Dendroclimatological Reconstruction for the Last Sub-Millennium in Central Japan

Annually resolved winter temperature and summer precipitation of Central Japan were reconstructed for the past 800 years, back to AD 1177 from an absolutely dated ring-width chronology of Chamaecyparis obtusa. This chronology was constructed from 300-year-old living trees and old logs of early modern and medieval origins that exist in hundreds in the Kiso Forest on the foothills of Mt. Ontake (3,063 m a.s.l.). In general agreement with the well established past climatic change in Europe, the reconstructed winter temperature showed three distinctively different phases, i.e. a cooling trend toward the mid 1200s possibly corresponding to the termination of the Medieval Warm Epoch, followed by a long cold spell corresponding to the Little Ice Age till the early 1800s, and then by a conspicuous warming trend continuing up to present.


136°E 138°E
(A Dendrochronology sampling sites I Weather stations) Log-top natural regeneration itself is not rare in Japan since, being free from shading by a thick cover of dwarf bamboo commonly encountered throughout the country, seedlings germinating on top of dead logs have a better chance of survival than those regenerating directly on mineral soil, especially when the nursery sapwood is moderately rotten and main tains adequate moisture holding capacity. However, old logs such as those used in this study are rather rare since the wann and moist monsoon climate means that they decay quickly, in a matter of several decades or so. Most probably, the cold subalpine climate, with sub-zero temperatures for nearly half the year (Figure 3 ), shading from sunlight by a dense canopy of 300-year-old Chamaecyparis stands (Sweda et al., 1985) and a thick ground cover of bamboo during the height of the growing season when the monthly mean temperature rises to 18°C might also have helped preserve the Chamaecyparis wood.  (after Sweda and Yonenobu, 1990) ..
The exi�tence of the medieval logs suggests that those of more recent origin had a better chance of being preserved. A search for apparently old logs with decaying surface but without old living trees on top, and subsequent crossdating against 300-year-old living trees revealed some 90 logs of more recent origin. These logs are identified here as 'early modem' to distinguish them from those of older and younger generations, though they match the medieval logs and living trees toward the earliest and latest ends respectively of the temporal spectrum of material.
While all the medieval and early modem logs were collected from a more or less protected alluvial basin at an altitude of 1,480 m a.s.l., the modem logs were obtained from living trees growing in the same stretch of 300-year-old Chamaecyparis forest but at two different sites some JO km away to the northwest at an altitude of 1,550 m a.s.l. on the dividing ridge between the Kiso and Hida Rivers. This choice of modem sampling sites was based on an assumption that trees growing on exposed ridges may well have higher climatic fidelity than those in protected sites (Sweda, 1986;Yamamoto et al., 1986).

Methods
Approximately 80 medieval, 80 early modem and 70 modem logs were collected. These samples were sawn off as disks except for a few medieval logs from which increment cores were taken due to the physical difficulty of sawing. After surfacing and polishing each diSk, ring widths were measured with a semiautomatic ring measurement device under microscopic magnification to a precision of 1/100 mm. Two series of measurements in different radial directions were made for each disk. The two series were then averaged, to give an individual tree chronology, which was subsequently standardized to give a series of ring width indices (RWis). In standardization, a derivative dy/dt =Mk exp (-kt) of the Mitscherlich's growtf:i equation (Sweda, 1984)  (1) was used since this function is known to fit Kiso Chamaecyparis well from our previous work (Yamamoto et al., 1986). In the above process of surfacing and measurement, several old logs unable to withstand physical processing and some dendrochronologically inferior samples failing to comply with our prescribed standard of sensitivity and chronology length (100 years or more) were discarded. This resulted in a total of 63 medieval, 74 early modern, and 69 modem logs that constitute the final synthesized standard chronology shown in Figure  4.

Synthesized Standard Chronology
The synthesized chronology was established by combining absolutely dated individual tree chronologies of the modem, early modem and medieval logs. The modem portion was first established by simply averaging individual modem chronologies. Then the early modem logs were crossdated against the synthesized chronology,. and subsequently added to it. Lastly the medieval logs were dated against the synthesized chronology partially extended with early modem logs, and then incorporated with the other data to complete the 807-year long standard chronology covering AD 1177 to 1983. The contribution of the three different generations of logs in the synthesized chronology is shown in Figure 4 as chronology depth expressed in terms of number of radial measurements involved.
In the above process of absolute dating, crossdating was conducted visually with disks as well as numerically with ring-width data. In visual crossdating, key years characterized as having particularly wide, narrow, dark rings etc. were identified, and their correspondence among disks was established. In numerical crossdating, a series of running crosscorrelations was calculated using shifting overlaps between the standard chronology and each individual log under examination. The latter was dated according to the timing of culminating cross correlation. Results from these two different methods showed good general agreement. In cases of disagreement, however, added weight was given to the results of visual comparisons where we had confidence in them, and the sample was discarded where this was not the case.
Before proceeding to dendroclimatic reconstruction, two major characteristics of this standard chronology have to be mentioned. One of them is the variability in ring width which changes in three distinctively different phases: before AD 1300, after AD 1800 and in-between. The high variability in the earliest phase is simply due to small sample size. On the other hand, the equally large magnitude of variability in the latest phase is attributable to the climatic sensitivity of the modem logs, deliberately sampled from sites on exposed ridges. As a mat t er of fact, the modem logs crossdated extremely well among themselves, whereas this was not the case for the older logs. Individual old logs showed variability of similar magnitude to that of modem logs, but having had grown in a protected basin, synchronous climatic component of variability was rather weak in comparison with randomly occurring inter-tree competition component, making the synthesized chronology complacent in the phase between AD 1300 to 1800. The other characteristic, or more aptly flaw from dendroclimatic point of view, in the synthesized dendrochr o nology is the depression spanning the two decades after 1960. This might be attributable to either a major typhoon in 1959 or the country-wide air pollution of the 1960s or both. In any case, this portion doesn't represent 9 1imatic variation per se and was excluded from the following dendroclimatic analysis.

DENDROCLIMATIC RECONSTRUCTION
Two major difficulties were encountered in the dendroclimatic reconstruction: the sharp recent growth depression referred to above, and the choice of weather stations with which to correlate the standard dendrochronology.
The first difficulty relates to the large non-climatic variability associated with this de pression, which occurs at a crucial time during the limited period of past instrumental climate observation. The data can either be truncated, left as it is, or corrected in some way or an other. Of these three possibilities, we have opted fpr the first, since it seemed most sound and appropriate for this preliminary climatic reconstruction. However, it resulted in a loss of a precious 20 out of only 80 years of instrumental climate observations available concurrently with the standard chronology.
The second difficulty is associated with the fact that, of four weather stations within an 80 km range surrounding Mt. Ontake, no one station represents the �ree-ring chronol ogy location particularly well. Again various approaches might have been adopted, such as choosing the most representative station, or taking a seasonally differentiated average of the stations depending on the seasonal weather pattern over Japan etc. Again, the most straight forward alternative was adopted, and the instrumental records of monthly mean temperature and monthly precipitation from the four stations were simply averaged. As a result, the fol lowing climatic reconstruction is based upon a mere 56-year-long monthly climate' data set running from 1904 to 1959 inclusively, the beginning limited by the availability of climate record and the end by the inadequacy of the dendrochronology.

Response Function Analysis
According to our field observations, tree growth in Kiso begins in April and ends by October, suggesting an appropriate definition of the growth year as beginning in November and ending in the following October, therefore preceding the calendar year by two months as shown in Figure 5.
Year t-2 Growth Year Year t-1 Year t Cale nd ar Ye ar Year t+1 A preliminary analysis using simple linear correlation ( Figure 6) indicates that annual tree growth (RWI) tends to be best correlated with temperature during winter months (De cember to April) and precipitation during summer months (May to August) than with any other combination of climatic factors and seasons as examined for up to three years prior to the growing season (Sweda 1992). Thus, as a first step in determining a response function, multiple linear regression of monthly winter temperatures and summer precipitation on RWI was performed and .the results tested for significance of the partial and multiple correlations.
In the t-and F-tests of the correlation coefficients (Afifi and Clark, 1984), involving a total of 27 predictor variables, i.e. five monthly mean temperatures and four monthly precipitation annually for three years, the significance of the partial correlations proved to be rather in consistent, whereas the multiple correlation was highly significant. This result indicates that the proposed model had too many predictor variables. Thus, in the subsequent analysis, the climatic variables were reduced from monthly to seasonal values. Then the response function of the form: RWI = aRWI' +bRWI" +cTw +dTw' +eTw" + f Ps+gPs' +hPs11 +i (2) Ta.tsuo Sweda. where RWI: the ring width index in growth year t, Tw: winter (December-April) mean temperature of growth year t,

437
Ps: summer (May-August) precipitation of growth year t, with prime (') and double prime (") indicating the previous growth year (t-1) and the year before that (t-2) respectively, and a ,..._, i: the regression coefficients, was iteratively reduced by eliminating non-significant predictor variables in the t-test of significance on individual partial correlation coefficients. As shown in Table 1, the significant variables turned out to be the first-order autocorrelation term, winter mean temperature of the same growth year Tw, and total summer precipitation of the previous growth year Ps'. In Figure 7 the RWireconstructed according to this reduced multiple regression (Model 2) is shown in comparison with the observed counterpart. The agreement is satisfactory as would be expected given the highly significant multiple correlation in Table 1.

Transfer Function Analysis
The dependence of current RWI on winter temperature of the same growth year and the RWI and summer precipitation of the year directly preceding growth indicates that these two climatic variables can be reconstructed from current and preceding RWis. (4) were calculated and tested. In the above multiple regression, the coefficients and error term for precipitation reconstruction were primed (1) to indicate that they should be different nu merically from those for. temperature reconstruction, but not in the sense according to which independent variables are primed in (2). As shown in Table 2, the winter temperature and summer precipitation reconstructed by transfer functions based on the whole period of con current observations revealed satisfactory multiple correlation with the observed counterparts.
For more rigorous verification, the whole period of concurrent observations was divided into early and late halves, and cross F-tests of significance were conducted, in which the climatic estimates reconstructed from each half of the observation were tested against the observed climate over the other half. Although the transfer function based on the late half scored satisfactory, the calibrated over the early half failed to verify significantly as shown in Table   2. A t-test of the significance of the partial correlation (not given in the table) revealed that the contribution of RWI' in multiple regressions (3) and (4) was low, indicating that further reduction in the independent variables was appropriate.
After removing RWI' the same verification procedure was repeated for transfer functions of the form: winter temperature These are no longer multiple regressions but simple ones. The verification is given in Table  3. The transfer function based on the early half again failed to verify significantly, while the correlation based on the late calibration proved to be significant at 1 % level.

Validity of Climatic Reconstruction
The above discussion indicates that the present dendroclimatic reconstruction is not fully valid from a rigorous statistical standpoint. However, judging from the satisfactory agreement between the observed and reconstructed RWis (Figure 7) and the overall agreement between the observed and reconstructed climates (Table 3), we have a good reason to believe that the validity of climatic reconstruction might be improved sufficiently by overcoming the difficulties mentioned earlier.
The major difficulty of limited climate data (only 56 years) could be largely overcome by correcting the recent section of the chronology. It is worth mentioning here that the depressed growth in the apparent chronology is inversely correlated with the pattern of total sulfur oxide emission in the country over the same period. Further experimentation in the choice of weather stations and of seasons to be predicted. would also improve the power of transfer functions.  (5) and (6). Estimates for two variables from one inevitably look alike. Though increase iil temperature should generally result in increases in evaporation, and eventually in precipitation, this is rather extreme.
In view of higher and more consistent correlation of RWI with temperature than with precipitation as seen in Figure 6 and Ta bles 1 through 3, it wou�d be reasonable to consider temperature reconstruction more realistic than precipitation counterpart. The reconstruction can be further improved by incorporating more independent variables such as wood density and earlywood/latewood ratio that may characterize other aspect of variability in tree growth.
Another inadequacy with the present reconstruction results from non homoscedasticity of the original tree-ring chronology mentioned earlier. In other words the complacency in the medieval and early modem chronology has made the corresponding climatic estimates complacent in comparison with the modem portion, which may not have been the case in reality. Correction for homoscedasticity of the original chronology by weighing with variance and sample size will certainly improve the reconstruction.

DISCUSSION
In spite of some shortcomings discusse!i in the preceding section, the reconstructed climate, especially temperature ( Figure 8) generally agrees with the past pattern of climatic change inferred from European data sets (e.g. Bradley andJones 1992, Mamer andKarlen 1984 Particularly noteworthy would be the general warming trend since the mid 1800s and more acute warming since the mid 1900s in association with the greenhouse effects of in creasing atmospheric carbon dioxide (Neftel et al. 1985, Bacastow andKeeling 1981). Most probably the former warming trend may correspond with some lag to increased C02 emis sion from deforestation and industrialization in Europe and North America since the Industrial Revolution, while the latter to accelerated use of fossil fuels by developed countries as well as industrial emergence and development of Asian �ountries, both seemingly triggered by the tennination of the World War TI (Houghton and Woodwell 1989).
Some of· the acute cooling may correspond to major volcanic eruptions, of which Katla in 1179, and Lak.i and Asama in 1783 are the most pronounced. No obvious relationship was found between the reconstructed temperature and solar minima such as Wolf, Sporer and Maunder, however, the tree-rings formed in these periods were complacent and devoid of significantly characteristic rings useful for visual crossdating.
To our knowledge, this is the longest absolutely-dated and annually-resolved dendro climatic reconstruction in Japan and possibly in Monsoon Asia. A more intensive search for older logs in the field may well result in an extension of the ring-width chronology and climatic reconstruction by a couple of hundred years. The quality of the dendroclimatic reconstructions would be improved by the use of data for other tree-ring parameters such as wood density, and early/late wood ratio etc.
One area with potential to produce similar reconstructions to those described here is undoubtedly Taiwan where Chamaecyparis formosensis over a thousand years in age are still available in relative abundance. Similar work would also be possible with similarly old Cryptomeria japoniCa found in Yakushima Island, Southern Japan. Though similar work might also be possible in the rest of Monsoon Asia, the prospect are not so good. Chronologies would be significantly shorter because of the limited availability of good dendrochronological materials in temperate zone, where forests have been depleted due to the long history of heavy human habitation, as well as in subtropical zone where a higher proportion of hardwood species makes chronology construction more difficult.