Daily rainfall variation over the strategically situated western Pacific island of Okinawa is investigated for 1982–2005. Although no trend in the annual rainfall was observed, analyses using a probability distribution function for each year revealed that the frequency of light (0–3 mm d−1) and heavy (26–50 mm d−1) rainfall events have a statistically significant decreasing trend, while extreme (>75 mm d−1) rainfall events have been increasing. Such events are associated with typhoons. An analysis of typhoons passing near Okinawa during 1982–2005 revealed that a cause of the increasing trend in heavy rainfall was increased rainfall amounts per typhoon rather than a rising number of strong typhoons approaching the island. In particular, typhoons which passed to the west of Okinawa, or which approached the island within 100 km, during the peak typhoon month of September showed a statistically significant increase in total rainfall of 6.5 mm y−1 over this time period. This is partially due to a slowing of the typhoon speeds in the latter half of the record. This work supports other recent studies that suggest changes to characteristics of stronger tropical storms are not restricted to the North Atlantic although, owing to the limited length of the record, we are not able to determine at this time whether the observed trends are unusual compared with natural multidecadal variability.
 Since the 1990s, a number of studies have reported increasing rainfall in some areas, apparently owing to global warming [e.g., Houghton et al., 1990; Hennessy et al., 1997; Intergovernmental Panel on Climate Change (IPCC), 2001]. One of these is the East Asian Monsoon area, for which the IPCC  have projected an increase in rainfall, along with an increase in interannual season-averaged precipitation variability. Despite some model divergence for some precipitation characteristics, according to the IPCC , rainfall in the boreal winter is very likely to increase over East Asia and Southeast Asia. Also, rainfall in summer is likely to increase in northern Asia, East Asia, South Asia and Southeast Asia, and extreme rainfall and winds associated with tropical cyclones are likely to increase in East Asia, Southeast Asia and South Asia.
Gu et al.  found that, ignoring ENSO and volcano impacts, the observed rainfall anomaly between 1979 and 2005 in the tropical region of 25°S–25°N, both over the ocean and the ocean plus land, had a statistically significant increasing trend. Over a slightly expanded region (30°S–30°N) for 1979–2003, Lau and Wu  found an increasing trend in the frequency of occurrence (FOC) of light (<1 mm d−1) and heavy rainfall (>20 mm d−1) and a decreasing trend in the FOC of intermediate rainfall (4–14 mm d−1). Light rainfall events mainly occurred over the tropical oceans. Intermediate intensity rain events were found over the warm pool regions and in areas adjacent to the Intertropical Convergence Zone (ITCZ) and the South Pacific Convergence Zone (SPCZ), while heavy rain was seen in the cores of these zones. Consistent with these findings concerning rainfall intensity variation over the tropical and subtropical regions, the Islands of Japan have experienced greater frequency of extreme high rainfall than of extreme low rainfall over the period 1998–2004, apparently influenced by global warming [Japan Meteorological Agency, 2005]. A possible cause of this increase in the tropical rainfall and its intensity is that there is an intensifying hydrological cycle in the tropical region in the warming environment [e.g., Yang et al., 2003; Wentz et al., 2007]. During the last 50 years, sea surface temperatures (SST) over the tropical oceans have been increasing, which is likely to impact significantly on regional atmospheric circulation [Kumar et al., 2004]. Increasing SST has been linked to increases in hurricane potential and destructiveness in the North Atlantic and North Pacific [Emanuel, 2005], and to a general global increase in the most destructive tropical cyclones [Webster et al., 2005]. This change in hurricane intensity could be linked to the increases in extreme rainfall events over the last few decades [Lau et al., 2008]. However, the evidence from the Northwest Pacific is complex. There is strong interannual and interdecadal variability in the number of typhoons in this area, linked to modulation in the El Niño cycle and the Pacific Decadal Oscillation [Chan, 2008; Lau et al., 2008; Chan and Xu, 2009]. Subregions of the northwest Pacific can be used to illustrate some of this variability with distinct decreases in typhoon numbers over SE China during the latter half of the twentieth century [Ren et al., 2006], little trend over the Philippines during the twentieth century as a whole [Kubota and Chan, 2009] and abrupt increases over Taiwan going into the first decade of the twenty-first century [Tu et al., 2009]. When analysis has concentrated on the characteristics of typhoons rather than their number, however, there has been some evidence that fits with the global view of Webster et al.  for an increase in severity. Thus, Rozynski et al. , in a study over 1960–2000, found evidence for an increase in the duration of stronger typhoons from coastal wind measurements in the region. However, Elsner et al.  in a study over the shorter period of 1981–2006 did not find any significant increase in wind speed for any typhoon strength in the area, in contrast to findings for the North Atlantic and Indian Oceans.
 Here we examine rainfall change over 1982–2005 for Okinawa, a small island at 26°N, 126°E in the northwestern Pacific. Okinawa's climate is controlled by the East Asian Monsoon, with its rainy season lasting from May to September, but with a midsummer decline in July [Ikema, 2008]. Being located in the open ocean near the west Pacific warm pool, Okinawa is within the zone of increasing tropical ocean rainfall highlighted by Gu et al. . More importantly, Okinawa is in the middle of the main belt of typhoon tracks in the northwestern Pacific (with about seven encounters per year; see Figure 1). Therefore, it is ideally placed for study of the variation in the impact and number of typhoons on rainfall in the region. The aims of this study are therefore to: (1) assess whether the large-scale increases in rainfall observed from satellites are seen over Okinawa during the period of daily observations (1982–2005); (2) assess whether any changes in Okinawan rainfall magnitude or rate are related to typhoon variability; and (3) determine if this variability is related to the number or strength of typhoons affecting the island.
2. Data and Methods
2.1. Okinawa Rainfall Data
 The basic data used in our study were daily rainfall records from the 14 weather stations operated by the Japan Meteorological Agency and the Okinawa General Bureau in Okinawa, available for 1982–2005 (Figure 2). To investigate rainfall variability at different intensities, probability distribution functions (PDFs) of rainfall according to its intensity (mm d−1) were initially derived as follows. First, we counted the number of days with different rainfall intensity using a 1 mm d−1 interval. Second, these data were divided by the total number of days in the period 1982–2005 to yield the probability of a particular daily intensity, or the frequency of occurrence (FOC). Third, we multiplied the number of days by the rainfall intensities to give an accumulated rainfall amount (ARA) for that intensity (Figure 3). The weather stations on Okinawa were spatially evenly distributed across the island, and so we used the average rainfall data for all 14 sites in our initial analyses. In order to derive clearer signals relating to the rainfall intensity, these data were further grouped into five principal ranges following a standard approach: 0–3 (light), 4–25 (intermediate), 26–50 (heavy), 51–75 (very heavy), and >75 mm d−1 (extreme) [e.g., Osborn et al., 2000].
2.2. Typhoon-Related Rainfall Data
 In order to investigate the complete influence of typhoon occurrence on rainfall trends over Okinawa, typhoons which approached the Island to within 100, 300, and 600 km were defined (Figure 4) using typhoon track data collected from the Digital Typhoon website (http://agora.ex.nii.ac.jp/digital-typhoon/) operated by Japan's National Institute of Informatics.
 Generally, typhoon-related rainfall in this region occurs a few days before and after the peak rainfall event as typhoons approach and then depart the Island system. In this study, rainfall events that occurred 1–2 days before and after the rainfall peak during a typhoon's approach were classified as typhoon rainfall (TR). This is consistent with rainfall data patterns from a number of typhoon-related case studies in this region [Ikema, 2008]. We also further classified TR according to the track of the typhoon, paying particular attention to the closest approach of the typhoon center to Okinawa. In this case, rainfall from typhoons that approached within 300 km of the Island was included as the variable TR300, and from typhoons in the 300–600 km range as TR600. The location of each typhoon at 6 hourly intervals was used to dynamically update these definitions. Typhoons that were further than 600 km from the Island were deemed less likely to be influential, and were not classified as yielding TR. Subtracting TR from all the rainfall data, the remaining will be classified here as nontyphoon rainfall (NTR).
2.3. Statistical Significance Tests
 The statistical significance tests for the trends computed in the study are based on standard linear regression analysis of variance modeling [e.g., Lau and Wu, 2007]. For the tests, no adjustment of degrees of freedom for possible autocorrelation in the residuals has been made [e.g., Wilks, 1995]. Another even more important limitation of the trend tests is that the relatively short data records available (24 years) compared with apparent multidecadal variability of tropical cyclone activity in the basin [e.g., Chan, 2008; Chan and Xu, 2009] does not allow one to determine whether the trends we find are long century-scale trends (as could be associated with greenhouse warming) or are temporary trends associated with multidecadal variability. Nevertheless, we hope to demonstrate that, while neither optimal in terms of data or exhaustive in use, these tests still allow us to elucidate some of the key controls on recent typhoon change in the vicinity of Okinawa.
3.1. General Observations
 For the 1982–2005 period, the highest FOC (Figure 3a) was achieved by the lowest intensity: 0–1 mm d−1 (25.0%). The ARA established that the 2–12 mm d−1 rainfall range contributes the highest amount (>900 mm for each band), indicating that light rainfall contributes most to the total (Figure 3b). The impact of extreme rainfall, over 90 mm d−1, was seen to be significant, with a few occurrences yielding >300 mm of rainfall. Most recorded rainfall events were from NTR (92.4%) and they themselves contributed 77.4% to the total rainfall amount (Figures 5a and 5b). TR300 contributed only 4.3% to the total FOC (Figure 6a), but provided 18.6% of total ARA (Figure 6b). TR600 contributed 3.3% of FOC (Figure 6a), just 1% lower than the TR300 FOC, but to 4.0% of ARA (Figure 6b); so significantly lower than the similar proportion for TR300. Thus, we found that typhoon rainfall could be most clearly represented by TR300, and have therefore used NTR and TR300 to demonstrate the difference in changes in nontyphoon and typhoon rainfall, respectively, over time.
3.2. Long-Term Trends in NTR and TR
 Variations for the annual average rainfall time series over the 14 weather stations on Okinawa are shown in Figure 7. Although the trend was found to be +5.1 mm y−1, this is not statistically significant. In order to remove the impact of rainfall totals and derive clearer signals with regard to extremes in the time series, the rainfall intensity data were grouped into five ranges, as described in section 2. Details of the statistics of these ranges are given in Table 1.
Table 1. Mean and Variance for FOC and ARA of Different Rainfall Bins
Rainfall, mm y−1
Units for FOC are given as percentages and for ARA as mm yr−1 for mean, and their squares for variance.
 For the overall rainfall, including NTR, TR300 and TR600, three rainfall ranges of light rainfall (0–3 mm d−1), heavy rainfall (26–50 mm d−1) and extreme rainfall (>75 mm d−1) show statistically significant trends, all at the 0.05 significance level, except for the heavy rainfall FOC, which is at a weaker, 0.10, level (Figure 8). The FOC indicates that light and heavy rainfall events were decreasing at a greater fractional rate than the increasing occurrence of extreme rainfall. In contrast, for ARA extreme rainfall is seen to have the largest increasing trend compared to decreasing trends of light and heavy rainfall. Note, however, that the balancing rise in days with no rainfall is weakly significant at the 0.10 level. These results indicate the significance of extreme rainfall events within the total rainfall record on Okinawa.
 In their analysis, Gu et al.  found that tropical rainfall over land showed a decreasing trend, but without statistical significance, while the rainfall over the ocean, and the total rainfall (ocean plus land), showed a significant increasing trend. In contrast, here, over Okinawa, we find that only some intensity classes show significant linear trends, rather than the time series as a whole.
 For NTR, two ranges, light (0–3 mm d−1) and heavy (26–50 mm d−1) rainfall, demonstrate decreasing trends in FOC (Figure 9a) and ARA (Figure 9b), with a statistical significance at the 0.05 or 0.10 levels. For TR300, intermediate rainfall (4–25 mm d−1) and extreme rainfall (>75 mm d−1) days were found to exhibit a statistically significant (0.05 level) increasing trend in FOC (Figure 10a). Although intermediate rainfall days (4–25 mm d−1) showed a significant (at the 0.05 level) increasing trend (+2.0 mm y−1) in ARA (Figure 10b), extreme rainfall displayed a considerably greater, if less significant (to 0.10 level), trend (+10.8 mm y−1).
 In summary, these data suggest that there is a general trend in NTR for decreasing rainfall amounts and in TR300 for increasing amounts. This trend in TR300 could be directly linked to regional changes in typhoon characteristics (i.e., frequency, tracks and intensity), and in order to investigate this further we have sought to elucidate key typhoon characteristics in this time period.
3.3. Changes in Typhoon Frequency and Location
 A possible reason for an increase in typhoon rainfall over Okinawa could be that the number of typhoons (or just the number of strong typhoons) approaching Okinawa may have increased, in line with suggestions made by Webster et al. , leading to increases in the FOC of extreme rainfall. To determine if the number of typhoon occurrences had increased over time, typhoons in the TR300 class with strengths, on the Saffir-Simpson scale (see http://www.nhc.noaa.gov/aboutsshs.shtml), of category 4 and 5 within 300 km of Okinawa, were counted for 1982–2005. Overall, the average number of typhoon approaches per year for categories 4 and 5 were found to be 1.4 and 2.4, respectively. This is a real, statistically significant difference at the 0.01 level in the mean annual occurrence of category 4 and 5 typhoons. However, neither the absolute number of typhoons, nor the number of strong typhoons, for either each category or their total, showed any statistically significant trend over this time period [Ikema, 2008].
 Another possible reason for increased typhoon rainfall intensity over Okinawa is that typhoons have been approaching closer to the Island over time, or that typhoons have more commonly passed to the west of Okinawa over time. Winds and rainfall are at their highest intensities around 30–100 km from a typhoon's center [Ogura, 1999]. During the typhoon season around Okinawa, southerlies prevail, and move typhoons northward. With cyclonic air movement around a typhoon, these southerlies enhance wind speed and moist air convergence from the south on the east side of a typhoon. Thus, intense wind and rainfall are concentrated on the eastern side of the eye in the Northern Hemisphere [Pu et al., 2002].
 To investigate this further, the number of typhoons were classified according to: (1) whether they approached within 100 km of Okinawa (TN100), (2) whether or not the typhoon's west side (TW) or (3) east side (TE) touched Okinawa. In order to avoid overlap between these classifications, TW or TE typhoons which closed to within 100km of Okinawa were counted in TN100. When correlated with the annual amount of typhoon rainfall (TR300), TE yielded a higher and more significant correlation (r = 0.43, p = 0.035) than TW (r = 0.20, p = 0.355). TN100 had an even stronger relation and good statistical significance (r = 0.54, p = 0.007), while the total of TN100 and TE produced the highest correlation (r = 0.72, p < 0.001). Therefore, we found that the combination TE + TN100, that is, the number of cyclones passing close to or to the west of Okinawa, is the best measure of typhoon track variability linked to TR300. However, investigation of changes in TE, TN100 and TE + TN100 for 1982–2005 showed that none of these typhoon track categories had a statistically significant linear trend for this period (Figure 11a). Their changes with season also showed no statistically significant trend (Figure 11b).
 We can therefore suggest that our observed trend in extreme rainfall at this location is not due to more typhoons affecting Okinawa in the study period. However, it is worth noting the abrupt change in TN100 seen around 2000 (Figure 11a). We shall return to this result in section 4.
3.4. Changes in Typhoon Rainfall Intensity
 Another explanation for the observed increasing trend in extreme rainfall events is that rainfall per typhoon is increasing, leading to more intense rainfall episodes, as suggested more generally by Lau et al. . Figure 12a shows monthly typhoon rainfall (TR300) from June to September and the total amount for the typhoon season. In Figure 12a, statistically significant increasing linear trends were derived for September (8.6 mm y−1, p = 0.031). This peak month of the typhoon season was the only one with a statistically significant trend, although most months show increases over the last decade. Dividing TR300 by TE + TN100 provided a measure of rainfall amount per most effective typhoon (for rainfall generation; TR300c). These data are shown in Figure 12b, and a statistically significant increasing trend in this variable was also obtained for September (6.4 mm y−1, p = 0.029). This suggests that the increasing trend of typhoon rainfall intensity over Okinawa is most likely attributable to greater rainfall from typhoons passing within 100 km of the island, or to its west, in September.
 With the results indicating a decreasing trend in NTR and an increasing trend in TR300, spatial-temporal changes in these rainfall categories over Okinawa were investigated to study any geographical bias. Figure 13 shows the total amount of NTR (<50mm d−1) and TR300 at the 14 weather stations. The PDFs of NTR (Figure 5) showed its dominance over Okinawa, both for FOC and ARA. This dominance of NTR applies to all the weather stations on the island. To see differences in the amount of NTR (<50mm d−1) and TR300 through 1982–2005, differences between their amounts in the first half of the period (1982–1993) and the second half of the period (1994–2005) were determined by subtracting the latter amount from the former amount (Figure 13). For NTR (<50mm d−1), all the weather stations show negative values, indicating the influence of a decreasing trend in NTR. For TR300, all the weather stations show positive values, indicating an increasing trend of TR300 across the Island. Overall, we can see that even though the topography of Okinawa dramatically varies about the Island, the general trends of decreasing NTR and increasing TR300 are found throughout the Island. The topography is at best a secondary component in this relationship.
 Another possible cause for increased rainfall per typhoon is a decrease in the typical speed of the storms, so that they spend longer near the Island. Average speeds of typhoons while they provided rainfall in Okinawa, that is, within the track for rainfall days counted for TR300, were calculated for the core typhoon season month of September. There is a statistically significant decrease, at the 0.05 level, in speed between the typhoons in the first half of the record (26.7 ms−1; 1982–1993) compared to the second (18.5 ms−1; 1994–2005) (Figure 14). Nevertheless, there is a weakly significant increasing trend, at the 0.10 level, in September TR300 rainfall rate (0.056 mm h−1 y−1) over this whole period.
4. Conclusion and Discussion
 Changes in ordinary and typhoon-related rainfall over the Island of Okinawa for 1982–2005 were detected using daily rainfall and typhoon track data. For NTR, rainfall intensities of 0–3 mm d−1 and 26–50 mm d−1 were found to have a decreasing trend in ARA. For TR300, rainfall intensities of 4–25 mm d−1 and >75 mm d−1 have an increasing trend in FOC. Particularly, extreme rainfall events (>75 mm d−1) showed a clear increasing trend in ARA. Okinawa is a representative island in the main typhoon alley of the western Pacific (Figure 1) and so these results are likely to have more general spatial applicability. They are compatible with the wider remote sensing rainfall trends of Lau and Wu  and also with Chen et al.’s  views of typhoons as the cause of extreme tropical rainfall events away from the ITCZ.
 Despite some evidence for an increasing number of strong typhoons over the Pacific basin as a whole [Emanuel, 2005; Webster et al., 2005], this trend has not been observed around Okinawa, at least for 1982–2005, and so such a trend is not responsible for the increasing trend of heavy rainfall events. However, it was found that the rainfall supply by typhoons for September has been increasing, leading to an increase in the number of extreme rainfall events (>75 mm d−1). This may be related to a slowing of the mean speed of strong typhoons during the period. Note that Wu et al.  have shown a change in the translation speed and a westward shift in track for typhoons in the western North Pacific between 1965 and 1983 and 1984–2003. This is qualitatively consistent with changes in typhoon daily rainfall rate, translation speed and track pattern in the recent warming period. Under the warming ocean environment, a positive correlation is reported between monthly typhoon rainfall and the SST anomaly in the western North Pacific [Rodgers et al., 2000; Chan, 2008; Tu et al., 2009]. This provides a supporting mechanism for increasing heavy rainfall from typhoons. In terms of typhoon formation, most of the initial disturbances from which typhoons later develop originate on the poleward edge of the ITCZ [Aoki, 1985; Nakazawa, 2000; Ma and Chen, 2009]. Thus, a trend toward more heavy rainfall in the ITCZ, as is consistent with Lau and Wu , could enhance moisture content in typhoons, and so supply more rainfall. Some modeling studies also report substantial increases in tropical cyclone rainfall rates in response to projected greenhouse warming over the twenty first century [Knutson et al., 2010]. Although, we found a weakly significant increasing trend in September typhoon rainfall rate alone in Okinawa, further studies are needed to detect a clearer signal of changes in typhoon rainfall, whose drivers may affect not only rainfall rate but also other typhoon characteristics. In addition, further study is needed, particularly with longer data sets, to determine whether the trends identified in our study for 1982–2005 represent a true long-term (e.g., century-scale change such as due to greenhouse warming) or whether they are a manifestation of multidecadal variability that appears as a trend in our relatively short data record.
 This study is compatible with other studies of this region. It shows little sustained trend in typhoon numbers, but with significant interannual variability (Figure 11a), consistent with Chan's view of typhoon behavior in the region [Chan, 2008; Kubota and Chan, 2009; Yuan et al., 2009]. However, as part of interdecadal variability, the abrupt increase after 2000 in the number of typhoons approaching within 100 km of the Island (Figure 11a) is consistent with Tu et al.'s  findings of an extension of the main typhoon track northward, as regional SST has increased.
 Understanding of typhoon change over the northwestern Pacific is clearly not straightforward, and strongly dependent on regionality, with track changes seemingly being responsible for the variety in the findings published in the literature [Chan and Xu, 2009; Tu et al., 2009]. Nevertheless, the rise in SST and the increasing evidence for increases in the impacts of the stronger storms in the region through extreme rainfall and wind changes means that understanding cyclone change in the northwestern Pacific through a combination of the study of tracks and characteristics is the way forward in defining their environmental risks.
 This work formed part of the Ph.D. thesis of T.I. He was funded by the Okinawa International Exchange and Human Resources Development Foundation. We would like to thank four anonymous referees whose comments significantly improved and focused the paper.