Improved nowcasting of precipitation based on convective analysis fields

Abstract

In quantitative prediction of precipitation the term 'nowcasting' usually refers to forecasts of up to 2-3 h ahead. Within this time range the precipitation field is in many cases closely related to its initial state. Thus there is prognostic information contained in the precipitation pattern observed at analysis time. Nowcasting methods are designed to make use of this information based on various extrapolation techniques. The local time evolution of a two-dimensional meteorological field (x,y) can formally be written V dt d t, (1) where V (u, v) is not necessarily a horizontal wind vector but a more general motion vector. In the case of convective cells, for example, V can be different from the wind at any atmospheric level. Purely advection-based nowcasting methods are based on the assumption that the first term on the right hand side of Eq. (1) can be neglected, and (t) (t 0) (2) following the motion. The local rate of change is then solely determined by the advection of existing patterns. Precipitation systems often exhibit a high Lagrangian - low Eulerian persistence, which justifies this assumption. Expressed in terms of scales, advection dominates if U L EVOL, (3) where EVOL is the time scale of Lagrangian evolution of the system, L is the length scale over which significant along-flow variations of precipitation occur, and U is the translation speed. The condition of Eq. (3) is fulfilled best for quasi-stationary (in the Lagrangian sense), fastmoving systems with sharp along-flow precipitation gradients such as fronts or squall lines. Of course such systems undergo a life-cycle of formation, intensification, weakening, and dissipation. They are typically composed of smaller-scale sub-structures which have a shorter evolution time EVOL. Nevertheless, as the inequality expressed by Eq. (3) illustrates, if precipitation gradients are sufficiently sharp (L sufficiently small) advection algorithms will give useful nowcasts even if the evolution timescale is short. Convective cells can develop in synoptic environments with weak mid-tropospheric winds, which allows them to remain more or less stationary with respect to the ground. Often the topography plays a role both in the triggering and the anchoring of convection to a certain area (Banta 1990). In such cases the translation speed U may become arbitrarily small so that inequality expressed by Eq. (3) is no longer fulfilled. A purely advection-based algorithm then degenerates to an Eulerian persistence forecast. Unfortunately, such quasi-stationary systems can lead to disastrous hydrological consequences. Due to the near-zero translation, extreme precipitation amounts may accumulate in a given area (Caracena et al. 1979). In general, precipitation systems move and evolve. Mathematically, this is equivalent to taking into account higher-order terms in the Lagrangian time evolution Eq. (2) dt d 2 (t t) dt (t) (t) (t t) d 2 2 2 0 0 0 (4)Attempts to improve the advection forecast by taking into account the linear term while neglecting higher-order terms have been of limited success due to the life-cycle behavior of precipitating convection. In order to mathematically represent an evolution (t) that undergoes a life-cycle of accelerated growth, decelerated growth, and weakening, a polynomial of at least third order would be necessary. Fitting such a polynomial requires observations at four consecutive times, with at least three of them being non-zero. Such a method can be used if a convective cell has already existed in the radar data for at least three time-steps. However, it is unlikely that the time and magnitude of the maximum intensity can be reliably predicted in this way unless the latest observation is already well within the stage of decelerating growth and thus itself close to the maximum. Consequently, algorithms designed to predict the formation of new convective cells, or convective initiation (CI), cannot be based solely on statistical methods but must include some physical considerations. The object-oriented cell evolution algorithm in the GANDOLF nowcasting system, for example, employs a mixed methodology of extrapolation, cell developmental stage classification, and physical relationships (Hand 1996; Pierce et al. 2000). One of the most important predictors of CI appears to be boundary-layer mass convergence. As shown in an experimental nowcasting study by Wilson and Schreiber (1986), CI in the High Plains area around Denver is closely associated with boundary-layer convergence zones ('boundaries'). Human forecasters at the Denver Stapleton Airport were able to predict thunderstorm initiation by monitoring radar-detected boundaries and associated cumulus cloudiness. This resulted in improved forecast skill compared to automated advection forecasts. The main problem was that the boundaries indicated the location area of new cell formation but not their precise position and initiation time. According to Wilson and Mueller (1993), small-scale (a few km) structures in the temperature, humidity, and wind field appear to determine where and when cells will form. The methodology of using boundaries to identify areas of incipient CI was later incorporated into the Autonowcaster (ANC) system (Mueller et al. 2003). A related approach is used in the GANDOLF system (Hand 1996). Convective cells are classified into different stages of development. Based on radar data and on a conceptual model of storm evolution, the current state of a cell is diagnosed and future states predicted. New ('daughter') cells close to existing ones are initiated if the near-surface mass-convergence predicted by a numerical weather prediction (NWP) model is sufficiently strong. During the 2000 Sydney Olympics Forecast Demonstration Project (FDP) various nowcasting systems were tested and evaluated (Pierce et al. 2004). These systems use different methods of determining precipitation motion vectors such as area tracking, individual cell tracking, and NWP model winds. Two of the systems, namely GANDOLF and ANC, have convective evolution and initiation capability. The main findings with regard to convective cell prediction in the Sydney FDP can be summarized as follows (Wilson et al. 2004). (1) Predictive skill above pure translation occurs when boundaries can be identified and used to nowcast cell evolution. (2) For nowcasts beyond 60 min, boundary characteristics are more important for storm initiation than early detection of cumulus clouds. (3) The accuracy of nowcasts even for periods 60 min is generally quite low. For the development of a cell initiation and evolution module within the INCA system these results served as a guideline. They indicate the importance of a high resolution analysis of wind, temperature, and humidity in the boundary layer. In Austria's alpine terrain, boundary layer mass-convergence is often related to the topography, which adds a deterministic component to cell initiation (Haiden 2004). With the current version of the INCA wind field analysis, the ability to correctly detect these convergence lines depends critically on the skill of the NWP model, and on the density of the surface station network. Only if a convergence line is either correctly predicted by the NWP model or captured by the surface station data it will be present in the INCA analysis and give a signal for CI or intensification. Section 15.2 gives a brief overview of the INCA analysis and nowcasting system used operationally at ZAMG. The advection nowcast which serves as a reference for the cell evolution forecasting experiments is described in Sect. 15.3. A discussion of the convective analysis fields used for the cell evolution algorithm is given in Sect. 15.4, and the algorithm itself is described in Sect. 15.5. Verification and parameter sensitivity are presented in Sect. 15.6, followed by some thoughts about orographic effects in convection initiation (Sect. 15.7). The concluding section summarizes the main problems and proposes priorities for possible further research in the area.