Sort Options:

• Sort By Last Name
• Sort By Title

Other:

• Print All

Sort by time-scale:

• Short
• Intermediate
• Long
• SHOW ALL

Evaluation of Precipitation Extremes Simulated by a Global 20-km-grid Atmospheric Model using L-moments Method

Evaluation of Precipitation Extremes Simulated by a Global 20-km-grid Atmospheric Model using L-moments Method

Kenji Kamiguchi (MRI, Japan)


Prediction of future change in extreme precipitation is very important and challenging study. Long time integration with high resolution model is required to do that. In this study, the annual maximum of daily precipitation (AMDP) simulated by a global 20-km-grid atmospheric model is evaluated by the rain gauge observation using L-moments method which is an excellent technique for extreme analysis.

The model used here is MRI-AGCM (Atmospheric General Circulation Model) whose horizontal resolution is 20-km. An AMIP climate simulation is conducted for 20-year.
L-moments are calculated in each grid using 20 samples of the AMDP. Meanwhile, rain gauge data contained in GDCN (Global Daily Climatology Network) is used as an observation. L-moments of the observation are calculated for the rain gauge stations whose observation length is the same or longer than 20-year. Though I recognize that there is room for further discussion about adequacy of direct comparison between grid value and station value, grid value is compared with station value at correspondent location.

The geographical pattern of the first order of L-moment (arithmetic mean of samples) in the model is in good agreement with that of the observation, reflecting realistic topography in the model.

However, value is smaller than the observed ones. As the third and the forth order of L-moment (called L-skewness and L-kurtosis, respectively) are plotted on L-moments diagram, the probability distribution function (PDF) of the AMDP roughly conforms to Gumbel distribution, both in the model and in the observation.

However, in the observation, there are many places whose L kurtosis and L-skewness are larger than that of the model.