Data Assimilation
in Climate and Energy
Systems Engineering

Due to recent climate change, the frequency and intensity of localized torrential rains have increased rapidly, leading to an expansion of damage from urban flooding and natural disasters. These localized heavy rains typically have a spatial scale of less than a few kilometers and a short temporal scale of about a few minutes; because they travel rapidly carried by the wind and their location changes in real time, accurately predicting the time and place of occurrence is extremely challenging.

There are two main factors that degrade the predictability of existing numerical weather prediction models (NWP). First, due to the nonlinearity of atmospheric systems, minute errors in initial conditions amplify exponentially over time, leading to prediction failures. Second, due to the limitations of model grid resolution, subgrid-scale phenomena cannot be directly calculated and must rely on parameterization, resulting in reduced forecasting accuracy.

In short, improving initial conditions is essential to reduce damage caused by localized torrential rains, which have become difficult to predict.