Assimilating GMS-5 Data into a Mesoscale Model to Improve Rainfall Forecasts for New Zealand

Abstract

Moisture information is derived from multichannel GMS-5 data and assimilated into the RAMS mesoscale model to improve rainfall forecasts in an operational mesoscale forecasting system for New Zealand. A new statistical algorithm to retrieve relative humidity profiles from GMS data is developed and characterised. The algorithm is trained on radiosonde sounding collocated with GMS imagery and forecast data. It uses Bayesian classification into relative humidity regimes and class-specific regression, and estimates regime-specific retrieval error covariances. The regime-specific error-covariances are investigated in detail, and pragmatic univariate assimilation scheme based on Newtonian relaxation. The impact of the data assimilation and other model parameters on rainfall forecasts in studied by performing several parallel series of model experiments for the SALPEX 96 extensive observing period and by comparing rainfall forecasts against rain gauge observations. These model experiments include the first with mesoscale data assimilation in New Zealand. The study also reports on a four month trail of the retrieval and assimilation schemes and their operational application. Validation and sensitivity studies with RAMS reveal that the mesoscale model can add significant value to rainfall forecasts from a global model, even without mesoscale data assimilation, particularly when the full RAMS microphysics scheme is used. The higher spatial resolution of the mesoscale model is better able to resolve the steep New Zealand orography and associated sharp rainfall gradients. Snow, graupel, and aggregates provide important enhancement mechanisms for rainfall in the Southern Alps. Modelling processes related to these hydrometeor species improves forecasts of heavy orographic rain and rainfall forecasts in the lee of the Southern Alps. The soil moisture initialization in RAMS affects forecasts of light rain. Increasing the size of the mesoscale model domain does not always improve rainfall forecasts in the data sparse New Zealand region. The algorithm to retrieve relative humidity profiles from GMS data shows significant skill in estimating broader vertical structures (r2 ≈60%), particularly at lower levels. Upper level estimates are dominated by the GMS 6.7 µm channel, whereas forecast data from a global model contribute significant information at lower levels. On New Zealand data, the algorithm performs significantly better than other algorithms currently used internationally. Compared to the error characteristics of a 12h global model forecast, the retrieval error shows significant regime-specific differences with smaller errors at upper levels in cloud-free regions and larger error at lower in cloudy regions. The retrievals capture more mesoscale detail and improve a first-guess from a 12h global forecast when combined according to the respective error characteristics. When assimilated into a mesoscale model during a 12h preforecast period the GMS retrievals can significantly improve rainfall forecasts. The impact of GMS moisture assimilation is strongest in slow-moving, light southerly, or convective situations, whereas no significant changes are noted in north-westerly storms with heavy orographic rain. Mostly, the impact of the assimilation on the flow field is small, and if the model fails to predict the correct flow the assimilation of the GMS retrievals can degrade rather than improve rainfall forecasts through incorrect advection. In the extended four month trail, the net-improvement of rainfall forecasts through the GMS data assimilation is relatively small.