Ensemble numerical forecasts of the sporadic Kuroshio water intrusion (kyucho) into shelf and coastal waters
The finite volume coastal ocean model downscaling ocean reanalysis and forecast data provided by the Japan Coastal Ocean Predictability Experiment (JCOPE2) are used to forecast sudden Kuroshio water intrusion events (kyucho) induced by frontal waves amplified south of the Bungo Channel in 2010. Two-month hindcast computations give initial conditions of the following 3-month forecasts computations which consist of ten ensemble members. The temperature time series computed by these ten members are averaged to compare with that actually observed in the Bungo Channel, where sudden temperature rises related to kyucho events are remarkable in February, August, and September. Overall, the intense kyucho events actually observed in these months are predicted successfully. However, intense kyucho events are forecasted frequently during the period of May through June even though intense kyucho events are absent during this period in the actual ocean. It is suggested that the present downscaling forecast model requires reliable lateral boundary conditions provided by JCOPE2 data to which numerous Argo data are assimilated to enhance the accuracy. In addition, it seems likely that the model accuracy is reduced by small eddies moving along the shelf break.
KeywordsJCOPE2 FVCOM Ensemble forecast Kyucho Argo
Western boundary currents such as the Kuroshio in the North Pacific are always accompanied by frontal waves on the onshore side. Unlike the Gulf Stream in the North Atlantic, the Kuroshio path is located very close to the western boundary (the southern coast of Japan in this case), so the frontal waves excited along the ocean currents frequently extend onto the shallow shelf and coastal waters (Akiyama and Saitoh 1993; Arai 2005) and rapidly alter the marine environment in these areas (Koizumi and Kohno 1994; Katano et al. 2007; Hirose et al. 2008). In fact, it has been reported that these frontal waves amplified at the shelf break cause sporadic Kuroshio water intrusion into various coastal waters south of Japan, where cultured fishes and fixed nets have been frequently damaged by sudden increases in both temperature and current speed associated with this warm water intrusion (Takeoka and Yoshimura 1988 and references therein). The Japanese term “kyucho” (meaning sudden stormy currents) has been used by the oceanographic community of Japan as well as fishermen to refer to these intrusion events, which have heretofore received attention from oceanographers (e.g., Uda 1953). Besides the disadvantages to inshore fisheries as mentioned above, kyucho events have an advantage in that catches of commercially high-valued fishes increase suddenly in coastal waters during these events, presumably because fish around shelf breaks avoid rapid temperature changes (Hashida 2011, personal communication). Thus, a research project in high demand is to establish kyucho forecasting similar to weather forecasting.
To date, it has been difficult to forecast ocean currents and hydrographic properties in shelf and coastal waters despite the development of high-performance computers such as the Earth Simulator (Masumoto et al. 2004) along with sophisticated numerical model codes. The development of numerical forecasting in these shallow waters seems to have been prevented by two obstacles. One is that the density of observation stations is too sparse to construct objectively analyzed datasets suitable for providing reliable initial conditions for models. For instance, one of the most powerful tools to obtain spatiotemporally dense oceanographic data is satellite altimetry, which provides sea surface height (SSH) data approximately once every 10 days with spatial resolution of 100 km, while both temporal and spatial scales required for resolving Kuroshio frontal waves are one order of magnitude smaller (see Table 1 in James et al. 1999). The other obstacle is uncertain in lateral boundary conditions. Even if a forecast model does a reasonable job of simulating ocean currents within the model domain, a wrong condition at lateral boundaries may reduce model accuracy.
It is difficult to detect kyucho events directly in JCOPE2 data because their horizontal resolution of 1/12 of a degree is unlikely to resolve Kuroshio frontal waves accurately (Isobe et al. 2004). Probably the simplest way to simulate kyucho events is to downscale JCOPE2 data using a numerical model suitable for reproducing small-scale features in shelf and coastal waters. The present study employs the finite volume coastal ocean model (FVCOM; Chen et al. 2003) for resolving complex topography in shallow waters using triangular cell grids. In fact, a hindcast computation using JCOPE2 data as boundary conditions of the FVCOM well reproduces the kyucho events in 2003 (Isobe et al. 2010) because mesoscale warm eddies impinging on the Kuroshio south of Japan trigger the kyucho occurrence and these eddies are well reproduced in the JCOPE2 analysis. In addition, the JCOPE2 model successfully forecasts Kuroshio meanders after removing the assimilation schemes incorporated into the model (Miyazawa et al. 2005), and FVCOM computations using JCOPE2 forecast data as lateral boundary conditions are therefore expected to forecast kyucho events as they occur in the actual ocean.
The objective of the present study is to examine the capability of ensemble kyucho forecasts by downscaling either physical processes or resolution in JCOPE2 analysis into a smaller FVCOM domain (i.e., one-way nested model) around the Bungo Chanel. In the present application, 3-month forecast computations follow 2-month hindcast computations, during which observation data are not assimilated into the FVCOM domain. This is because, as mentioned above, the fine structure of Kuroshio frontal waves is likely to be destroyed by assimilating sporadic observations insufficient for coastal waters. The insufficient accuracy that may appear in the present forecast computations is caused by the uncertainty of small-scale processes generated in the FVCOM domain as well as that included within JCOPE2 data. It is worthwhile to specify causes reducing the capability of the forecasts to improve this “first generation” of kyucho forecast models.
2 Data and methods
2.1 Kyucho events in the Bungo Channel
2.2 Model setup and nests
Perturbations generated by connecting two different models are removed in sponge regions (shading in Fig. 4), where all modeled variables are restored to JCOPE2 reanalysis data. The coefficient (a reciprocal of time) of restoring terms diminishes linearly from 1/6 h−1 at the lateral boundaries to zero at the inner edge of the sponge regions. The width of the sponge regions is doubled along the eastern boundary to carefully retain the shape of warm eddies propagating westward as baroclinic Rossby waves and/or advection due to the Kuroshio Countercurrent. The FVCOM domain is divided into 23 layers vertically, and 23-layer JCOPE2 data are interpolated linearly to each FVCOM layer in the sponge regions.
Two datasets are combined to give depth data to each triangular cell grid of the FVCOM. The topographic dataset with 1/12° resolution (ETOPO5) provided by the US National Geophysical Data Center is used for the area south of 31.5° N, while the 500-m gridded topographic dataset provided by the Marine Information Research Center, Japan is used for the area north of 31° N. The above two topographic datasets are blended using a coefficient varying linearly in space between 31° N and 31.5° N latitudes. The depth in each triangular cell grid is determined by interpolating the nearest 16 topographic data weighted by the inverse distance.
2.3 Procedures of forecast computations
Three-month forecasts follow the above hindcasts, with the results at the end of the computations used as initial conditions (Fig. 5). Three-month period is chosen as the forecast duration because JCOPE2 provides us with 3-month daily forecast data, which are set to lateral boundaries in the same manner as in the hindcast computations. In addition, modeled SST is restored to monthly averaged JCOPE2 forecast data on a timescale of 5 days so that small-scale features revealed in the SST field of the original JCOPE2 analysis do not destroy those in the FVCOM domain. Forecast winds such as those of the Numerical Analysis and Prediction System of the Japan Meteorological Agency are not used to drive the present forecast model because, as mentioned above, short-term wind fluctuations are not critical to kyucho occurrences and because the use of forecast winds complicates interpretations of the forecast accuracy. Monthly averaged wind stresses computed using QSCAT data in the period 2002 through 2008 drive the model domain during the course of the forecast computations. Tidal forcing is not included in the forecast computations as in the hindcast computations.
4.1 Reliability of the lateral boundary conditions
As mentioned above, the kyucho events actually observed during the two shaded periods in Fig. 2 are all forecasted (mid-February and a (b) and d in Fig. 9) in the present experiment, while the kyucho events revealed in the forecast computations are not always observed in the actual ocean (run 3, run 4, and c). In general, the accuracy of downscaling computations depends strongly on the reliability of lateral boundary conditions, and it is thus worth investigating to what extent the boundary conditions provided by JCOPE2 data are reliable. The number of Argo floats south of Japan is likely to be crucial to enhancing the reliability of JCOPE2 data because hydrographic data assimilated to the reanalysis model are derived from numerous Argo floats and because the number of Argo floats carried passively by ocean currents is unlikely to be stable south of Japan.
Comparison of potential density anomalies in the forecast runs and those observed by Argo floats: the number of Argo floats (n) during each forecast run within the box outlined by the broken line in Fig. 4, correlation coefficient (r), and skill (s)
However, Fig. 9 apparently shows that the number of Argo floats is not a unique factor in determining the forecast accuracy. For instance, the sudden temperature rise c in run 6 is not present in the actual time series despite the forecast computation using the lateral boundary conditions corrected by numerous Argo data. Furthermore, the forecast computation goes well in run 1 (Figs. 7 and 9) in spite of there being the least Argo floats among all forecast runs. Hence, factors unrelated to the number of Argo floats south of Japan should be specified to improve this forecast model developed for coastal waters.
4.2 Modeled processes reducing the forecast accuracy
5 Conclusion possible improvement of the kyucho forecast model
The present study investigated whether numerical models downscaling JCOPE2 reanalysis and forecast data are capable of forecasting ocean circulation in coastal waters. The target oceanic phenomenon in the present study is the kyucho events generated in the Bungo Channel in 2010, when sudden temperature increases caused by Kuroshio water intrusion are observed in February, August, and September. In particular, the Kuroshio front south of the Bungo Channel is favorable for the kyucho occurrence in February because it is located close to the channel owing to the existence of a stable warm eddy south of Japan. This stable eddy is also revealed in the forecast computation (run 1), and the kyucho event thus occurs in the forecast model as in the actual ocean (Figs. 7 and 9). In addition, the kyucho events in August and September (a (b) and d in Fig. 9) are successfully forecasted in the present model, where mesoscale eddies south of Japan appear consistently with those observed as shown by relatively high correlation coefficients and skills during both runs 5 and 6 (Table 1).
The establishment of assimilation methods applicable to coastal waters will enhance the accuracy of hindcast computations (and hence, the initial condition of forecasts). Besides the assimilation methods, the present forecast model has shortcomings that we should overcome to establish operationally practical kyucho forecast models. In particular, kyucho events that did not occur in the actual ocean often appear in the model (Fig. 9). For instance, the forecast kyucho event in mid-August (c in Fig. 9) is prevented from forming in the actual ocean by the cold eddy moving along the shelf break (Fig. 12). Apparently, this cold eddy is advected by the Kuroshio along the shelf break south of the Bungo Channel, and it is thus considered that short-term fluctuations in the current speed cause the detachment of the eddy in mid-August. If this is the case, the suggestion is that the daily wind forcing and/or tidal currents at the shelf break, which are omitted in the present model, are required to enhance the forecast accuracy.
Kyucho events accompanied by large temperature rises are frequently forecasted during the period May through June when intense kyucho events are absent in the actual ocean (Fig. 9). The forecast computations during this period are conducted using lateral boundary conditions that pass through areas with less Argo data. In fact, runs 5 and 6 using lateral boundary conditions, which undergo assimilating numerous Argo data, result in relatively high and stable correlation coefficients and skills between forecast and reanalysis density profiles and successfully forecast the kyucho events actually observed in August and September. The suggestion is that one of the most efficient improvements in forecasting coastal waters is to deploy numerous Argo floats offshore, but this improvement is beyond a scientific issue.
Professor Lie-Yauw Oey and all other members of the scientific committee provided the authors with the opportunity to present this topic at the 3rd International Workshop of Modeling the Ocean (IWMO-2011), which we greatly appreciated. Comments of two anonymous reviewers are very helpful in improving the manuscript and are appreciated. This work is supported by the Japan Society for the Promotion of Science KAKENHI (21244073).
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.
- Isobe A, Fujiwara E, Chang PH, Sugimatsu K, Shimizu M, Matsuno T, Manda A (2004) Intrusion of less saline shelf water into the Kuroshio subsurface layer in the East China Sea. J Oceanogr 60:853–863Google Scholar
- Koizumi Y, Kohno Y (1994) An influence of the kyucho on a mechanism of diatom growth in Shitaba Bay in summer. Bull Coastal Oceanogr 32:81–89 (in Japanese with English abstract)Google Scholar
- Masumoto Y et al (2004) A fifty-year eddy-resolving simulation of the world ocean—preliminary outcomes of OFES (OGCM for the Earth Simulator). J Earth Sim 1:35–56Google Scholar
- Uda M (1953) On the stormy current (Kyutyo) and its prediction in the Sagami Bay. J Oceanogr Soc Japan 9:15–22 (in Japanese with English abstract)Google Scholar