Exploiting new observations and data assimilation techniques for improved forecasting of convective precipitation

Brief periods of intense rainfall can lead to flash flooding with the potential to cause millions of pounds of damage to property, and to threaten lives. Accurate flood warnings even just a few hours ahead can allow preparations to be made to minimize damage. In order to improve the prediction of these events, more accurate forecasts of heavy rainfall are needed, which can then be used to inform flood prediction and warning systems. The UK Met Office is developing a new numerical weather prediction system with the goal of improving severe weather forecasts. This is a computer model that solves mathematical equations representing atmospheric motions and other physical processes such as cloud formation, with a horizontal grid spacing of 1.5km. This allows a more accurate representation of fine-scale features and explicit representation of storms, but the results are still dependent on the accuracy of the starting conditions or initial data describing the current state of atmospheric variables such as winds, pressure, temperature and humidity. Initial conditions are usually estimated using a sophisticated mathematical technique known as data assimilation that blends observations with model information, taking account of the uncertainties in the data. In this project, we propose fundamental research to reduce initial condition errors. The work will be carried out in a partnership between the Universities of Reading, Surrey and the Met Office.

We plan to investigate ways of extracting the maximum information from weather radar observations of precipitation and moisture in the lower parts of the atmosphere. Although rainfall is usually well observed by weather radar, severe precipitation can cause the radar beam to lose energy, and thus the weaker returned signal may be misinterpreted, giving a lower rain-rate than in reality. We will develop algorithms to correct for this and other problems caused by severe rainfall. Recently, we have also developed techniques to infer humidity information about the lower atmosphere, and we plan to optimize the method and investigate the observation error characteristics, to prepare for this data to be assimilated by the Met Office.

One of our goals is to use observations to provide information on the small scales without degrading the large scale weather patterns, which are themselves likely to be accurate. However, currently much of the small scale observational information is being lost by ignoring correlations between observation errors. We will develop a generic approach for treating observation correlations for a range of observation types. We will investigate mathematical methods that both capture the maximum amount of information contained in the observations, while still being practical for operational computations, which have to take place within a limited time frame.

Another goal is to develop innovative ways of treating moist processes that are largely absent from present-day assimilation systems. We plan to design and test efficient and effective ways of assimilating moisture information that respect the intricate dynamical and physical relationships that operate in the atmosphere. If successful, such new approaches will allow better use of cloud and rain affected observations than at present.

Predicting convective rain is made harder by the fact that some events are inherently unpredictable, even with good data assimilation and models, due to their high sensitivity to even small errors in the initial conditions. Further studies will be made to look at the dynamical reasons for the low predictability of such events using diagnostics derived from models and observations.

The primary impact of this project will be on the ability of the Met Office to forecast high impact rainfall, and on downstream systems and users of those forecasts.

Proposed work on improving the quality and accuracy of radar reflectivity measurements and derived rainfall estimates will have impact in a number of ways. Firstly, there will be an immediate impact via the Met Office STEPS/UKPP numerical weather prediction (NWP) post-processing system. This objectively merges extrapolation forecasts with output from numerical weather prediction (currently the 1.5 km grid-spacing model, updated every 3 h, at best) to produce an 'optimal' forecast every hour. The output is used to drive hydrological models by the Environment Agency; in particular, it is used to drive the G2G model operated within the Joint Met Office/ Environment Agency Flood Forecasting Centre (FFC), which gives a national view of short-term flood risk, including un-gauged catchments. It is also used as guidance for the Extreme Rainfall Alerts, along with the UK convective-scale ensemble system, UK-MOGREPS. Improved radar-derived rainfall will also have a direct impact on the quality of the NWP forecast, as rainfall is assimilated and, incidentally, enable a more useful verification of any rainfall forecasts to be performed.

Longer term improvements will come primarily through the implementation of a nowcasting configuration of the NWP system. This could become a replacement input into STEPS/UKPP and so naturally feed through to the same users. The Met Office have a key deliverable of their Public Weather Service R&D plan to have an 'Operational implementation of next-generation NWP-nowcasting system' by March 2017. This would potentially include implementation of direct assimilation of radar reflectivity by 4DVAR rather than rainfall via latent heat nudging. This brings advantages of consistency and coupling to model dynamics, but will benefit enormously from the more accurate data, the better characterization of their errors and the more efficient use of data via improved representation of error covariance structure provided by this project. The latter will also have impact through better use of the information in the radar Doppler wind data and satellite observations already assimilated.

The NWP-based nowcasting system would also be the route through which the other main outputs of the project deliver impact, though they would also have impact in the existing NWP system if the nowcasting system were delayed. Use of radar refractivity provides a major source of low-level humidity information, vital for good forecasts of high-impact rainfall. This information is currently very sparse, mainly coming from surface synoptic stations but increasingly from aircraft as the land and take off from major airports. While we cannot predict the amount of forecast improvement, we know that deficiencies in low level moisture are a major contributor to forecast error.

A better (and more adaptive) choice of moisture-variable(s) in the Data Assimilation system will have impact on the assimilation system, especially assimilation of refractivity and reflectivity (for which it may prove essential); this is a current goal in the Met Office's rolling plan for improvements.
Many of these impacts will also have benefit for other forecast systems; techniques to model observation error covariance will be applicable in other systems and to other observation types. Application of refractivity will also be applicable in other radar systems.

Of course, the ultimate aim of this project is impact on the public (and thus the insurance industry) through improved flood warnings, especially of pluvial and fluvial flooding of small, fast-response catchments. This will bring benefits through reduced loss of life and damage to property.