Search Results

You are looking at 1 - 5 of 5 items for

  • Author or Editor: Shunji Kotsuki x
  • Refine by Access: All Content x
Clear All Modify Search
Shunji Kotsuki
and
Craig H. Bishop

Abstract

Recent numerical weather prediction systems have significantly improved medium-range forecasts by implementing hybrid background error covariance, for which climatological (static) and ensemble-based (flow-dependent) error covariance are combined. While the hybrid approach has been investigated mainly in variational systems, this study aims at exploring methods for implementing the hybrid approach for the local ensemble transform Kalman filter (LETKF). Following Kretschmer et al., the present study constructed hybrid background error covariance by adding collections of climatological perturbations to the forecast ensemble. In addition, this study proposes a new localization method that attenuates the ensemble perturbation (Z-localization) instead of inflating observation error variance (R-localization). A series of experiments with a simplified global atmospheric model revealed that the hybrid LETKF resulted in smaller forecast errors than the LETKF, especially in sparsely observed regions. Due to the larger ensemble enabled by the hybrid approach, optimal localization length scales for the hybrid LETKF were larger than those for the LETKF. With the LETKF, the Z-localization resulted in similar forecast errors as the R-localization. However, Z-localization has an advantage in enabling us to apply different localization scales for flow-dependent perturbation and climatological static perturbations with the hybrid LETKF. The optimal localization for climatological perturbations was slightly larger than that for flow-dependent perturbations. This study also proposes optimal eigendecomposition (OED) ETKF formulation to reduce computational costs. The computational expense of the OED ETKF formulation became significantly smaller than that of standard ETKF formulations as the number of climatological perturbations was increased beyond a few hundred.

Open access
Shunji Kotsuki
,
Steven J. Greybush
, and
Takemasa Miyoshi

Abstract

With the serial treatment of observations in the ensemble Kalman filter (EnKF), the assimilation order of observations is usually assumed to have no significant impact on analysis accuracy. However, Nerger derived that analyses with different assimilation orders are different if covariance localization is applied in the observation space. This study explores whether the assimilation order can be optimized to systematically improve the filter estimates. A mathematical demonstration of a simple two-dimensional case indicates that different assimilation orders can cause different analyses, although the differences are two orders of magnitude smaller than the analysis increments if two identical observation error variances are the same size as the two identical state error variances. Numerical experiments using the Lorenz-96 40-variable model show that the small difference due to different assimilation orders could eventually result in a significant difference in analysis accuracy. Several ordering rules are tested, and the results show that an ordering rule that gives a better forecast relative to future observations improves the analysis accuracy. In addition, the analysis is improved significantly by ordering observations from worse to better impacts using the ensemble forecast sensitivity to observations (EFSO), which estimates how much each observation reduces or increases the forecast error. With the EFSO ordering rule, the change in error during the serial assimilation process is similar to that obtained by the experimentally found best sampled assimilation order. The ordering has more impact when the ensemble size is smaller relative to the degrees of freedom of the dynamical system.

Open access
Shigenori Otsuka
,
Shunji Kotsuki
, and
Takemasa Miyoshi

Abstract

Space–time extrapolation is a key technique in precipitation nowcasting. Motions of patterns are estimated using two or more consecutive images, and the patterns are extrapolated in space and time to obtain their future patterns. Applying space–time extrapolation to satellite-based global precipitation data will provide valuable information for regions where ground-based precipitation nowcasts are not available. However, this technique is sensitive to the accuracy of the motion vectors, and over the past few decades, previous studies have investigated methods for obtaining reliable motion vectors such as variational techniques. In this paper, an alternative approach applying data assimilation to precipitation nowcasting is proposed. A prototype extrapolation system is implemented with the local ensemble transform Kalman filter and is tested with the Japan Aerospace Exploration Agency’s Global Satellite Mapping of Precipitation (GSMaP) product. Data assimilation successfully improved the global precipitation nowcasting with the real-case GSMaP data.

Full access
Shunji Kotsuki
,
Kenta Kurosawa
,
Shigenori Otsuka
,
Koji Terasaki
, and
Takemasa Miyoshi

Abstract

Over the past few decades, precipitation forecasts by numerical weather prediction (NWP) models have been remarkably improved. Yet, precipitation nowcasting based on spatiotemporal extrapolation tends to provide a better precipitation forecast at shorter lead times with much less computation. Therefore, merging the precipitation forecasts from the NWP and extrapolation systems would be a viable approach to quantitative precipitation forecast (QPF). Although the optimal weights between the NWP and extrapolation systems are usually defined as a global constant, the weights would vary in space, particularly for global QPF. This study proposes a method to find the optimal weights at each location using the local threat score (LTS), a spatially localized version of the threat score. We test the locally optimal weighting with a global NWP system composed of the local ensemble transform Kalman filter and the Nonhydrostatic Icosahedral Atmospheric Model (NICAM-LETKF). For the extrapolation system, the RIKEN’s global precipitation nowcasting system called GSMaP_RNC is used. GSMaP_RNC extrapolates precipitation patterns from the Japan Aerospace Exploration Agency (JAXA)’s Global Satellite Mapping of Precipitation (GSMaP). The benefit of merging in global precipitation forecast lasts longer compared to regional precipitation forecast. The results show that the locally optimal weighting is beneficial.

Open access
Masahiro Momoi
,
Shunji Kotsuki
,
Ryota Kikuchi
,
Satoshi Watanabe
,
Masafumi Yamada
, and
Shiori Abe

Abstract

Predicting the spatial distribution of maximum inundation depth (depth-MAP) is important for the mitigation of hydrological disasters induced by extreme precipitation. However, physics-based rainfall-runoff-inundation (RRI) models, which are used operationally to predict hydrological disasters in Japan, require massive computational resources for numerical simulations. Here, we aimed at developing a computationally inexpensive deep learning model (Rain2Depth) that emulates an RRI model. Our study focused on the Omono River (Akita Prefecture, Japan) and predicted the depth-MAP from spatial and temporal rainfall data for individual events.

Rain2Depth was developed based on a convolutional neural network (CNN), and predicts depth-MAP from 7-day successive hourly rainfall at 13 rain gauge stations in the basin. For training the Rain2Depth, we simulated the depth-MAP by the RRI model forced by 50-ensembles of 30-year data from large-ensemble weather/climate predictions. Instead of using the input and output data directly, we extracted important features from input and output data with two dimensionality reduction techniques (principal component analysis (PCA) and the CNN approach) prior to training the network. This dimensionality reduction aimed to avoid overfitting caused by insufficient training data. The nonlinear CNN approach was superior to the linear PCA for extracting features. Finally, Rain2Depth was architected by connecting the extracted features between input and output data through a neural network.

Rain2Depth-based predictions were more accurate than predictions from our previous model (K20), which used ensemble learning of multiple regularized regressions for a specific station. Whereas the K20 can predict maximum inundation depth only at stations, our study achieved depth-MAP prediction by training only the single model Rain2Depth.

Free access