Parameter Estimation Using an Evolutionary Algorithm for QPF in a Tropical Cyclone - Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications - page 695

27.1 Introduction

Numerical weather prediction (NWP) models have uncertainties involved in
the subgrid-scale physical processes, most of which have to be parameterized
( Navon 2009 ). In the formation of cloud and precipitation, the convection and
microphysical processes are important, and interactions among hydrologic, bound-
ary layer and land surface processes are conducted mostly by cumulus convection
( Arakawa 2004 ). Parameterizations, including convective parameterization (CP),
contain numerous parameters whose values are not known precisely. Parameters
in CP scheme can affect the model performance and hence the forecast accuracy.
Therefore optimization of parameters can potentially improve the accuracy of
numerical forecasts.
Genetic algorithms (GAs) have been used for global search by combining the use
of random number generation and information from precious iterations to evaluate
and improve a population of points at a time ( Goldberg 1989 ). They are based on
natural genetic and selection mechanism, and ideas of constructing an optimization
procedure are borrowed from Genetics ( Holland 1975 , 1992 ). With capability of
achieving global optimal solution, GAs have been applied to various atmospheric
problems ( Fang et al. 2009 ; Jackson et al. 2004 ; Kishtawal et al. 2003 ; Singh
et al. 2005a ,b).
A standard GA had been applied to a heavy rainfall case in Korea by Lee
et al. ( 2006 ) to improve the quantitative precipitation forecasting (QPF) through
optimization of both a physical and a computational parameter. This study focuses
on optimal parameter estimation in a CP scheme to improve the QPF in the Weather
Research and Forecasting (WRF) model using a micro-GA. Section 27.2 describes
the parameters to be optimized, the typhoon case and the experiments design, and
Sect. 27.3 describes the computational procedures of the micro-GA. Results are
presented in Sect. 27.4 , and conclusions are provided in Sect. 27.5

27.2 Methods

27.2.1 Description on Parameters

Deep convection is generally parameterized when horizontal grid spacing is greater
than about 10 km but CP is typically avoided at higher resolution ( Kain et al. 2008 ).
In this study, we will investigate the resolution dependency of performance of the
Kain-Fritsch (KF) CP scheme ( Kain 2004 ; Kain and Fritsch 1993 ), in the process
of GA optimization, by comparing three simulation runs - one with no CP scheme,
another with the default KF scheme, and the other with the improved KF scheme.
Here, two parameters are optimized by GA.
One parameter to be optimized is a convective time scale ( T c ). As illustrated by

Fritsch and Chappell ( 1980 ), the CP problem is mainly to determine T c and the grid