Project topics and materials
Get Complete Project Material File(s) Now! »
FORECASTING RAINFALL USING NUMERICAL WEATHER PREDICTION SYSTEMS
General concepts
The science of weather forecasts has developed significantly over the past decades with most of the improvement of skill due to the development of NWP models (Stensrud, 2007; WMO, 2009). The grid resolution of NWP systems has decreased steadily over the years and is now ranging from a few kilometres to 15 kilometres in regional models. Since only weather systems that are eight or more times larger than the grid spacing can be effectively resolved by NWP, smaller subgrid-scale weather phenomena still need to be approximated through parameterization schemes within the NWP model (Stensrud, 2007; Lean et al., 2008; Warner, 2011). This is particularly the case with convective systems. Individual convective elementsrequire grid spacing of between 25 m and 1000 m to be resolved directly. Stensrud (2007) and Warner (2011) describe a variety of parameterization schemes that exist. These schemes include convective schemes, and also schemes dealing with heat transfer, the boundary layer, soil, vegetation and cloud cover. Each of these parameterization schemes introduces inaccuracies due to their particular inherent approximations that can affect important features of subgrid-scale convection, which impact on forecasts of rainfall in severe thunderstorms.
Trigger functions that determine the activation time and place for convection schemes also play an important role. These uncertainties affect forecasting features of convective precipitation such as onset time, location of convection, duration of the rain event and the diurnal cycle of convection (Stensrud, 2007, Lean et al., 2008). Accurately representing the diurnal cycle of convection by convective parameterization schemes is a major problem in NWP models, although promising new research indicates that this deficiency will eventually be resolved (Bechtold et al., 2014) By definition, the amount of rainfall at a grid point represents the average amount of rain forecast over a box with the size of the model resolution and centred over the grid point (Stensrud, 2007; Warner, 2011). As mentioned earlier, NWP models often have difficulty in accurately forecasting the exact rainfall location, onset timing of convection, and even total amount of rainfall (Theis et al., 2005; Stensrud, 2007; Lean et al., 2008, Bechtold et al., 2014).
This can differ between model configurations when, for example, the same basic model system is used, but with different initial conditions or parameterization schemes. According to Lean et al. (2008) precipitation forecasts also change when increasing the resolution for the same general model configuration. In their study, they compared rainfall prediction of the Unified Model (UM) from the United Kingdom Met Office at configurations for 12 km, 4km and 1 km resolutions. Convective parameterization was switched off in the 1 km version. Generally, a significant improvement of rainfall prediction was found with smaller grid box resolution, mainly due to the explicit modelling of precipitation compared to using a parameterization scheme necessary at 12 km. Typically, it was found that the 12 km modeltends to initiate convection on average 1-2 hours too early, and it tends not to have a realistic shower structure, which could lead to an underestimation of rainfall peaks. Additionally, the 12 km model tends to produce too much light rain and too little heavy rain compared to observations due to the averaging of rain over a larger grid box compared to the 4 km and 1 km models.
Addressing uncertainty of rainfall forecasting
From the previous discussion, rainfall forecasting is indeed quite problematic and difficult to predict accurately (see also Ebert, 2001; Theis et al., 2005). Nevertheless, it is the most important meteorological variable required to increase the lead-time of flash flood forecasts. When attempting to use a single NWP model to predict rainfall accurately for small river basins, as used in the SAFFG and SARFFG, the uncertainty linked to model prediction of rainfall is significant and can become quite challenging, as described in Section 4.2.1. This refers particularly to the uncertainty of the location of convective elements, amount of rain and the problem of the timing of convective precipitation during the day. These challenges can be addressed through an ensemble of NWP forecasts, as discussed in Chapter 3 (Ebert, 2001; Theis et al., 2005; NRC, 2006; Toth et al., 2007; Collier, 2007; He et al., 2009; Warner, 2011; Landman et al., 2012).
CHAPTER 1 INTRODUCTION
1.1 THE NEED FOR IMPACT FORECASTING
1.2 PRESENT INTERNATIONAL STATUS OF IMPACT FORECASTING
1.3 PROBLEM DEFINITION
1.4 PURPOSE, GOALS AND SPECIFIC OBJECTIVES OF THE STUDY
1.4.1 Purpose of the study
1.4.2 Main goals
1.4.3 Specific objectives
1.5 SCOPE OF THE STUDY AND CONTRIBUTION TO SCIENCE
1.6 STRUCTURE OF THE STUDY
CHAPTER 2 EARLY WARNING SYSTEMS AND FLASH FLOOD DISASTERS
2.1 INTRODUCTION
2.2 WEATHER-RELATED DISASTERS
2.3 EWS AS A COMPONENT OF DISASTER RISK REDUCTION
2.4 WARNINGS AGAINST FLOOD DISASTERS
2.5 THE SOUTH AFRICAN FLASH FLOOD GUIDANCE SYSTEM
2.6 SUMMARY
CHAPTER 3 DECISION MAKING UNDER FORECAST UNCERTAINTY
3.1 INTRODUCTION
3.2 UNDERSTANDING FORECAST UNCERTAINTY
3.3 THE PROBLEM OF FORECAST UNCERTAINTY FOR USERS IN EARLY WARNING SYSTEMS .
3.5 SUMMARY .
CHAPTER 4 INCREASING THE PREDICTIVE LEAD-TIME OF FLASH FLOOD GUIDANC
4.1 INTRODUCTION
4.2 FORECASTING RAINFALL USING NUMERICAL WEATHER PREDICTION SYSTEMS
4.3 DEVELOPING A FLASH FLOOD OUTLOOK METHODOLOGY
4.4 RESULTS AND CASE STUDIES
4.5 DISCUSSION
4.6 CONCLUSION
CHAPTER 5 THE IMPACT MODEL FOR THE SEVERE WEATHER IMPACT FORECASTING SYSTEM1
CHAPTER 6 CONCLUSIONS AND RECOMMENDATIONS
