Development of a Two-way Nested LETKF System for Cloud-resolving Model - Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications - page 481

20.1 Introduction

In the last decade, local heavy rainfalls that developed in urban areas (e.g.
Tokyo Metropolitan area) in summer have influenced urban functions, and have
occasionally caused urban flash floods (e.g. Nerima heavy rainfall in 1999, Itabashi
heavy rainfall in 2005). To mitigate damages from local heavy rainfalls, accurate
forecasts are desired. A few experiments on local heavy rainfalls have been
performed so far with variational data assimilation systems. For instance, Kawabata
et al. ( 2007 ) reproduced the Nerima heavy rainfall that occurred in the Tokyo
Metropolitan area by assimilating GPS, (Global Positioning System) precipitable
water vapor, (PWV, amount of water vapor in a column) and radial wind of
Doppler radars with JMANHM (Japan Meteorological Agency Non-hydrostatic
Model)-4DVAR (4 dimensional variational data assimilation system), and pointed
out that low-level convergence of water vapor is essential to reproduce local heavy
rainfalls.
Besides variational data assimilation methods, ensemble Kalman filters (EnKFs)
can provide initial conditions that are close to actual fields by assimilation of
observation data. Some previous studies have used the EnKF for mesoscale appli-
cations and have obtained promising results (e.g., Snyder and Zhang 2003 ; Zhang
et al. 2004 , 2006 ; Dowell et al. 2004 ; Tong and Xue 2005 ; Xue et al. 2006 ; Meng
and Zhang 2008 ; Seko et al. 2011 ; Miyoshi and Kunii 2012 ). In addition to accurate
initial conditions, EnKFs used as assimilation systems provide the probability
of heavy rainfalls and a number of rainfall forecasts. Especially in the forecast
of local heavy rainfalls, horizontal convergence in which local heavy rainfalls
are generated is generally relatively weak and the predicted distribution of local
heavy rainfalls is widely spread. Thus, it should be considered that the predicted
distribution is a part of the fields that have probability density distributions.
Because of these merits, Local Ensemble Transform Kalman Filter (LETKF, Hunt
et al. 2007 ) based on the JMANHM ( Saito et al. 2006 ), known as the NHM-LETKF
( Miyoshi and Aranami 2006 ), was used as the data assimilation system in this
study.
As mentioned before, local heavy rainfalls most often are generated in mesoscale
convergences. Even if convergence is relatively weak, mesoscale convergences
need to be reproduced by assimilation. In this study, mesoscale convergences were
produced by the LETKF system with a grid interval of 15 km (Outer LETKF).
Besides the position of convergence, rainfall intensity of local heavy rainfalls is
also important. Then, the LETKF systems with a grid interval of 1,875 km (Inner
LETKF), which reproduce positions and intensities of intense convection cells, are
deployed within the domain of the Outer LETKF.
As pointed out in Kawabata et al. ( 2007 ), convergence of low-level water vapor
is indispensable in reproducing local heavy rainfalls. Because GPS-derived PWV or
slant water vapor (SWV, water vapor amount along paths from GPS satellites to GPS
receivers) and horizontal wind or radial wind observed by Doppler radars provide
information about low-level convergence of water vapor, this data is expected to