, 569–585. Colle, B. A., K. J. Westrick, and C. F. Mass, 1999: Evaluation of MM5 and Eta-10 precipitation forecasts over the Pacific Northwest during the cold season, Wea. Forecasting, 14, 137–154. Das, S., R. Ashrit, G. R. Iyengar, et al., 2008: Skills of different mesoscale models over Indian region during monsoon season: Forecast errors, J. Earth Syst. Sci., 117, 603–620. Davis, C., T. Warner, J. Bowers, et al., 1999: Development and application of an operational, relocatable, mesogamma-scale weather analysis and forecasting system, Tellus, 51A, 710–727. Deb, S. K., T. P. Srivastava, and C. M. Kishtawal, 2008: The WRF model performance for the simulation of heavy precipitating events over Ahmedabad during August 2006, J. Earth Syst. Sci., 117, 589–602. 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. Francis, P. A., N. Vinayachandran, and S. S. C. Shenoi, 2013: The Indian Ocean forecast system, Curr. Sci., 104, 1354–1368. Hong, S. Y., J. Dudhia, and S. H. Chen, 2004: A revised approach to ice microphysical processes for the bulk parameterization of clouds and precipitation, Mon. Wea. Rev., 132, 103–120.

NO. 5

VISHNU AND FRANCIS: HIGH-RESOLUTION WRF MODEL SIMULATIONS OF SURFACE WIND

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. Kain, J. S., 2004: The Kain-Fritsch convective parameterization: An update, J. Appl. Meteor., 43, 170–181. Mlawer, E. J., S. J. Taubman, P. D. Brown, et al., 1997: Radiative transfer for inhomogeneous atmosphere: RRTM, a validated correlated-k model for the longwave, J. Geophys. Res., 102, 16663–16682. Nair, S., and V. Narayanan, 1980: Diurnal variation of lower tropospheric winds (0–3 km) over Thumba during 18–20 August 1976, Mausam, 31, 409–414. Premkumar, K., M. Ravichandran, S. R. Kalsi, et al., 2000: First results from a new observation system over the Indian Seas, Curr. Sci., 78, 323–330. Raju, P. V. S., P. Jayaraman, and U. C. Mohanty, 2012: Prediction of severe tropical cyclones over the Bay of Bengal during 2007–2010 using high-resolution mesoscale model, Nat. Hazards, 63, 1361–1374.

463

Sela, J., 2009: Implementation of the Sigma Pressure Hybrid Coordinate into GFS, NCEP office Note-461, 1–25, available at: http://www.emc.ncep.noaa.gov/officenotes/FullTOC.html. Skamarock, C. W., J. B. Klemp, J. Dudhia, et al., 2008: A Description of the Advanced Research WRF Version 3, NCAR Technical Notes, Boulder, 101pp. Sousounis, P. J., T. A. Hutchinson, S. F. Marshall, 2004: A comparison of MM5, WRF, RUC, ETA performance for great plains heavy precipitation events during the spring of 2003, in: Preprints 20th Conference on Weather Analysis and Forecasting, Seattle, Amer. Meteor. Soc., vol. J24.6. Taylor, K. E., 2001: Summarizing multiple aspects of model performance in a single diagram, J. Geophys. Res., 106, 7183–7192. Verspeek, J., M. Portabella, A. Stoffelen, et al., 2013: Calibration and Validation of ASCAT Winds, OSI SAF technical report, SAF/OSI/KNMI/TEC/TN/163, 34pp. Wang, J., and H.-J. Wang, 2013: Forecasting of wind speed in Rudong, Jiangsu province by the WRF model, Climatic Environ. Res. (in Chinese), 18(2), 145–155.