Causes of prediction failure

Causes of Prediction Failure

Major Causes of Local Heavy Rain Prediction Failure

Initial condition error

One of the key problems of the NWP is the accuracy of initial conditions. Equations governing atmospheric motion (Navier-Stokes equations, mass-continuous cubic thermodynamic energy equations) are defined as initial value problems, and small errors in initial states can be quickly amplified through nonlinear dynamics.

DA emerged to address this issue, which combines background and available observations to produce an analysis, an improved estimate of atmospheric conditions.

In particular, errors in the initial conditions of water vapor directly affect the prediction of precipitation. Accurately analyzing water vapor in the initial chapter of the numerical forecast model is one of the essential conditions for improving prediction accuracy.

Nonlinearity and chaos

Nonlinear equations governing atmospheric flow have chaos solutions that are sensitive to the initial condition
The governing equation fails to include the mechanical interaction of the entire turbulent transition spectrum
Due to the limitations of available computer resources, the governing equation has to be severely limped.
Roundup errors associated with computer precision exponentially amplify uncertainty over time

The Butterfly Effect is a phenomenon in which a minute initial condition difference in a complex system is non-linearly amplified over time, resulting in a completely different result. Edward Lorenz published the discovery in 1963 in a paper called 'Deterministic Nonperiodic Flow'.

In the existing numerical forecasting model, small errors in the initial stage are gradually amplified along the flow of the atmosphere, resulting in a butterfly effect that spreads naturally at various scales. Since weather forecasts are affected by the butterfly effect, which greatly expands over time, the Korea Meteorological Administration and others use an ensemble forecasting method that performs predictions several times with slightly different initial conditions.

Model Grid Resolution Limit

The Korea Meteorological Administration operates the Global Zone Model (GDAPS) with a horizontal grid resolution of about 25 km, the Regional Forecast Model (RDAPS) with a resolution of about 12 km, and the Local Forecast Model (LDAPS) with a resolution of about 1.5 km.

Since convection systems of local heavy rain develop on a scale of a few kilometers or less, numerical models must use limited computer resources and cannot be simulated by decomposing the lattice indefinitely small due to the observational limitations of the current observational system.

Gray zone: In the gray zone area, where the horizontal grid size of the numerical forecast model is between several hundred meters and about 10 kilometers, the rising airflow in the convective cloud cannot be explicitly fully interpreted, so the use of subgrid convective cloud parameters is still necessary.

Parameterization Error

Variations and errors in concentrated precipitation are due to the uncertainty of initial conditions, the unphysical parameterization scheme, and the uncertainty of subgrid-scale convection scheme. The causes of erroneous predictions result in the complexity of subgrid scale processes, parameterization of physical processes, inferior data assimilation techniques, or lack of computing power to run the model with very high resolution.

In addition, a key cause of error in convection scale numerical forecasting (CS-NWP) is the very rapid growth of initial conditions and model errors due to the nature of highly chaotic convection scale dynamics.

The current operational numerical forecasting model cannot explicitly interpret convection activity because the horizontal resolution is at least a few km. Therefore, there is still great uncertainty in how to parameterize the wetting process on a subgrid scale.