Evaluating Spatiotemporal Patterns and Trends of Drought in Japan 1 Associated with Global Climatic Drivers

: Drought disasters, such as water scarcity and wildfires, are serious natural disasters in 9 Japan that are also affected by climate change. However, as drought generally has widespread 10 impacts and the duration of drought can vary considerably, it is difficult to assess the spatiotemporal 11 characteristics and the climatic causes of drought. Therefore, to identify the drought homogeneous 12 regions and understand climatic causes of regional drought over Japan, this study provides a 13 spatiotemporal analysis for historical droughts patterns and teleconnections associated with global 14 climatic drivers. The trends of meteorological elements, which are the basis of drought index 15 calculation, was first assessed. Then, drought characterized by the Self-calibrating Palmer Drought 16 Severity Index (scPDSI) was investigated. Trends and patterns of drought were identified through 17 the trend-free pre-whitening Mann-Kendall test and distinct empirical orthogonal function. The 18 continuous wavelet transform and cross wavelet transform together with wavelet coherence were 19 utilized to depict the links between drought and global climatic drivers. The results are described as 20 follows: (1) the trends of precipitation were insignificant. However, temperature and potential 21 evapotranspiration increasing trends were detected over Japan; (2) the drought trend over Japan 22 varied seasonally, increasing in spring and summer and decreasing in autumn and winter; (3) two 23 major subregions of drought variability—the western Japan (W region) and most of the 24 northernmost Japan near the Pacific (N region) were identified; (4) wildfires with large burned area 25 were more likely to occur when the scPDSI was less than -1; and (5) the North Atlantic Index (NAOI) 26 showed the strongest coherence connections with Distinguished Principle Components-1 among four 27 climatic drivers. Additionally, Distinguished Principle Components-2 showed stronger coherence 28 connections with NAOI and Arctic Oscillation Index. This study is the first to identify homogeneous 29 regions with distinct drought characteristics over Japan and connect the drought in Japan with the 30 global climatic drivers. 31 this we adopted TFPW-MK to analyse the time series trends. The specific details TFPW-MK found Yue et The main are listed as follows: between PDOI and DPC2 during the same period. Besides, the leading relationship of 442 time-frequency became 14~20 months between 2002 and 2010. The correlation between ONI and 443 drought was relatively weak compared with the other three climatic factors. The results indicated 444 that the ONI (Figure 14h) led the DPC2 by 14~18 months from 1995 to 2010.


Introduction
. Contrasted with permanent aridity in arid areas, drought is a 36 temporary reduction in precipitation or water availability over an extended period (Hisdal, 2000) and 37 can last for months or years. Considering the duration, intensity, geographical extent, and broad 38 effects of droughts, it is difficult to determine drought occurrences and their effects (Asong et al., 39 2018). Additionally, drought is considered to be the most complex and impenetrable extreme climate 40 event, affecting more people than any other natural hazard (Hagman, 1984 (Mishra and Singh, 2010). In particular, the deficit in 44 precipitation combined with high evapotranspiration losses during a drought would increase the risk 45 of other disasters such as wildfires (Sarris et al., 2014). The risk of water scarcity would be 46 exacerbated by long-term drought, which poses challenges to water resources management (Iglesias 47   To analyse the key global climatic drivers of drought events over Japan, the following four 159 global climatic indices were investigated. The Arctic Oscillation Index (AOI) (Li et al., 2003) and 160 North Atlantic Oscillation Index (NAOI) (Li et al., 2003) 167 The Mann-Kendall (MK) test, which was proposed by Mann (Mann, 1945) and modified by 168 Kendall (Giglio et al., 2015), is widely used for analysing the change trends in hydrometeorological 169 time series (Liu et al., 2015;Yue et al., 2003aYue et al., , 2003b. The advantage of the MK test is that the time 170 series does not require any special form for the probability distribution function, which means it is 171 less sensitive to potential interference from outliers in the data (Serrano et al., 1999). However, this 172 test requires that the data should be independent. Some hydrometeorological time series may usually 173 display serial correlation. This will increase the probability that the MK test detects a significant 174 trend, altering the magnitude estimate of serial correlation . To efficiently eliminate 175 the effect of the serial correlation on the MK trend test,  proposed the trend-free 176 pre-whitening MK (TFPW-MK) test. Before the MK test, the time series is first detrended and 177 pre-whitened. In this paper, we adopted TFPW-MK to analyse the time series trends. The specific 178 details of TFPW-MK can be found in . The main steps are listed as follows: 179

Trend-Free Pre-whitening Mann-Kendall test
Step 1. Using the Theil-Sen approach (Sen, 1968) estimates the slope b of the trend in the time 180 series. If the slope is equal to zero, then it is unnecessary to continue conducting the trend analysis. 181 https://doi.org/10.5194/nhess-2020-416 Preprint. Discussion started: 6 January 2021 c Author(s) 2021. CC BY 4.0 License.
If the slope differs from zero, then it is assumed to be linear, and the time series are detrended by the 182 following equation: 183 Step 2. The lag-k serial correlation coefficient rk of the detrended series Xt ' is computed using 185 Equation (7), and then, Autoregressive (k) (AR) is removed from Xt ' by Equation (8). 186 This pre-whitening procedure after detrending the series is referred to as the TFPW procedure. 189 After applying the TFPW procedure, the time series should be independent. 190 Step 3. The identified trend Tt and the residual Yt ' are blended by the following equation: 191 Step 4. The MK test is applied to the blended series to assess the significance of the trend. We 193 can obtain the statistic Z through the calculation of the TFPW-MK test and measure the degree to 194 which a trend is consistently decreasing or increasing. In the bilateral trend test, if |Z|＞Z1-α/2 is at a 195 desired confidence level α, the original hypothesis is unacceptable; that is, the time series trend is not 196 statistically significant at the 1-α confidence level. The confidence levels of 0.05 and 0.1 are equal to 197 the Z values of 1.96 and 1.64, respectively. Thus, the trend can be classified according to the Z value 198 (

201
The empirical orthogonal function (EOF), which deals with temporal and spatial functions, is 202 used to extract the spatiotemporal modes based on the data variance representations. The EOF was 203 introduced into meteorology and climate research by Lorenz (1956)   represents the signal energy at a specific scale (period) and time (Asong et al., 2018). In this paper, 224 the time-frequency domain of DPCs was analysed by CWT. The specific calculation process for 225 CWT can be found in Torrence et al. (1998). Notably, the CWT brings about a cone of influence 226 (COI) that delimits a region of the WPS beyond which the edge effects become significant and the 227 power could be suppressed (Torrence et al., 1998). 228 https://doi.org/10.5194/nhess-2020-416 Preprint. Discussion started: 6 January 2021 c Author(s) 2021. CC BY 4.0 License. Also, the cross wavelet transform (XWT) (Torrence et al., 1998) and wavelet coherence (WCO) 229 (Torrence et al., 1999) can examine the relationship between the DPCs and the global climatic 230 driving factor. WCO reveals local similarities between two time series and may be found to be a 231 local correlation coefficient in the time-frequency plane; that is, their possible teleconnection can be 232 identified by WCO (Asong et al., 2018). Similar to the CWT, the parts outside of the COI should 233 also be interpreted with caution. The specific XWT and WCO analysis methods can also be found in 234 Torrence et al. (1998;. 235

237
To better understand the drought trends in Japan, the variation characteristics of meteorological 238 elements (precipitation, temperature, and potential evapotranspiration) were first analysed by 239 anomaly curves. Figure 1 shows the anomaly curves of precipitation, temperature, and potential 240 evapotranspiration, which reflects the difference between the annual value of meteorological 241 elements and the overall mean over Japan. For precipitation, there was a weak increasing trend. Next, the TFPW-MK trend test results were calculated and then interpolated over Japan by 259 inverse distance weighted (IDW) (see Figure 2). For precipitation, except for some parts of the 260 northernmost region, Japan showed increasing trends. However, all the precipitation trends were 261 insignificant. For temperature, the whole of Japan showed strong increasing trends. The trends in 262 northeast Japan was most significant. For potential evapotranspiration, the trends of potential 263 evapotranspiration were similar to that of temperature. It was evident that almost all of Japan showed 264 increasing trends in potential evapotranspiration, except for several parts of the northernmost region. 265 The trends of potential evapotranspiration in most of eastern Japan were insignificant and only 266

272
In addition, in order to reduce the impact of using a single data on the robustness of the results, 273 the trend analysis results of different meteorological elements using different datasets were 274 compared (As shown in the Appendix Ⅰ). The results showed that the CRU dataset used in this paper 275 was proven to be credible. 276

277
For drought, the average scPDSI series of Japan from 1960 to 2018 is shown in Figure 3. A 278 drought month event is defined to occur when the scPDSI is less than -1 (Ye et al., 2019a). The 279 results indicated that the two driest periods occurred in 1983~1988 and 1994~1997. In these two 280 periods, the minimal values of scPDSI were -3.80 and -3.42, which both reached the level of 281 severely dry, as shown in Table 1 For the corresponding drought temporal characteristics, the DPC scores are displayed in Figure  340 9. The DPC1 scores showed a decreasing trend, which meant that the W region became drier. In a wet spring, when the scPDSI was positive, the burned area of western Japan was less than 381 100 ha. The three springs with severe wildfires, when the burned area was larger than 300 ha, were 382 accompanied by drought events in which the scPDSI was less than -1. Although there were fewer 383 wildfire occurrences in the N region than in the W region, these two regions followed a similar 384 pattern. A total of six wildfires with burned areas of over 60 ha occurred in the N region. The scPDSI 385 values corresponding to these six wildfires were all negative, and four of them experienced drought 386 (scPDSI ≤ -1). When the scPDSI was more than 1, there were only six wildfire occurrences in the N 387 region, and the burned area was less than 60 ha. Whether in the W or N region, there was, to a certain extent, a relationship between the wildfire 389 burned area and drought. Indeed, the drought did not necessarily lead to large wildfires, but a lack of 390 soil moisture could increase the risk of severe wildfires. Understanding the effects of drought on 391 other natural disasters could encourage scholars to pay more attention to drought research in Japan.

395
Due to the complexity in the causes of drought, the relationship between drought and global 396 climatic drivers needs to be discussed. Figure 13  For the W region, Figure 14a~b shows that sporadic but significant coherence was found 422 between DPC1 with AOI, NAOI, PDOI, and ONI. For AOI (as shown in Figure 14a), a positive 423 correlation occurred between approximately 96 and 128 months from the 1970s to the mid-1990s, 424 while a temporary negative correlation was found during 1992~2000 in the range of 20 to 32 months. 425 The coherence between DPC1 and NAOI is shown in Figure 14b. It was obvious that the NAOI led 426 the DPC1 in phase by approximately 84 to 112 months from the 1980s to 2000s. However, over the 427 1995~2010 period, DPC1 lagged the NAOI, ranging from 36 to 72 months. In Figure 14c, the 428 directions of the arrows were somewhat messy, meaning that the coherence between PDOI and 429 DPC1 was slightly difficult to determine. But a relatively obvious relationship was that DPC1 lagged 430 PDOI by approximately 8 to 16 months from 1992 to 2005. Also, three strong coherence bands were 431 observed in Figure 14d, one of which was nearly a positive correlation between the DPC1 and ONI 432 in the 1970s. And the ONI led the DPC1 by approximately 12 to 16 months in the 1998~2010 period. 433 From 1998 to 2000, the coherence band was mainly concentrated over 140~168 months. 434 The coherence was stronger in DPC2 than in DPC1. As shown in Figure 14e, a band of 435 approximately 112~128 months of high energy was observed during the period of the mid-1970s to 436 mid-2000s, while the regions beyond the COI were ignored due to edge effects. In this period, the 437 coherence was initially unstable, but after 1985, the relationship between the DPC2 and AOI 438 gradually showed positive correlations. The NAOI (Figure 14f The foregoing analysis has shown the teleconnection between drought over Japan and global 451 climatic drivers. This paper mainly focused on qualitative analyses to determine the climatic driving 452 factors that have the most obvious impact on drought rather than focusing on quantitative analyses. 453 For DPC1, the most significant coherence between DPC1 and NAOI occurred from the 1980s to 454 2000s, and the dominant frequency of DPC1 appeared in 1985~2005, which suggested that the 455 periodic features of drought in the W region were likely to be affected by the NAO. For the N region, 456 the significant coherence between drought with the AO and NAO appeared throughout the research 457 period, but because of the edge effects, the coherence analysis mainly revolves around the parts in 458 the COI. This finding also showed that the complexity in the cause of the drought was affected by 459 more than one climatic driving factor. The CWT and WTC analysed the relationship between 460 drought and global climatic driving factors from only a statistical point of view, so the underlying 461 physical process was not the target of this research. However, this paper would be meaningful for 462 determining the climatic driving factors that affect the occurrence of drought, and in-depth research 463 on predicting droughts through climatic driving factors is required. 464 Additionally, this paper identified the two global climatic drivers, AO and NAO, which had the 465 most significant effects on drought over Japan. But the effects of these two global climatic drivers on 466 regional climate is not limited to Japan. Actually, in the Korean Peninsula across the sea from Japan, 467 the spring drought events in this region have also been confirmed to be affected by the NAO (Kim et 468 al., 2017). Also, the majority of UK recorded droughts in recent history showed a clear relationship

479
Investigation of the coherence connections between drought and global climatic driving factors 480 is significant for a better understanding of drought. This paper focused on Japan, which has less 481 available drought-related research, as the research area and provided a comprehensive analysis of 482 drought patterns over Japan during the period from 1960 to 2018 using the scPDSI drought index. 483 The relationship between wildfire burned area and the scPDSI was analysed. Wavelet analysis was 484 applied to detect the climatic driving factors that had the strongest relationship with drought, which 485 overcame the insufficiency of classical drought analysis in determining the cause of the drought. 486 The main conclusions obtained from this paper are summarized as follows: (1) The potential 487 evapotranspiration and temperature showed increasing trends throughout almost all of Japan, while 488 the changing trend in precipitation was not significant. The changes in potential evapotranspiration 489 and precipitation were most obvious in summer, whereas there was little difference in the 490 temperature in different seasons. (2) On average, 1983~1988 and 1994~1997 were the two driest 491 periods in Japan. Also, the droughts were greater in spring and summer and weaker in autumn and 492 winter. (3) DEOF was used to identify two major subregions of drought variability-the western 493 region (W region) and most of the northernmost region near the Pacific (N region). The 494 corresponding scores of DPC1 and DPC2 showed a trend of decreasing (increasing in drought) and 495 increasing (decreasing in drought), respectively. (4) When scPDSI was less than -1, wildfires with 496 larger burned areas were more likely to occur. (5) The global climatic driving factors that showed the 497 strongest cross-correlation with DPC1 and DPC2 were the NAOI and AOI together with the NAOI, 498 respectively, and their corresponding coherence times were 1980~2010 and 1975~2005, respectively. 499 The outputs of the paper provide a reference for the future study of drought prediction in Japan. 500 Through the identification of drought subregions with similar drought spatiotemporal characteristics, 501 it can be helpful for drought risk management at the regional scale over Japan. The analysis of the 502 climatic causes of drought in these subregions can be useful for choosing suitable impact factors for 503 drought predictions. 504  523 Appendix Ⅱ: Based on the data mentioned in Appendix Ⅰ, the scPDSI based on the different datasets 524 were also compared during the period of 1958~2012 (As shown in Figure A4).