W.E. Wright*, B.T. Guan, Y.-H. Tseng, E. R. Cook, K.-Y. Wei, S.-T. Chang

Reconstruction of the Springtime East Asian Subtropical Jet and Western Pacific pattern from a millennial-length Taiwanese tree-ring chronology

Climate Dynamics

W.E. Wright*, B.T. Guan, Y.-H. Tseng, E. R. Cook, K.-Y. Wei, S.-T. Chang

* Laboratory of Tree-Ring Research, The University of Arizona, Tucson, Arizona, , +1-520-954-4766, FAX +1-520-621-8229

Electronic Supplementary Material

I. Justification for radiosonde data truncation at the International Geophysical Year (1958)

Prior to the first International Geophysical Year in 1958, only a relatively small number of locations were recording balloon radiosonde data of the middle and upper troposphere on a daily basis. Pibal (pilot balloon) wind speed data are available for many locations prior to 1958, but above-surface pressure and temperature measurements are not usually available until the beginning of RAOB (rawinsonde/radiosonde) measurements. The inadequate spatial coverage is especially important for studies involving teleconnections, because mathematically accurate upper air data can still yield poor spatial correlation coefficients if insufficient numbers of data points are available for precise estimation of the non-measured locations. Considering a region bounded by 110E-180E and 20N-50N, the region used to produce the index of East Asian Subtropical Jet (EAJ) activity [Park, 2010], we see that data coverage is good back to 1957 over the western part of a region but that the coverage drops quickly prior to that year (Supplemental Figure 1), and prior to 1957 there were only a few radiosonde locations within the latitudinal range of about 29˚N and 35˚N [Park, 2010] where the mean EAJ core exits the Asian continent immediately north of Taiwan in the cool season (Supplemental Figure 2).

II. High Pass Filtering

High-pass filters are not commonly used in dendroclimatology outside of tree-ring chronology production, when some form of high-pass filtering is usually used to remove the biological (non-climate) growth trend(s) in tree ring width time series (Fritts 1976). Detrending for this purpose is often an iterative process wherein low and medium frequency changes that are not common to most of the ring width time series are removed or minimized by the curve fitting. Much rarer is a common tree growth pattern that requires more extreme filtering of the tree-ring data.

MTM spectral analyses of the Taiwan tree-ring data (CIL STD) and of the climate parameters that showed significant relationships in the tree-ring/climate regressions (March-May means of NCEP/NCAR Reanalysis 700 hPa mean temperature, the East Asian Subtropical Jet and the Western Pacific pattern) indicates that the power in those time series is dominantly in the higher frequencies. In addition, MTM spectral coherence analyses revealed that the range of significant frequencies was very similar between the tree ring data and the climate data. Consequently, we investigated high pass filtering of the data to enhance the previously identified relationships. Three methods of high pass filtering were explored: Box (running average), Binomial, and LOESS. The lowest significant frequency in the tree ring and climate data was about 4 years, suggesting that filtering with windows of 4 years or 5 years should be chosen (or the equivalent, for Binomial filtering), but Windows of 2, 3 and 6 years were also calculated to assess the influence of filtering outside the logical window size range. Roughly equivalent treatments using the Binomial method were determined by applying MTM spectral analysis to the results (Supplemental Table I). Each method at a given window was applied to the Taiwan tree-ring standard chronology and to the March-May mean of the Western Pacific pattern, and then the correlation coefficient and coefficient of determination from simple linear regressions of the two time series were calculated to allow comparison of results from the different methods and window sizes. The decision about which method and window size to apply in subsequent analyses was based on the significance of the regression and recognition that the significant frequencies in the spectral coherence analyses covered a range between 2.7 and 3.8 years (Supplemental Table I). Given the upper limit of the significant frequencies, the only high pass filter results considered for subsequent analyses were filters of four years and higher.

Supplemental Table I. Comparison of high-pass filtering methods. The frequencies significant at 95% and correlation coefficients from regression of the CIL STD chronology on the March-May average of the Western Pacific pattern are presented. Also presented is the variance explained by each regression.

Significant frequencies
Method/Window / Tree Ring chronology / March-May Western Pacific pattern / correlation coefficient/
variance explained
Box/ 2 year / 2.3-2.5 / 2, 3, 3.7 / -0.71/0.50
LOESS/ 2 year / 2.3-2.5 / 2, 3, 3.7 / -0.71/0.50
Box/ 3 year / 2.3-2.5 / 2, 3, 3.7 / -0.71/0.50
LOESS/ 3 year / 2.3-2.4 / 2, 3, 3.7 / -0.71/0.50
Box/ 4 year / 2.5, 3.5 / 3.1, 3.8 / -0.64/0.41
LOESS/ 4 year / 2.5, 3.5 / 3.1, 3.8 / -0.67/0.45
Box/ 5 year / 2.5, 3.5 / 3.1, 3.8 / -0.60/0.36
LOESS/ 5 year / 2.4, 3.5 / 3.1, 3.8 / -0.67/0.45
BOX/ 6 year / 2.4, 3.7 / 3.8 / -0.55/0.30
LOESS/ 6 year / 2.4, 3.7 / 3.1, 3.8 / -0.60/0.36
Binomial/ 2 point / 2.5 / 3.1, 3.8 / -0.64/0.41
Binomial/ 3 point / 2.5, 3.8 / 3.1, 3.8 / -0.57/0.32
Binomial/ 4 point / 3.5, 6.1 / 3.8 / -0.49/0.24

3. CIL STD 5YR chronology and CRU TS3.1 regional surface vapor pressure

An additional climate parameter, March-May CRU TS3.1 mean surface (sea level) vapour pressure (CRU SURF VP; 21N-26N, 119E-121E), attained statistical significance against the CIL STD 5YR time series following 5-year high-pass LOESS filtering (Supplemental Figure 3). EAJI and WPI relations to the surface vapour pressure around Taiwan are indicated by significant relationships revealed by regressions of both EAJI 5YR and WPI 5YR on the mean of March-May CRU SURF VP 5YR for the region specified above (WPI- r = 0.60, N = 51, p ≪ 0.0001; EAJI- r = 0.47, N = 51, p < 0.0005). Mean March-May temperatures close to the tree site elevation also show a highly significant relationship to the March-May surface vapour pressure (March-May CRU SURF VP 5YR/ March-May NNR 700 hPa T 5YR; r = 0.72, N = 51, p ≪ 0.0001), but the correlation coefficient from regression of March-May surface temperature on March-May 700 hPa temperature over the same coordinates is much lower (March-May NNR SURF T 5YR/ NNR March-May 700 hPa T 5YR; r = 0.55, N = 51, p ≪ 0.0001), indicating that the significant correlation between the surface (sea level) vapour pressure and the 700 hPa mean temperature is not a spurious relationship. A likely mechanism linking the surface vapor pressure to the 700 hPa temperature and tree growth at the CIL site is the daily orographic lifting of water vapor to the height of the cloud forest (Still et al. 1999), between 850 hPa and 700 hPa, where a persistent cloud layer forms most afternoons (Mildenberger et al. 2009). The mean height of the cloud layer in a cloud forest is determined by a combination of the vapor pressure and temperature of the air mass (Foster 2010), so fluctuations in the surface vapor pressure, the source of the vapour for the cloud layer, can cause changes in the cloud density at the elevation of the trees, with attendant changes in the sunlight available for photosynthesis.

4. Mann-Kendall tests of trend significance on time series of 33-year running variance, following methods designed to account for autocorrelation (Hamed and Rao 1998; Yue and Wang, 2004), indicated significance for running variance in CIL STD 5YR time series produced using LOESS and Box filtering. Box filtering was not used in the reconstructions presented in the main manuscript, but analysis of the running variance is presented as evidence that the trend in the running variance is independent of the method of indexing. Both are plotted in Supplemental Figure 4. Also plotted in Figure 4 are 33-year running variance for the WPI 5YR and EAJI 5YR time series. Running variance of the climate indices are presented to show recent downward trends similar to the recent trend in the CIL STD 5YR chronology.

Hamed KH, Rao AR (1998) A modified Mann-Kendall trend test for autocorrelated data. Journal of Hydrology 204 (1-4):182-196. doi:10.1016/s0022-1694(97)00125-x

Foster P (2010) Chapter 4. Changes in Mist Immersion. In: Bruijnzeel LA, Scatena FN, Hamilton LS (eds) Tropical Montane Cloud Forests. International Hydrology Series. Cambridge University Press, New York, pp 57-66

Fritts HC (1976) Tree rings and climate. Academic Press, London

Mildenberger K, Beiderwieden E, Hsia YJ, Klemm O (2009) CO2 and water vapor fluxes above a subtropical mountain cloud forest-The effect of light conditions and fog. Agricultural and Forest Meteorology 149 (10):1730-1736

Park JS, Jhun JG, Kwon M (2010) Prominent features of large-scale atmospheric circulation during spring droughts over northeast Asia. International Journal of Climatology 30 (8):1206-1214

Still CJ, Foster PN, Schneider SH (1999) Simulating the effects of climate change on tropical montane cloud forests. Nature 398 (6728):608-610

Yue S, Wang CY (2004) The Mann-Kendall test modified by effective sample size to detect trend in serially correlated hydrological series. Water Resources Management 18 (3):201-218. doi:10.1023/b:warm.0000043140.61082.60

Supplemental Figure captions

Supplemental Figure 1. Radiosonde stations by year within the boundaries used by Park (2010) to develop the EAJ index (20N-50N and 110E-180E).

Supplemental Figure 2. Map of radiosonde stations within the boundaries used by Park (2010) to develop the EAJ index (20N-50N and 110E-180E)

Supplemental Figure 3. Significant spatial correlations between the CIL STD 5YR chronology and 5YR LOESS filtered CRU TS3.1 surface vapor pressure. The arrow indicates the position of the CIL tree site.

Supplemental Figure 4. 33-year running variance, CIL STD 5YR LOESS, CIL STD 5YR Box, EAJI 5YR and WPI 5YR. Note the downturn in the late 20th century in all the time series.

Supplemental Figure 1

Supplemental Figure 2

Supplemental Figure 3.

Supplemental Figure 4

4