Dendroclimatic reconstruction of precipitation for sites in the southern

The Holocene 11,2 (2001) pp. 203-213

Dendroclimatic reconstruction of

precipitation for sites in the southernCanadian Rockies

Emma Watson and Brian H. Luckman

(University of Western Ontario, Department of Geography, London, OntarioN6A 5C2, Canada)

Received 30 June 1999; revised manuscript accepted 21 May 2000

AHOLOCENERESEARCHPAPER

Abstract: Seven Douglas fir (Psetudotsuiga menziesii) tree-ring chronologies were developed from sites in Banffand Jasper National Parks, Alberta, and near Cranbrook, British Columbia. The chronologies ranged in lengthfrom 191 to 691 years and display a strong precipitation signal. Multiple regression models were developedthat calibrate >50% of the variance in the instrumental records and were used to reconstruct annual (pAugust-July or pJuly-June) precipitation for Banff, Jasper and Cranbrook. Comparison of these reconstructions withrecords from adjacent areas indicates that periods of reduced precipitation occurred over this region in c. AD1760-1775, the 1790s, the 1840s-1870s, the 1890s and c. 1920-1945. Periods of significantly greater precipi-tation occurred during the 1 880s-1920s and in the latter half of the twentieth century. The Banff reconstructionextends back to AD 1430 and also shows major drier intervals c. 1470-1510, the 1570s and 1615-1660. Thesereconstructions indicate regionally coherent precipitation patterns that fluctuate on decadal timescales and maybe linked to changes in atmospheric circulation pattems.

Key words: Dendroclimatology, tree-rings, precipitation reconistruction, precipitation variability, Pseudotsugamnenziesii, Canadian Rockies.

IntroductionAlthough the first dendrochronological studies in the CanadianCordillera utilized valley-floor, moisture-sensitive sites (Schulman,1947; Drew, 1975), little subsequent work has been carried outin this environment. Studies elsewhere in North America havedemonstrated that lower forest border sites are well suited toprovide proxy-records of precipitation (Fritts, 1974; Case andMacDonald, 1995; Woodhouse and Meko, 1997). RecentlyCook et al. (1996; 1999) have used a dense network of precipi-tation-sensitive chronologies to develop a gridded drought(Palmer Drought Severity Index (PDSI)) reconstruction for theentire continental United States. Unfortunately, no equivalentdata exist for Canada. The need for a greater understanding ofprecipitation variability and the necessity to develop palaeocli-matic time-series to evaluate climate change have been ident-ified as 'urgent scientific problems requiring attention'(Houghton et al., 1996: p. 46). Climate predictions based uponthe relatively short instrumental climate records may be unre-liable as the records may not represent the full range of themeasured climatic variable (Woodhouse and Overpeck, 1998).Precipitation reconstructions derived from tree-rings offer con-siderable potential to further our understanding of precipitationvariability, particularly for mountain areas where precipitation

© Arnold 2001

is highly variable and measurement stations are sparsely dis-tributed. This paper presents tree-ring-based annual precipi-tation reconstructions for two sites in the Alberta Rockies(Banff and Jasper) and for Cranbrook, British Columbia(located in the adjacent Rocky Mountain Trench).

Chronology developmentPrevious workExploratory dendrochronological work in the southern CanadianCordillera was carried out by Edmund Schulman in the 1940s(Schulman, 1947). Subsequently, scientists from the Laboratoryof Tree-Ring Research in Tucson (LTRR) collected Douglas firchronologies from Banff (Powerhouse, Tunnel Mountain andExshaw), Jasper (Pyramid and Patricia Lake), Kamloops and Pen-ticton (plus Pinus Ponderosa chronologies from the latter twoareas) as part of the first large network of tree-ring sites in westernNorth America (Drew, 1975; Fritts, 1991; Table 1). These LTRRchronologies extend only to 1965, generally have low sampledepth and poor replication in the earliest parts of the record. TheBanff Powerhouse site was resampled and a short ring-width anddensity chronology developed from 12 cores by Robertson andJozsa (1988).

0959-6836(01)HL457RP at Ebsco Electronic Journals Service (EJS) on December 18, 2008 http://hol.sagepub.comDownloaded from

http://hol.sagepub.com

204 The Holocene 11 (2001)

Table 1 Summary statistics for selected tree-ring chronologies from the southem Canadian Cordillera

Site name Lat. Long. Elev. Years n3 r % Mean' Auto Std. Mean(m) Missing sensitivity corr. dev. ring Coeff. of SSS2

width variation >0.85

1997 COFECHA chronologies (Douglas fir)Prairie de la Vache 52048' 117058' 1150 1768-1996 29 0.83 0.06 0.34 0.64 0.56 1.10 0.51 1781(4)4Pyramid Lake 52055' 118006' 1200 1672-1996 27 0.86 0.05 0.42 0.62 0.55 0.89 0.62 1710(5)Maligne Canyon 52057' 117057' 1200 1730-1996 26 0.78 0.00 0.34 0.72 0.67 1.16 0.58 1780(6)Lake Annette 52055' 117058' 1150 1796-1996 16 0.76 0.00 0.32 0.61 0.65 1.37 0.47 1817(3)Powerhouse 51012' 115031' 1432 1306-1996 53 0.77 0.06 0.33 0.72 0.35 0.58 0.61 1430(4)Perry Creek 49034' 115050' 760 1745-1996 22 0.73 0.00 0.29 0.74 0.54 0.93 0.58 1780(5)Wasa Lake 49047' 115045' 610 1762-1996 34 0.79 0.00 0.33 0.67 0.74 1.40 0.53 1826(3)

LTRR COFECHA chronologies (Douglas fir)Pyramid Lake 52055' 118006' 1128 1630-1965 16 0.83 0.00 0.42 0.58 0.54 0.90 0.60Patricia Lake 52054' 118005' 1128 1540-1965 36 0.81 0.02 0.42 0.64 0.61 0.91 0.67Powerhouse 51012' 115031' 1432 1410-1965 21 0.82 0.14 0.37 0.73 0.30 0.47 0.64Banff 51010' 115033' 1310 1460-1965 47 0.79 0.30 0.41 0.66 0.36 0.58 0.61Exshaw (CRN)6 51004' 115011 1310 1560-1965 22Tunnel Mountain 51010' 115033' 1402 1570-1965 25(CRN)

Mean statistics N5Douglas fir (this table) 11 0.80 - 0.36 0.67 0.53 0.94 0.57Englemann spruce (St George, 1997) 21 0.63 - 0.18 0.81 0.32 0.71 0.45Alpine larch (Colenutt, personal communication; 19 0.77* - 0.36 0.65 0.22 0.44 0.50*Luckman etal., 1995)Limber pine (Case and MacDonald. 1995) 3 - - 0.26 0.54 0.33 0.74 0.45Whitebark pine (Youngblut, 1999) 3 0.56 - 0.18 0.84 0.19 0.41 0.46

'For definition, see Fritts (1976).2Subsample signal strength (SSS) is used to define the portion of a chronology with a strong common signal for dendroclimatic studies (Wigley et al.,1984; Briffa and Jones, 1990).n3 = Number of cores.4The number of trees required to attain an SSS of 0.85.N5 = number of chronologies.6Only available as standardized chronology.

New chronology developmentIn 1997, we collected 16 new tree-ring width chronologiesfrom undisturbed, open-grown, low-elevation, moisture-sensi-tive stands of Douglas fir (1 1 chronologies) and Ponderosa pine(five chronologies) at 13 sites in the Canadian Rockies and Bri-tish Columbia (Figure 1; Table 1). A primary goal of this sam-pling programme was to relocate and, where possible, resamplethe LTRR sites and update their chronologies. Chronologies

124 W 122 W 1205'W 11B W

JASPER,

KAMLOOPS

-VANCOUVEERPENTICTON*

t< .-

1x

developed from the eastern and western halves of this networkrepresent two distinct groupings, and only the eastern chrono-logies are presented here. The tree-ring sites selected werelocated close to Historical Canadian Precipitation Network(HCPN) meteorological stations. These records have been cor-rected for inhomogeneities and missing values by the Atmos-pheric Environment Service (AES) of Environment Canada(Mekis and Hogg, 1997). Corrected temperature records were

116W 114°W 112W

i * Meteorological Station

i Limber Pine Chronology

0 100 200 300

Kilometres

-52 N - O The Cartographic Section. Dept of Geography, U.W

11 'W 108'W 1W'

v* 0Towers RidgerBANFF CALGARY

~~~~~~~~50'88

'FORT MACLEOD

('Lundbreck Falls 1MAPE

CRANBROQK

U n ited' States122'W 120'W 118°W 116`W 114,W 112'W 110° W

-48 N158 W

Figure 1 Location map showing meteorological stations and selected tree-ring chronology sites.

- r | _ _ _ _ _ _.1 --- 48°N7-108'W

at Ebsco Electronic Journals Service (EJS) on December 18, 2008 http://hol.sagepub.comDownloaded from


Emma Watson and Brian H. Luckman: Dendroclimatic reconstruction of precipitation in the Canadian Rockies 205

also supplied by AES (for details, see Vincent, 1998) andLuckman and Seed (1995).

Site descriptionsFive well-drained xeric/subxeric sites were sampled in themontane ecoregion of Banff and Jasper National Parks, Alberta(Figure 1). Soils at these sites were primarily Eutric Brunisols(Holland and Coen, 1982). Although the LTRR developedchronologies from three sites near Banff (Exshaw, Tunnel Moun-tain and Powerhouse), only the Powerhouse site was resampledin this study. Robertson and Jozsa (1988) described this site as'one of the best tree-ring sites in Canada'. Old (up to 691 years),open-grown Douglas fir trees grow at the edge of a former ice-marginal terrace, 30-50 m above the present valley floor. Severalof the oldest trees grow at the terrace edge, a steep eroded slopewith hoodoo development, and have their roots exposed bymass wasting.

Douglas fir were sampled from four sites in Jasper NationalPark, approximately 250 km northwest of Banff (Figure 1). ThePyramid Lake site was previously sampled by LTRR in 1966 andby parties from UWO in 1990 and 1995 (Luckman et al., 1996).Open-grown trees were cored along a south-facing bedrock ridgeoverlooking Pyramid Lake. The oldest trees were located near thetop of the ridge and LTRR identification tags were discovered onseveral sampled trees. We did not resample LTRR's adjacentPatricia Lake site. Three previously unsampled sites were coredin the Jasper area to examine the spatial variability in the ring-width signal. At Lake Annette large, young open-grown firs werecored from a valley-floor site. At Maligne Canyon, open-grownfirs were sampled along a south-facing slope overlooking the can-yon. Trees were sampled from an open-forest site growing on alow, west-facing bedrock ridge overlooking the wet, shrubby openmeadow at Prairie de la Vache, approximately 10 km SSE of Jas-per. Fire scars were found on a limited number of trees at all ofthe Jasper sites.

Douglas fir samples were also collected from two previouslyunsampled sites near Cranbrook, B.C., in the floor of the RockyMountain Trench (Figure 1). The Perry Creek site was on anunlogged bluff overlooking the Perry Creek Logging Road. TheWasa Lake site was an east-facing, well-drained, grassy bluffdominated by open-grown Douglas fir and Ponderosa pine.

Chronology development and characteristicsSamples were prepared, measured and cross-dated using standardprocedures (Stokes and Smiley, 1968) and all dating was verifiedusing the computer program COFECHA available in the Den-drochronology Program Library (DPL; Grissino-Mayer et al.,1996). Ring-width measurements were carried out to .001 mmusing a Velmex Unislide traversing table and digital counter.ARSTAN (also available in the DPL) was used to standardize rawring-width measurements with modified negative exponential ornegative linear trend models and to combine these standardizedindices into site chronologies (Cook et al., 1990; Briffa et al.,1996). These deterministic growth models are considered to beconservative (Cook, 1985) in that they are likely to retain mostof the low-frequency climatic information which is resolvablegiven the length of the series.

Selected raw ring-width series from the earlier LTRR Power-house data set were used to increase sample depth for the 1997chronology. However, series from those trees known to be pre-viously cored by LTRR were excluded to avoid artificially inflat-ing chronology statistics. Samples from earlier UWO collectionswere added to the 1997 Pyramid Lake and Maligne Canyonchronologies. Although the LTRR Patricia Lake chronologyextends back to 1540, the raw ring-width measurements available

Summary statistics for the UWO 1997 and LTRR Douglas firchronologies are given in Table 1. Similar characteristics for otherspecies in the region are also shown. The Douglas fir chronologiesdisplay high interannual variability and a strong common signal.Subsample Signal Strength values (SSS; Table 1) exceed 0.85with only 3-6 trees at a site which suggests that long, high-qualitytree-ring chronologies may be developed from a limited numberof samples at these sites (see Briffa and Jones, 1990).Ten of these standardized chronologies are presented in Figure

2. Correlations between all 13 chronologies over their commonperiod (1796-1965) are presented in Figure 3. The two Power-house chronologies are almost identical back to about 1560, butthe 1997 chronology is 100 years longer. In the early to mid-sixteenth century, both chronologies display a double peak whichis more pronounced in the 1997 chronology. The LTRR Banff1966 chronology is a composite of the Exshaw and Tunnel Moun-tain chronologies and is strongly correlated with the Powerhouse1997 chronology. Differences in the amplitude of peaks andtroughs between the Banff chronologies (Figure 2) probablyreflect varying sensitivity of the trees to prolonged wet or dryconditions due to differences in site characteristics (e.g., soils,aspect, slope) or microclimatic differences between these sites.The low-frequency patterns in the Patricia Lake and two Pyra-

mid Lake chronologies are very similar, but the shorter Jasperchronologies display more variability (Watson, 1998). Neverthe-less, all the Jasper chronologies are well correlated with the 1997Pyramid Lake chronology, suggesting that this chronology exhib-its the strongest regional signal. Although correlations betweenthe six Jasper chronologies and the five Banff chronologies range

from 0.25 to 0.58 (Figure 3), the low-frequency patterns in thechronologies are quite similar (Figure 2).The correlation between the two Cranbrook area chronologies

(Wasa Lake and Perry Creek) is not particularly high (r = 0.56).However, both chronologies exhibit enhanced growth in the early1900s and suppressed growth during the late 1920s and 1930s.Following this well-documented drought, both chronologies dis-play enhanced growth until the last quarter of the twentieth cen-

tury when growth has been declining (particularly at the PerryCreek Site). The Cranbrook area chronologies are more highlycorrelated with the Banff chronologies than those from Jasper(Figure 3).

Tree-ring/climate relationships

Precipitation and temperature records from meteorological sta-tions located in Banff, Jasper and Cranbrook (Figure 1) were usedto evaluate the climate signal contained in the 13 chronologies on

a monthly, seasonal and annual basis.

Monthly relationshipsSimple correlations were computed between the 13 chronologiesand monthly precipitation and temperature records from the clos-est meteorological station for a 19-month period (previous Aprilto current October). All but two of the significant correlationsbetween ring widths and precipitation are positive, indicating thathigher precipitation generally results in enhanced tree growth(Figure 4). However, the highest monthly correlations differbetween the three areas sampled. The five Banff chronologies are

all most highly correlated with precipitation from July of the pre-vious year (r = 0.40-0.53), usually followed by May of the growthyear (r = 0.28-0.39). Jasper chronologies are predominately corre-

lated with conditions during the growth year: five of the sixchronologies are most highly correlated with May precipitation (r= 0.41-0.52) followed by February. The Cranbrook chronologiesare most highly correlated with previous August (Wasa Lake) or

previous December (Perry Creek). The Banff and Cranbrookfrom ITRDB only extend back to 1700 and afford no extensionto our 1997 Pyramid Lake chronology.




1300 1350 1400 1450 1500 1550 1600 1650 1700 1750 1800 1850 1900 1950 20003- 50

Powerhouse 1997 40

2-30

10

0 Po ehos 10S,= _,96 30 74

-30

20

2 Tunnel Mountain

2_ N

n-

ling-width Indices25 yr Cubic Splinelumber of Samples

,> WtX iAi-g iET;-

1300 1350 1400 1450 1500 1550 1600 1650 1700 1750 1800 1850 1900 1950

Year-50

-- 40

- 30

- 20

10

0

Perry Creek 30

10

I~~~~~~~~~

2 Pyramid Lake 1997c

m5

-40

-- 30

- 20

--10

-- 0

2000

-30

-20

10

Cc 0 L.1 10

2

I Pyramid Lake 1965.,E- .E - A- w .

0-- _/_t-2

u --u

2- Prairie de la Vache - 30

1 A~~~~~~~~~~~~~'720

10

01700 1750 1800 1850 1900 1950 2000

Year

C-

CD

a

cn0)

cn

1500 1550 1600 1650 1700 1750 1800 1850 1900 1950 2000

Year

Figure 2 Selected Douglas fir tree-ring chronologies for the study area.

chronologies apparently show a much broader climate windowwith significant correlations with spring and summer months inboth the current and previous years (Figure 4). These differencesin the response of the Banff, Jasper and Cranbrook chronologiesmay be associated with differences in site characteristics or theprecipitation regimes experienced in these regions, or may beattributable to differences in the time period over which the corre-

lations were calculated.Correlation coefficients computed between the 13 chronologies

and monthly temperatures are lower than those with precipitationwith maximum chronology-temperature correlations rangingfrom -0.24 to -0.48. Monthly correlation patterns for tempera-

tures are similar but of opposite sign to those for precipitation(Figure 4). Maximum correlations for the Banff chronologiesoccur with previous July (r = 0.33-0.48). Five of six Jasperchronologies are most highly correlated with temperature in cur-

rent March or May (r = 0.33-0.38). The negative correlationsbetween ring width and temperatures in part reflect inverse

Wasa Lake

='~1 '

3 --

2 -

1-

0-wx 2

-1..cc

2_^

a

0*

co70

Cl)BR

IXxa)

3.1

cc:

10

!O

10

Powerhouse 1965




Banff Jasper

ttt"0+ +4 tRtf +6

Powerhouse97(P097) I_ 1@ >147 147l..SPowerhouse66(P066)Banff(BA) | --X - i;037;A

Exshaw(EX) s .27 0.28Tunnel Mtn.(TU) T:k0:02 .5t 1 7Pyramid lake 97(PY97) _ _

Pyramid Lake 66(PY66)I_Patricia Lake(PA) - 1.00

Maligne Canyon(MC) >0.1-019

LakeAnnette(LA)Prairie de laVache(PD) 0.50-0.70

0.30-0.49Wasa Lake(WA) 0.20-0.29

Perry Creek(PC)

Figure 3 Correlations between moisture-sensitive Dougllogies for the southern Canadian Cordillera for the common1965; n = 170). All correlations are significant at the p =

BaA

Cran-brook

0e 3 0.2+0.26:P 0.23.2650;0a:M

Seasonal and annual relationshipsTree growth (i.e., ring width) often correlates more highly withseasonal precipitation totals than with individual monthly values.However, correlations between several combined monthly totals(e.g., May-June, June-July, May-July and J-anuary-March) andring-width data were not much higher than the single-month cor-

relations. Previous studies (e.g., Duvick and Blasing, 1981; Stock-ton and Meko, 1983; Case and MacDonald, 1995) have shownthe best results with annualized data that combine present andprior year effects. Correlation analyses were run for 13 different

0.31 12-month periods from January to December of the previous year_022!0.25- to January to December of the growth year. Maximum corre-

0-0-4g+-t-33- lations all involved those intervals spanning parts of the presentand previous summers (Table 2). The Banff chronologies corre-lated most highly with pJuly-June whereas the majority of Jasperchronologies were most highly correlated with the pAugust-July

[as fir chrono- period. Maximum annual correlations for the Cranbrook chrono-period (1796- logies occur with either pJuly-June or pAugust-July precipitation..05 level. Results for the LTRR and 1997 Pyramid Lake and Powerhouse

chronologies are very similar when calculated over the same inter-val. The three LTRR chronologies developed from Banff sites

Temp.- Prec. (Tunnel Mountain, Powerhouse and Exshaw) show a decrease inanff Correlations correlation with Banff climate data that appears to parallel their

respective distances from the meteorological station.I dsacs mtoooia

M -6.16

Jo -0.66-u 0.66A -0.55

S -0.53

N=100

-6

S -" 1t-0

0

M -0.33J -0.18Ju -0.51A -0.39S -0.36

N =59

ui 2 M -022

-0'1 i am Ju -0.30. -1 ~~~~~~~~~~~~A-0.36

*-2 S -0.570. A M J J A S NN D J F M A M J J A S Aa)D N =77

Figure 4 Number of statistically significant (p = .05) correlations betweenthe standard tree-ring chronologies and monthly instrumental temperatureand precipitation data. The chronology sites are grouped based on themeteorological station used for analysis. Correlations are normally positivewith precipitation and negative with temperature, though statistically sig-nificant results for prior December at Banff have the opposite sign. Thesingle best monthly correlation is identified for each chronology. Corre-lations between monthly temperature and precipitation values for May-September are shown in the right-hand column. All correlations are nega-

tive and bolded values are significant at p = 0.05.

relationships between precipitation and temperature in certainmonths (Figure 4). Higher temperatures also increase water stressduring the growing season (Fritts, 1976). Maximum correlationsfor the Cranbrook chronologies (0.24-0.27) are lower than thosefor the other sites. The fact that the chronologies are sensitive toboth precipitation and temperature suggests that it may be possibleto reconstruct climatic indices that incorporate both parameters(e.g., PDSI; see Cook el al., 1988; Stahle et al., 1988). Thoughfew PDSI data are presently available, preliminary results usinguncorrected daily data for Jasper provided by W. Skinner(personal communication, 1998) show encouraging results. Thecorrelation between May PSDI (1953-1994) and the 1997 Pyra-mid Lake chronology was 0.79.

Development of climate reconstructions

Calibration and verification proceduresStepwise multiple linear regression analysis was used to developstatistical relationships between precipitation and the most highlycorrelated tree-ring chronologies. In order to account for laggedand autocorrelated growth effects in the tree-ring chronologies,forward (t+1) and backward (t-1) lagged tree-ring indices were

Table 2 Maximum correlations between annual precipitation andmoisture-sensitive Douglas fir chronologies from the southern CanadianCordillera

Chronology Annualization period N'

pJun pJul pAugMay Jun Jul

Prairie de la Vache 0.59 0.62 0.60 58JPyramid Lake..0.61 ~0.67 0.71 58JPyramid Lake (66) 0.60 0.65 0.66 29JPatricia Lake (66) 0.69 0.71 0.69 29JMaligne Canyon 0.56 0.64 0.67 58JLake Annette 0.55 0.62 0.63 58JPowerhouse 0.66 0.71 0.63 99BPowerhouse (66) 0.67 0.75 0.67 70BBanff (66) 0.68 0.73 0.66 70BTunnel (66) 0.69 0.77 0.66 70BExshaw (66) 0.57 0.58 0.55 70BPower+Tulnnel 0.68 0.74 0.65 99BWasa Lake 0.66 0.67 0.70 76CPerry Creek 0.68 0.70 0.68 76CPCI3Xe ;;0ti t:S :X:t0.74 0.76 0.75 76C

Precipitation reconstructions were developed for the relationships high-lighted in grey. All correlations are statistically significant at the p.01 level.'Number of years (J = Jasper, B = Banff and C = Cranbrook).2Chronology for the combined Powerhouse and Tunnel Mountain ring-width series.3First principal component of the Wasa Lake and Perry Creek chrono-logies.




included as potential predictors in the regression analyses. Thestatistical quality of the precipitation estimates derived from thetransfer functions were evaluated using several parametric andnon-parametric statistical measures over three different intervals(Table 3). The conventional data-splitting approach (Gordon,1982) provides a limited assessment of the temporal stability ofthe tree-ring/climate relationship. The alternative, 'leave-out-one'procedure (Gordon, 1982) utilizes the entire length of record.Initially, a single observation is withheld and the relationshipbetween precipitation and the tree-ring predictor(s) is calibratedfrom the remaining observations and used to estimate theexcluded observation. This procedure is repeated to generate a setof independent precipitation estimates of equal length to the orig-inal record. Although the parameters for the final, full calibrationmodel are still not verifiable, this technique uses the full data setto develop independent estimates which can be used to verifyreconstruction skill (Gordon, 1982). This method is similar to thepredicted residual error sum of squares (PRESS) procedure(Draper and Smith, 1981) and to cross-validation techniquesdescribed by Michaelsen (1987), Loaiciga et al. (1993) andWoodhouse and Meko (1997).

Development of individual station reconstructionmodelsAlthough six precipitation reconstructions (Banff, Jasper, Kam-loops, Westwold, Penticton and Cranbrook) have been developed(Watson, 1998), only the eastern reconstructions are presentedhere. These reconstructions explain a high percentage of the vari-ance in their respective instrumental record and are well verified(Table 3). Additionally, analysis of the Durbin-Watson D statisticand results of the Kolmogorov-Smirnov test (Table 3) indicatethat the residuals from the three models do not display statisticallysignificant serial correlation and do not differ significantly fromthe theoretical normal distribution. Visual inspection of plots ofthe standardized residuals from the three regression modelsagainst time (not shown) also reveal no major violations of theassumptions of the general linear model (Draper and Smith,

1981). The calibration models described below were used toreconstruct precipitation back to the year when SSS falls below0.85 (Table 1).

BanffSeveral stepwise multiple regression analyses were conducted forthe period previous July to current June using various combi-nations of the five standardized Banff chronologies. The longest,best-verified reconstruction with the lowest standard error was

generated from a multiple regression conducted between theinstrumental precipitation record (pJuly-June) and a chronologydeveloped from the combination of the 1997 UWO Powerhousechronology and the Tunnel Mountain chronology (1570-1966).

Figure 5 clearly illustrates the ability of the regression modelto predict both high and low frequency variations in the instru-mental data. The correlation between the chronology and theinstrumental data for the full calibration model (1896-1994) was

0.77; the adjusted RK value of 0.57 revealed that slightly less than60% of the variance in the Banff precipitation record was cali-brated (Table 3). The estimates derived from the calibration modelpassed all three verification tests (Table 3) suggesting that the fullcalibration equation can provide reliable (i.e., time stable) esti-mates of Banff (pJuly-June) precipitation. The full calibrationmodel was used to reconstruct Banff precipitation back to 1430(Figure 6). Note that this Banff reconstruction is slightly revisedfrom an earlier version (Watson, 1998; Luckman and Watson,1999; Luckman, 2000).

JasperMultiple regression analyses were conducted between variousgroupings of the six chronologies from Jasper National Park andthe Jasper precipitation record for two annual periods (pJuly-Juneand pAugust-July). The best reconstruction model was developedfrom the simple linear relationship between the 1997 PyramidLake chronology in the current year and annual (pAugust-July)precipitation for the full period 1937-1994. Although the R andadjusted R2 values (0.71 and 0.50 respectively; Table 3) are

Table 3 Statistical evaluation of the calibration and verification models and the residuals from the three full reconstruction models

Model Calibration N' n2 Standard R R2d Verification r RE3 Sign test4 K-S5 Durbin6-period error period (agree/ test Watson D

dis.)

Banff (Meteorological Station elevation: 1389 m)Early 1896-1944 3 1 (t) 54.55 0.80* 0.64 1945-1994 0.64* +0.47 35/15Late 1945-1994 3 2 (t,t+1) 56.78 0.68* 0.45 1896-1944 0.78* +0.60 36/13Full 1896-1994 3 3 (t,t+l,t-1) 55.79 0.77* 0.57 1896-1994 0.74* +0.56 78/21 PASS 2.04

Jasper (Meteorological Station elevation: 1061 m)Early 1937-1965 3 1 (t) 59.55 0.68* 0.44 1966-1994 0.75* +0.55 25/4Late 1966-1994 3 1 (t) 45.10 0.75* 0.55 1937-1965 0.68* +0.46 24/5Full 1937-1994 3 1 (t) 52.76 0.71* 0.50 1937-1994 0.69* +0.49 47/11 PASS 1.73

Cranbrook (Meteorological Station elevation: 939 m)Early 1921-1958 3 1 (t) 68.47 0.77* 0.58 1959-1995 0.70* +0.58 26/11Late 1959-1995 3 2 (t,t-l) 63.97 0.78* 0.59 1921-1958 0.72* +0.53 25/13nsFull 1921-1995 3 3 (t,t+l,t-1) 66.93 0.79* 0.60 1921-1995 0.76* +0.58 56/19 PASS 2.36

Verification tests were conducted using (a) the traditional data-splitting approach (early and late periods) and (b) for the full length of the instrumentalrecords using an independent set of estimates developed using the leave-out-one procedure (Gordon, 1982).'Number of potential predictors.2Number of predicators retained in the model.3Positive reduction of error values (RE) are regarded as an indication of successful reconstruction of the climatic variable (Fritts, 1976).4In all but one case (ns), departures from the mean in the instrumental and estimated series are significantly associated (p = .05) (for details, see Fritts, 1976).5The Kolmogorov-Smimov test results show that the distribution of the residuals does not differ significantly from the theoretical normal distribution.6Results show that statistically significant serial correlation is not present (.01 level) in the residuals.*Statistically significant .05 level.




Banff

1890

Jasper

Actual PrecipitationReconstructed Precipitation

v i I , ~ ~ ~ ~ ~~~ ~ ~ ~ ~ ~ ~ ~ ~~~~~~~Ii.I , . .

18901900 1910 1920 1930 1940 1950 1960 1970 1980 19902000Year

700 -650 -600 -

E 550-E500-

0450400~.2 400 -

C..(2 350

300 -250200 -

Cranbrook

11 .1

1890 1900 1910 1920 1930 1940 1950Year

1960 1970 1980 1990 2000

Figure 5 Comparison of instrumental and reconstructed precipitation records for the calibration period at Banff (1896-1994, R = 0.77), Jasper (1937-1994, R = 0.71) and Cranbrook (1921-1995, R = 0.79).

slightly lower than those for the Banff reconstruction, the precipi-tation estimates are very similar to instrumental values during thecalibration interval (Figure 5). Estimates based on the early, lateand full reconstruction models passed all three verification tests.The reconstruction also correlated significantly with a less rigor-ously corrected Jasper precipitation record (Luckman and Seed,1995) for years outside the calibration interval (1916-1936; r =

0.67). As the ability of the regression model to predict precipi-tation outside the calibration interval has been demonstrated, thering-width indices from the Pyramid Lake chronology were sub-stituted into the transfer function yielding precipitation estimatesback to 1710 (Figure 6).

CranbrookThe two Cranbrook chronologies are highly correlated (0.70) withannual (pJuly-June or pAugust-July) precipitation (Table 3).Multiple models were evaluated using (1) the two chronologiesand (2) the first eigenvector (89% explained) of the two chrono-logies. In both cases, annual (pAug-July and pJuly-June) precipi-tation was regressed against predictor chronologies (in years t, t+1and t-1). The best model used the first eigenvector (years t, t+land t-1) to reconstruct pAugust-July precipitation with R andR2d values for the model of 0.79 and 0.60 respectively (Table 3).These values are higher than those for the Banff and Jasper recon-

structions and the ability of the model is clearly displayed in Fig-ure 5. The model passed all three verification tests for the earlyand full reconstructions but estimates derived from the late modelmarginally failed the sign test conducted against instrumental datafrom the early period (Table 3). This is not considered to be a

major problem. The full reconstruction model was used toestimate annual Cranbrook precipitation back to 1827 (Figure 6).

Comparison of the reconstructions

The Banff, Jasper and Cranbrook reconstructions are of compara-ble statistical quality to those developed for arid sites in theUnited States and have slightly better calibration and verificationstatistics than the other precipitation reconstructions developed forwestern Canada (Case and MacDonald, 1995; Sauchyn and Beau-doin, 1998). These reconstructions extend the instrumentalmeteorological records by 465 years for Banff, 226 years for Jas-per and 95 years for Cranbrook.

The Banff, Jasper and Cranbrook reconstructionsMean annual precipitation for the 1430-1994 period at Banff(493.30 mm) is slightly lower than that recorded over the periodof instrumental observations (508.39 mm, 1890-1994), suggestingthat the twentieth century was slightly wetter than the preceding400 years (Figure 6). The most prolonged and extremely dry per-

iod is reconstructed for the interval c. 1470-1510 (also identifiedfor Califomia by Hughes and Brown, 1992) and is followed bythe wettest conditions in the entire reconstruction in c. 1515-1550.Dry intervals lasting for an average of roughly 30 years were

centred in approximately 1570 (Stahle et al., 1985; 2000; Wood-house and Overpeck, 1998), 1630, 1705, 1760 and 1850 (the latterfour were noted for Nebraska by Weakley, 1943). Although thereare fewer wet intervals in the reconstruction, they are generally

700650 -600 --

550 -500450 -

400 -350 -

EE

a

0.

co

600 -

550 --E 500-E= 450-0Z< 400-Q. 350 -% 300-

250

,\nn

vtll

--I




800 Banff

20014301450 1500 1550 1600 1650 1700 1750 1800 1850 1900 1950 2000

- 700-

. 600-c

*- 500-

*= 400-a..

cl- 300-

YearJasper

1600 1650 1700Year

1750 1800 1850 1900 1950 2000

Precipitation-- Mean

25 yr Cubic Spline+ 2 Standard Errors

1500 1550 1600 1650 1700 1750 1800 1850 1900 1950 2000Year

Figure 6 Precipitation reconstructions for Banff (1430-1994), Jasper (1710-1994) and Cranbrook (1827-1995). Reconstructed precipitation, a 25-yearfilter and 2 standard error confidence intervals are shown. The confidence intervals incorporate an estimate of the increase in uncertainty of the reconstructionassociated with decreasing sample depth back in time.

more extreme (i.e., deviate further from the mean) than dry events.Major wet intervals are reconstructed for the periods approxi-mately 1585-1610, 1660-1680, 1870-1885, 1895-1910 and from1950 onwards. The latter two are also known from the instrumen-tal record (Figure 5).The means of the reconstructed (1710-1994) and instrumental

records for annual (pAugust-July) precipitation at Jasper are

almost identical (444.93 and 441.81 mm; Figure 6). The mostsevere and prolonged dry periods were reconstructed for the per-

iods 1820-1865 and 1915-1945. The most pronounced wetterintervals were centred around 1785, 1805 and 1880. The short(1827-1995) Cranbrook reconstruction (Figure 6) shows slightlylower mean precipitation (408.50 mm) for the 1827-1921 periodthan the instrumental record (416.31 mm). After c. 1890, precipi-tation increased dramatically and remained above average untilapproximately 1920. A dry interval was reconstructed for the per-

iod c. 1920-1940 followed by a 40-year wetter interval (c. 1940-1980). Reconstructed annual precipitation has declined since the1980s.

Comparison with reconstructions from adjacentregionsCase and MacDonald (1995) developed a 487-year reconstructionof annual (pAugust-July) precipitation for the Rocky Mountainfoothills of southwestern Alberta. Their reconstruction (Figure 7)is developed from two limber pine tree-ring sites approximately60 and 200 km east and southeast of Banff (Towers Ridge andLundbreck Falls; Figure 1). Sauchyn and Beaudoin (1998) have

developed a 313-year annual (pAugust-July) precipitation recon-

struction for Maple Creek, Saskatchewan, using Picea glauca(white spruce) from the Cypress Hills of southeastern Alberta andsouthwestern Saskatchewan (Prairies reconstruction, Figure 7).The R'dJ for this reconstruction was 36.5%. PDSI reconstructionsdeveloped by Cook et al. (1996; 1999) for northeastern Wash-ington and northern ldaho are also plotted on Figure 7. To facili-tate comparison, the precipitation reconstructions have been con-

verted to z-scores and all series are smoothed with a 25-yearsmoothing spline. The vertical shading on Figure 7 indicates inter-vals after 1700 when the majority of the records (four of six from1700-1830; five of seven from 1830-1990) show below aver-

age precipitation.Table 4 presents correlations between the Banff, Jasper, Cran-

brook, Foothills and Prairies precipitation reconstructions forapproximately 50-year intervals and for the maximum common

paired-intervals. Table 4 and Figure 7 show that considerable spa-

tial and temporal variations exist in the relationships between thereconstructions. The trends for Banff and Jasper are strikinglysimilar for c. 1770-1815 and c. 1865-1990, but are less similarin the first halves of the eighteenth and nineteenth centuries. Apartfrom the interval 1800-1840, low-frequency trends in the Foot-hills reconstruction after the mid-1700s are quite similar to thosein the Banff and Jasper reconstructions. The period 1751-1800 isthe only interval when all of the reconstructions (except Cran-brook, which begins in 1827) are significantly correlated witheach other (Table 4). This interval is dominated by two major

onA _L14301450

1500 150I1500 1 550

800 -- Cranbrook- 700-E I-600-=

<, 500-

*= 400--a,wCX 300-

200 -

10014301450




1450 1500 1550 1600 1650 1700 1750 1800 1850 1900 1950 2000

31Prairies:

::(1694-1984) o- C)C ~ 0 )OL)(2.0

Foothills: \ \(1519-1980)

0.0

-20 :3.0-

1.0 (1442-1982)/

3.01

-2.01

-~Jasperj.(1722-1982)

-2.0

2 P6 N.Idho

1450 1500 1550 1600 1650 1700 1750 1800 1850 1900 1950 2000

Year

Figure 7 Comparison of low-frequency patterns in precipitation reconstructions for the southern Canadian Cordillera and adjacent areas. The 'Prairies'

record is a reconstr-uction from Maple Creek, Saskatchewan (Sauchyn and Beaudoin, 1998) and the 'Foothills' record is from Case and MacDonald (1995).

The PDSI reconstructions are from Cook et al. (1 996) and refer to grid points 8 and 16 of their network, immediately south of the Canadian border. The

vertical shading indicates intervals after 1700 when the m-ajority of the records (four of six from 1700-1830; five of seven fronm 1830-1990) show bclow-

average precipitation.

droughts (the 1 760s and 1 790s) which are coherent throughout

the region.

Comparison of the Banff and Foothills record prior to 1750

shows a similar but slightly offset pattern through the 150OIs (note

that the Foothills reconstruction also exhibits a pronounced double

peak in the early 1500Os) and less agreement in the latter half of

the seventeenth century (r = 0.32; 1651 1700). Although present

in the Banff and Jasper reconstructions, the 1 790s drought is much

more extreme in the Foothills and Prairies reconstructions. Several

reconstructions developed for the Great Plains of the United States

also identified a major drought in the 1790s (e.g., Weakley, 1943;

Duvick and Blasing, 1981; Blasing et al., 1988). The 1850-1860s

drought identified in the Banff, Jasper, Cranbrook and Foothills

reconstructions has also been identified in studies from the Great

Plains of the United States (e.g., Meko, 1992; Stockton and Meko,

1983) eastern California and the western and southwestern United

States (see Woodhouse and Overpeck, 1998). During the first half

of the twentieth century, the Banff, Jasper, Cranbrook and Foot-

hills reconstructions are very similar but the reconstructions sug-

gest the mountains are wetter during the latter half of the present

century (Figure 7). With the exception of the late 1700s, the pat-

tern shown by the reconstructed precipitation record from the

0

CIO

0

CZ

CuCZ,

0)




Table 4 Correlations between selected annual precipitation reconstructions over c. 50-year intervals

Banff Jasper Cranbrook Prairies

Jasper Cran. Prairies Foothills Cran. Prairies Foothills Prairies Foothills Foothills

1701-1750 -0.09 -0.09 0.61 -0.10 -0.02 0.111751-1800 0.59 0.30 0.61 0.39 0.46 0.341801-1850 0.37 - -0.25 0.50 - -0.23 0.22 - - -0.041851-1900 0.32 0.58 0.36 0.56 0.38 0.03 0.26 0.21 0.56 -0.021901-1950 0.50 0.65 0.31 0.53 0.39 0.42 0.02 0.33 0.44 0.181951-1992 0.38 0.39 0.24 0.52 0.26 0.17 0.27 0.44 0.34 0.30Full 0.39 0.58 0.16 0.47 0.27 0.11 0.22 0.28 0.41 0.14

Note that the reconstructions are for different annual precipitation periods (pJuly-June or pAugust-July) and are of different length (Figure 7). Correlationsbetween the Foothills and Banff reconstructions are lower but still significant during the sixteenth and seventeenth centuries: 1507-1550 (0.36), 1551-1600 (0.46). 1601-1650 (0.51) and 1651-1700 (0.32). Statistically significant (p = 0.5) correlations appear in bold.

Prairies is quite different from the other records shown in Figure7. These discrepancies may reflect a different precipitation regimein the Cypress Hills or differences in response between the speciesused for these reconstructions. Further data are needed to resolvethese questions.The reconstruction of periods with coherent low-frequency pat-

terns in precipitation between sites suggest times of strong, com-mon, regional controls on precipitation. Differences in the low-frequency trends between sites may result from local site or spec-ies effects in individual chronologies, but they may also be indica-tive of much larger-scale climatic patterns. The fact that precipi-tation variability is not always coherent throughout the regionsuggests that there may be multiple controls whose relative influ-ence on precipitation varies through time. A larger network ofsites is needed to evaluate these possibilities. Reconstructionsdeveloped for Kamloops, Westwold and Penticton (Watson, 1998)are of comparable quality to those at Banff, Jasper and Cranbrookand show similar low-frequency variability in the nineteenth andtwentieth centuries. These variations are also present in recentPDSI reconstructions from the adjacent United States (Cook et al.,1996; Figure 7) suggesting that low-frequency precipitation varia-bility during this interval was largely controlled by regional-scaleatmospheric circulation patterns. Similar regional control mayoccur during the fifteenth to eighteenth centuries, but more longreconstructions are needed to evaluate this possibility.

Conclusions

This paper has demonstrated that high-quality annual precipitationreconstructions can be developed from moisture-sensitive Douglasfir ring-width chronologies from low-elevation sites in the Canad-ian Cordillera. Regression-based calibration models captured atleast 50% of the variance in the instrumental precipitation records.Long-term trends in the Banff, Jasper and Cranbrook reconstruc-tions are similar to each other and to reconstructions from theRocky Mountain Foothills and PDSI reconstructions from theadjacent United States (Washington and Idaho). They indicate thatregionally extensive drier intervals occurred over the southernRockies through the AD 1700s (particularly the 1760s and 1790s),the 1850- 1860s, the 1 890s and during the historically documenteddrought of the 1920-1940s. Significantly greater amounts of pre-cipitation occurred during the 1870-1880s, 1900-1910s and the1950-1970s (in the mountains only). The reconstructed precipi-tation record for Banff suggests significant drier intervals alsooccurred c. AD 1470-1510, in the 1560-1570s and the 1630-1650s with a major wetter interval c. 1520-1550. The coherencybetween reconstructions in the last three centuries indicates thatthe individual calibration models are well developed. It also sug-

gests that the low-frequency behaviour of precipitation across theregion may be controlled by synoptic or larger-scale circulationfeatures. Further work is needed to explore these relationships inthe present instrumental record and to develop a comparable net-work of long, high-quality, drought-sensitive chronologies acrossthe southern Canadian Cordillera to examine the coherency ofthese low-frequency precipitation patterns over time.

Acknowledgements

This project was funded by the Atmospheric Environment Service(AES) of Environment Canada (since renamed the MeteorologicalService of Canada) and the Natural Sciences and EngineeringResearch Council of Canada. We thank Walter Skinner (AES) forhis support of this project and Lucie Vincent and Eva Mekis ofAES for providing information regarding the climate records.Thanks also to Cliff Whyte ,who assisted in locating suitable sam-pling sites in Banff and Jasper National Parks; to David Mercerand Susan Muleme (Cartographic Section, UWO) for cartography;and Keith Briffa for comments on the manuscript. For assistancein the field and/or laboratory we wish to thank Nicholas Neary,Heather Luckman, Bonnie Robinson and Rob Wilson. Thanks alsoto R. Case, G. MacDonald and D. Sauchyn for providing tree-ring data from their published work.

References

Biasing, T.J., Stahle, D.W. and Duvick, D.N. 1988: Tree ring-basedreconstruction of annual precipitation in the south-central United Statesfrom 1750 to 1980. Water- Resources Research 24, 163-71.Briffa, K.R. and Jones, P.D. 1990: Basic chronology statistics and assess-ment. In Cook, E.R. and Kairiukstis, L.A., editors, Methods ofdendroch-ronology: applications in the environmental sciences, Dordrecht: Kluwer,137-52.Briffa, K.R., Jones, P.D., Schweingruber, F.H., Karlen, W. and Shiya-tov, S.G. 1996: Tree-ring variables as proxy-climate indicators: problemswith low frequency signals. In Jones, P.D., Bradley, R.S. and Jouzel, J.,editors. Climatic variations and for-cing mechanisms ofthe last 2000 years,NATO ASI Series, Vol. I 41. New York: Springer-Verlag, 9-41.Case, R.A. and MacDonald, G.M. 1995: A dendroclimatic reconstructionof annual precipitation on the western Canadian prairies since AD 1505from Pinus flexilis James. Quaternarv Research 44, 267-75.Cook, E.R. 1985: A time series analysis approach to tree-ring standardiz-ation. PhD thesis, University of Arizona, Tucson.Cook, E.R., Kablack, M.A. and Jacoby, G.C. 1988: The 1986 droughtin the southeastem United States: how rare an event was it? Journal ofGeophysical Research 93, 14257-60.Cook, E.R., Meko, D.M., Stahle, D.W. and Cleaveland, M.K. 1996:




Tree-ring reconstructions of past drought across the coterminous UnitedStates: tests of regression method and calibration/verification results. InDean, J.S., Meko, D.M. and Swetnam, T.W., editors, Tree rings, environ-ment and humanity, Radiocarbon, 155-69.

1999: Drought reconstructions for the continental United States.Journal of Climate 12, 1145-62.Cook, E.R., Shiyatov, S. and Mazepa, V. 1990: Estimation of the meanchronology. In Cook, E.R. and Kairiukstis, L.A., editors, Methods ofden-drochronologv: applications in the environmental sciences, Dordrecht:Kluwer, 123-32.Draper, N. and Smith, H. 1981: Applied regression analysis (secondedition). New York: Wiley.Drew, L.G. 1975: Tree-ring chi-onologies of western North America: VICanada and Mexico. Laboratory of Tree-Ring Research, Chronology Ser-ies No. 1. Tucson: University of Arizona.Duvick, D.N. and Blasing, T.J. 1981: A dendroclimatic reconstruction ofannual precipitation amounts in Iowa since 1680. Waiter ResourcesResearch 17, 1183-89.Fritts, H.C. 1974: Relationships of ring widths in arid-site conifers tovariations in monthly temperature and precipitation. Ecological Mono-graphs 44, 411-40.

1976: Tree rings and climate. London: Academic Press.1991. Reconstructing large-scale climatic patterns from tree-ring

data. Tucson: University of Arizona Press.Gordon, G.A. 1982: Verification of dendroclimatic reconstructions. InHughes, M.K., Kelly, P.M., Pilcher, J.R. and LaMarche, V.C. Jr, editors,Climate from tree rings, Cambridge: Cambridge University Press, 58-61.Grissino-Mayer, H.D., Holmes, R.L. and Fritts, H.C. 1996: The Inter-national Tree-Ring Data Bank Programn Library, version 2, users manual.Laboratory of Tree-Ring Research, Tucson. Arizona.Holland, W.D. and Coen, G.M. 1982: Ecological (biophysical) landclassification of Banif and Jasper National Parks, Vol HI: soil and veg-etation resources. Alberta Institute of Pedology Publication No. SS-82-44.Houghton, J.T., Meira Filho, L.G., Callander, B.A., Hlarris, N., Kat-tenberg, A. and Maskell, K. 1996: Technical summary. In Houghton.J.T., Meira Filho, L.G., Callander, B.A., Harris, N., Kattenberg, A. andMaskell, K., editors, Climate change 1995: the science of c limate change,Intergovernmental Panel on Climate Change. Cambridge: Cambridge Uni-versity Press, 9-47.Hughes, M.K. and Brown, P.M. 1992: Drought frequency in central Cali-fornia since 101 Bc recorded in giant sequoia tree rings. Climate Dynamics6, 161-67.Loaiciga, H.A., Haston, L. and Michaelsen, J. 1993: Dendrohydrologyand long-term hydrologic phenomenoni. Reviews of Geophysics 31, 151-71.Luckman, B.H. 2000: The Little Ice Age in the Canadian Rockies. Geo-morphology, 32, 357-84.Luckman, B.H. and Seed, E.D. 1995: Fire-climate relationships andtrends in the Mouintain National Parks. Final Report: Contract C2242-4-2185, Parks Canada.Luckman, B.H. and Watson, E. 1999: Precipitation reconstruction in thesouthern Canadian Cordillera. Preprints 10th Symnpositum on GlobalChange Studies, 79th Annual Meeting, American Meteorological Society,Dallas, Texas, 10-15 January, 296-99.

Luckman, B.H., Colenutt, M.E., Kavanagh, T.A., Lusted, J.R. andDiraddo, C. 1996: Field investigations in the Canadian Rockies in 1995.Report submitted to Parks Canada, Alberta Parks and B.C. Parks Service.Luckman, B.H., Colenutt, M.E., Kavanagh, T.A. and Seed, E.D. 1995:Field investigations in the Canadian Rockies in 1994. Report submittedto Parks Canada, Alberta Parks and B.C. Parks Service.Mekis, E. and Hogg, W.D. 1997: Rehabilitation and analysis of Canadiandaily precipitation time series. Preprints AMS 10th Conference on AppliedClimatology, Reno NV., 20-23 October, 300-304.Meko, D.M. 1992: Dendroclimatic evidence from the Great Plains of theUnited,States. In Bradley, R.S. and Jones, P.D., editors, Climate since AD1500, London: Routledge, 312-30.Michaelsen, J. 1987: Cross-validation in statistical climate forecast mod-els. Journal of Climate and Applied Meteorology 26, 1589-600.Robertson, E.O. and Jozsa, L.A. 1988: Climatic reconstruction from treerings at Banff. Canadian Journal of Forest Research 18, 888-900.Sauchyn, D.J. and Beaudoin, A.B. 1998: Recent environmental change inthe southwestern Canadian Plains. The Canadian Geographer 42, 337-53.Schulman, E. 1947: Dendrochronologies in southwestern Canada. Tree-Ring Bulletin 13, 10-24.Stahle, D.W., Cleaveland, M.K. and Hehr, J.G. 1985: A 450-yeardrought reconstruction for Arkansas, United States. Nature 316, 530-32.

1988: North Carolina climate changes reconstructed from tree rings:AD 372 to 1985. Science 240, 1517-19.Stahle, D.W., Cook, E.R., Cleaveland, M.K., Therell, M.D., Meko,D.M., Grissino-Meyer, H.D., Watson, E. and Luckman, B.H. 2000:Epic 16th century drought over North America. EOS, 81(12), 121, 125.St George, R.S. 1997: Tree-ring reconstruction ofsummer temperaturesin Banffand Jasper National Parks. Unpublished MSc thesis, Departnmentof Geography, University of Western Ontario.Stockton, C.W. and Meko, D.M. 1983: Drought recurrence in the GreatPlains as reconstructed from long-term tree-ring records. Joturnal of Cli-mate and Applied Meteorology 22, 17-29.Stokes, M.A. and Smiley, T.L. 1968: An introductioni to tree-r-ing dating.Chicago: University of Chicago Press.Vincent, L.A. 1998: A technique for the identification of inhomogeneitiesin Canadian temperature series. Journal of Climate 11, 1094-104.Watson, E. 1998: Dendroclimatic reconsttrction ofprecipitation in Bri-tish Columbia and Alberta. Unpublished MSc thesis, Department ofGeography, University of Western Ontario.Weakley, H.E. 1943: A tree-ring record of precipitation in western Nebra-ska. Journal of Forestry 41, 816-19.Wigley, T.M.L., Briffa, K.R. and Jones, P.D. 1984: On the average ofcorrelated time series, with applications in dendroclimatology and hydro-meteorology. Journal of Climate and Applied Meteorology 23, 201-13.Woodhouse, C.A. and Meko, D.M. 1997: Number of winter precipitatiolldays reconstructed from southwestern tree rings. Journal of Climlate 10,2663-69.Woodhouse, C.A. and Overpeck, J.T. 1998: 2000 years of drought varia-bility in the central United States. Bulletin ofthe American MeteorologicalSociety 79, 2693-714.Youngblut, D.K. 1999: An investigation into the dendr-oclimatic potentialoala long-lived whitebark pine chronology from the Canadian Rockies.Unpublished BSc thesis, Department of Geography, University ofWestern Ontario.



Documents

Dendroclimatic reconstruction of precipitation for sites in the southern