Impact of observations from Arctic drifting buoys on the reanalysis of surface fields



[1] We investigated the impact of Arctic ice-drifting buoys on an experimental ensemble reanalysis called ‘ALERA’. The ALERA, where the buoy data are assimilated, includes the analysis ensemble mean and spread for each prognostic variable. In the data set, ensemble spreads of surface variables were found to be small only in the regions of densely aggregated buoys. Comparing the ALERA and the data set without the assimilation of surface pressure data observed by the buoys, differences in the ensemble mean and spread between two data sets were locally large, modifying air temperature and winds near the surface. Examining the effect of Arctic-buoy distribution on long-term reanalysis data sets, it was found that the amount of cross-ensemble spreads derived from common reanalysis is very sensitive to the number of buoys. This suggests that data set accuracy might be more vulnerable to deterioration in the near future due to fewer opportunities for buoy deployments over the sea ice.

1. Introduction

[2] The Arctic regions provide challenging environments for data assimilation. Although atmospheric reanalysis products are given as uniformly gridded data sets, in situ meteorological stations, which are difficult and expensive to establish and maintain, are sparsely distributed over the Arctic Ocean, causing analysis uncertainty. Limitations have been found in these data sources at high latitudes [Bromwich and Wang, 2005; Bromwich et al., 2007]. A high-resolution reanalysis project such as the Arctic System Reanalysis ( [Fan et al., 2008] has been initiated; however, a general lack of data over the Arctic Ocean for assimilation probably bears problems concerning accuracy of the data set.

[3] However, the International Arctic Buoy Programme (IABP: has been providing reasonably good surface pressure data in the sea-ice-covered Arctic region since 1979. This programme has assembled meteorological and oceanographic data for real-time operational forecasting requirements, and research. Moreover, many Arctic climate changes were first observed and studied using data from the IABP [Walsh et al., 1996; Rigor et al., 2000]. In addition to supporting such studies, the IABP observations have been essential for forcing, and validating global weather and climate models through assimilation. For example, the buoy data have been assimilated into the National Centers for Environmental Prediction - National Center for Atmospheric Research reanalysis data sets (hereafter NCEP1) [Kalnay et al., 1996].

[4] The IABP buoys are generally deployed during summer and drift with the sea ice. Most of the buoys exit the Arctic Ocean through the Fram Strait by the Transpolar Drift Stream (TDS) [e.g., Inoue and Kikuchi, 2007]. Due to the recent weakened Beaufort Gyre, the broader and faster TDS quickly sweeps buoys away from the Russian coast [e.g., Rigor et al., 2002]. In addition to this, recent abrupt decreases in sea ice have already created difficulties in deploying buoys over the sea ice. With the larger area of open water in summer, there are fewer opportunities to deploy buoys and to obtain surface data for data assimilation, which is also the case in winter. This situation is expected to become more serious in the future, suggesting that analysis error will also increase in the Arctic and deteriorate forecasting skills. Therefore, it is important to investigate the impact of the Arctic drifting buoys not only to better understand the actual atmospheric state but also to clarify errors in the estimated state. For this purpose, we evaluated the effect of Arctic drifting buoys by an observing system experiment.

2. Description of ALERA

[5] ALERA is an experimental reanalysis data set produced by the atmospheric general circulation model (AGCM) for the Earth Simulator (AFES) [Ohfuchi et al., 2004] and a four-dimensional local ensemble transform Kalman filter (LETKF) [Hunt et al., 2007; Miyoshi and Yamane, 2007] system. The ALERA data set was generated in an analysis cycle using a 40-member ensemble forecast with AFES 2.2 [Enomoto et al., 2008] at a T159 (approximately 80 km) and L48 resolution for the period from May 2005 to January 2007. The assimilated observations were adapted from the operational numerical weather prediction system of the Japan Meteorological Agency (JMA) with quality-control flags. Satellite observations, with the exception of satellite-based wind data, were not assimilated. Despite the lack of satellite-observed radiances, Miyoshi et al. [2007a] confirmed that the accuracy of ALERA is comparable to those of the JMA operational analysis and NCEP1 except for the upper atmosphere above 30 hPa and Southern Hemisphere high latitudes. ALERA is open for scientific use and is distributed at

[6] In this study, we regarded the 6-hourly ALERA data as the control run (CTL). When producing ALERA data, patches of uniform grid spacing were used in LETKF for error covariance localization. Contraction of the patches in the zonal direction near the poles caused some discontinuities, particularly in the ensemble spread, due to excessive localization. Although Miyoshi et al. [2007b] have proposed a new version of LETKF that does not require local patches, we used the version with local patches to allow for comparisons with the original ALERA product. As will be shown below, discontinuities did not significantly affect our evaluation of the impact of buoys in the Arctic region. In the test run (ARC), the same observational data set, except without surface pressure observations north of 70 °N, were assimilated using the identical system that produced ALERA. These products include the analysis ensemble mean and analysis ensemble spreads of wind, temperature, humidity, dew point depression, and geopotential height at 17 pressure levels between 10 and 1000 hPa. For this study, we examined the period from June 2006 to January 2007, during which many drifting buoys were deployed just before the International Polar Year (IPY). Further, the sea-ice extent in summer 2006 was relatively larger than that in the summers of 2005, 2007, and 2008. Therefore, this period is suitable for evaluating the impact of Arctic buoys.

3. Effect of Arctic Drifting Buoys

[7] Figure 1a shows the ensemble spread of sea level pressure (SLPsprd) for CTL in August 2006. Because of the dense network of surface meteorological stations over land, the SLPsprd was extremely small in those areas (less than 1 hPa). In contrast, SLPsprd over the Arctic Ocean was generally much larger (more than 2 hPa) than that over land. However, over the Beaufort Sea, where there were relatively numerous drifting buoys on multi-year ice, the SLPsprd was relatively small. These characteristics were also found in other variables (e.g., air temperature and wind speed) and other months. The ARC results show large SLPsprd over the Beaufort Sea (Figure 1b). The larger SLPsprd in the eastern Arctic suggests that buoy data improved the analysis accuracy both locally and over the entire Arctic region.

Figure 1.

Analysis ensemble spread of SLP in August 2006 for (a) CTL (b) and ARC. Dots depict the positions of the Arctic drifting buoys. The thick line denotes ice concentration grater than 15%. The area enclosed by the dashed line is used in Figure 2.

[8] Figure 2a shows a time series of the difference (ARC − CTL) in the ensemble spread of SLP (ΔSLPsprd) averaged over a dense-buoy area (the area is shown in Figure 1). The ΔSLPsprd was always positive (1–3 hPa) during the analysis period, suggesting that buoy data are effective in maintaining SLP accuracy in all seasons, even though the buoys drift. The stepwise increase in ΔSLPsprd at the beginning of August would have been caused by the deployment of six additional buoys in this region during the month (shading in Figure 2a). This result confirms that densely located drifting buoys contribute to more accuracy in SLP. As a whole, the difference (ARC − CTL) in the SLP ensemble mean (ΔSLPmean) had a negative bias (bars in Figure 2b) with larger amplitude in fall and winter than in summer partly due to the enhanced Beaufort high and frequent passage of cyclones in fall and winter. However, there was no clear correlation between ΔSLPmean (Figure 2b) and ΔSLPsprd (Figure 2a) although the correlation between ΔSLPmean and SLPmean itself (black line in Figure 2b) was high. This result suggests that without the buoy data, the spatial patterns of high and low pressure systems over the Beaufort Sea tend to be smoothed. The smoother SLP field (and higher pressure levels) would dampen the pressure gradients and winds.

Figure 2.

Time series of (a) ΔSLPsprd (black line) and monthly number of buoys reporting on the GTS in the area (shading), (b) ΔSLPmean (bars) and SLPmean for CTL (black line), and (c) ΔT1000mean. Averaging was done over the area shown in Figure 1 (75°N–85°N, 130°W–180°W).

[9] The negative bias of ΔSLPmean should result in modulated air temperature near the surface. Figure 2c shows a time series of the ensemble mean difference of air temperature at 1000 hPa (ΔT1000mean). During summer, ΔT1000mean was not large (∼0.5 K), likely because the air temperature over the ice surface during the melting season tends to remain near freezing. This feature could be verified even in August 2006, when many drifting buoys were intensively deployed for the IPY (Figure 2a). After summer, however, ΔT1000mean showed a positive bias (approximately 1 K as a monthly average) when ΔSLPmean had a negative bias associated with the Beaufort high (black line in Figure 2b). This warm bias would likely result in underestimation of surface cooling if these data were used as forcing data in a coupled model, because fall is the beginning of the freezing season.

[10] In order to understand how ΔSLPmean affects wind fields, we examined the relationship between ΔSLPmean and SLPmean for CTL. Figure 3a shows a significant negative correlation is evident over the Beaufort Sea, and an area of positive correlation is clearly visible over the east Siberian Sea, suggesting that a situation with large ΔSLPmean occurs under such a dipole SLP pattern. Because events with negative ΔSLPmean occurred frequently (Figure 2b), a western-low and eastern-high SLP pattern was dominant. This condition is a typical SLP pattern during winter, and enhances sea-ice drift over the Transpolar Drift Stream. A large negative ΔSLPmean should modify wind fields in the areas with strong SLP gradients. Figure 3b shows correlation (shading and contour) and regression (vector) maps of the difference between wind fields at 925 hPa (ΔU925mean and ΔV925mean) and ΔSLPmean. The weakened Beaufort high with negative ΔSLPmean suppressed the clockwise wind force. If these forcing data were used in coupled ice-ocean models, sea-ice motion would be damped in those areas. Assuming that the free drift of ice is 1% of wind speed, an anomalous wind speed of 1 m s−1 would lead to an ice-drift error of ∼200 km over 7 months.

Figure 3.

(a) Correlation coefficients of the SLPmean for CTL with the ΔSLPmean in Figure 2b from June to December 2006. The contour line indicates the SLPmean. (b) Correlation and regression coefficients of ΔU925mean (contour) and ΔV925mean (shading) with the ΔSLPmean at the 90% or greater statistical significance level.

4. Summary and Discussion

[11] Data impact on the ALERA was assessed by comparing data sets with (CTL) and without (ARC) the assimilation of Arctic drifting buoy data. Relatively small SLP spread was found in CTL over the Arctic Ocean during the analysis period of densely distributed buoys, reducing the uncertainty in SLP projection. Although very limited satellite data assimilation could possibly magnify the impact of withholding the buoy data, difference between ARC and CTL is valid to gain an insight into the impact of buoys on the surface fields. Comparison between ARC and CTL revealed a negative bias of ΔSLPmean over the Beaufort Sea that locally modified air temperature and winds near the surface. An incidental finding is that an increase in the number of buoys likely influenced ΔSLPsprd, suggesting that a sustainable Arctic buoy network is indispensable for tracking and interpreting the ongoing Arctic climate change. Our results are presumably exaggerated to a certain extent because an extreme approach by an all-or-none comparison was performed. However, a situation of lower number of ice-drifting buoys is overhanging due to fewer opportunities to deploy them during summer and due to accelerated ice drift.

[12] Arctic buoys reporting on the Global Telecommunication System (GTS) have been deployed since 1979. Therefore, it is possible to determine how the number of buoys affects surface data sets. To assess the relationship between cross-ensemble spreads derived from reanalyses and the number of buoys from a long-term climatological perspective, we selected four atmospheric reanalysis data sets: NCEP1; NCEP-Department of Energy (DOE), hereafter NCEP2 [Kanamitsu et al., 2002]; the European Centre for Medium-Range Weather Forecasts (ECMWF) 40-year reanalysis (ERA-40) [Uppala et al., 2005]; and the Japanese 25-year Reanalysis (JRA-25) [Onogi et al., 2007]. These data sets are available on a global 2.5° × 2.5° latitude-longitude grid. Here, we examined the data sets for the period of January 1979 to August 2002. Even though these data were constructed with assimilation models, the quality of the reanalyses vary with time (e.g., due to the number of buoys). We calculated a regression map of the cross-ensemble spread by using monthly data sets and the number of Arctic buoys north of 70 °N (Figure 4a). The area with a high buoy concentration clearly correlated negatively with the magnitude of the cross-ensemble spread, i.e. the spread decreases as the number of buoys increases. Along the Eurasian coastal region, the relationship was very weak because few buoys were deployed initially.

Figure 4.

(a) Regression of cross-ensemble spread derived from NCEP1, NCEP2, ERA-40, and JRA-25 reanalyses with the number of IABP buoys located at latitudes higher than 70°N (unit: hPa per 10 buoys). Shading indicates statistical significance at the 90% or greater level. September buoy positions before and after August 2002 are shown by red and black crosses, respectively. (b) Time series of the number of IABP buoys reporting on the GTS. Dots indicate the data for September. The data on the black line (from September 2002) were not used for calculation of the regression map in Figure 4a.

[13] After August 2006, buoy numbers increased from 25 to 60, partly because of the start of the IPY (black line in Figure 4b). The spread should be decreasing in the western Arctic; however, buoy network is becoming compacted (e.g., in the eastern Beaufort Sea) partly due to large scale changes in wind conditions during winter and fewer opportunities to deploy buoys from the pack ice during summer. We must consider these issues and proceed with caution when using reanalysis results for the Arctic.

[14] We have shown that the observations provided by the IABP are vital for analysis of accurate meteorological fields. The IABP has been developing and deploying buoys that may be deployed in the open ocean during summer and survive freeze up during fall, which obtains sustainable meteorological data from the ‘Blue Arctic’.


[15] Comments from I. G. Rigor and anonymous reviewers were very helpful. ALERA, which was produced by the JMA, JAMSTEC and Chiba Institute of Science, is used in this study. We used the Earth Simulator under support of JAMSTEC. The authors express their appreciation to staff at NCEP, ECMWF, JMA for making freely available the reanalysis data sets. The buoy data were obtained from the web site of the Department of Fisheries and Oceans, Canada, and posted on the GTS by participants of the IABP.