Texas Drought History Reconstructed and Analyzed from 1698 to 1980

David W. Stahle Department of Geography, University of Arkansas, Fayetteville, Arkansas

Search for other papers by David W. Stahle in
Current site
Google Scholar
PubMed
Close
and
Malcolm K. Cleaveland Department of Geography, University of Arkansas, Fayetteville, Arkansas

Search for other papers by Malcolm K. Cleaveland in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

A selected group of nine climate-sensitive tree-ring chronologies from old post oak trees are used to reconstruct the June Palmer Drought Severity Index (PDSI) from 1698 to 1980 for two large regions in northern and southern Texas. Analysis of tree growth and monthly climate variables indicate that the June PDSI is the most robust climate signal evident in these chronologies, and principal component analysis (PCA) reveals a north-south geographic pattern in the relationships between the regional tree-ring time series. Serially random amplitude series from the first two significant eigenvectors of tree growth, which explain 65% of the total variance in the tree-ring data, were entered into stepwise multiple regression as predictors of regionally averaged June PDSI in north and south Texas for the common interval 1931–80. The regression models explain 59% and 60% of the variance in north and south Texas June PDSI, respectively, and both reconstructions are well verified against independent June PDSI data available on a statewide basis from 1888 to 1930. The weak persistence present in the observed June PDSI series was added to the serially random tree-ring reconnections prior to verification, using autoregressive modeling procedures.

The mean and variance of June PDSI during the 50-yr period of meteorological observation (1931–80) appear to be representative of the last 283 yr, but significant changes in average June PDSI for Texas appear to have occurred over both 30 and ∼90-yr time intervals. Moderate or more severe June droughts (PDSI ≤ −2.0) have an estimated recurrence probability of over 90% each decade, and the risk of extreme June drought (PDSI ≤ −4.0) is estimated at over 50% every 15 yr in north Texas and every 10 yr in south Texas. The reconstructions faithfully reproduce the frequency domain properties of the actual June PDSI, and marginally significant spectral peaks are present at 2.3 yr and between 14 and 18.67 yr in both reconstructions. Significant interannual persistence of June moisture extremes apparent in the statewide June temperature, precipitation, and PDSI data from 1888 to 1982 is also present in both regional reconstructions from 1698 to 1980. The reconstructions indicate that the risk for below average June moisture conditions increases to at least 65% in north and south Texas in the summer following a June drought (PDSI ≤ −2.0). Interannual persistence is also indicated for June wetness anomalies and may have some modest value in statistical forecasts of growing season moisture conditions in Texas.

Abstract

A selected group of nine climate-sensitive tree-ring chronologies from old post oak trees are used to reconstruct the June Palmer Drought Severity Index (PDSI) from 1698 to 1980 for two large regions in northern and southern Texas. Analysis of tree growth and monthly climate variables indicate that the June PDSI is the most robust climate signal evident in these chronologies, and principal component analysis (PCA) reveals a north-south geographic pattern in the relationships between the regional tree-ring time series. Serially random amplitude series from the first two significant eigenvectors of tree growth, which explain 65% of the total variance in the tree-ring data, were entered into stepwise multiple regression as predictors of regionally averaged June PDSI in north and south Texas for the common interval 1931–80. The regression models explain 59% and 60% of the variance in north and south Texas June PDSI, respectively, and both reconstructions are well verified against independent June PDSI data available on a statewide basis from 1888 to 1930. The weak persistence present in the observed June PDSI series was added to the serially random tree-ring reconnections prior to verification, using autoregressive modeling procedures.

The mean and variance of June PDSI during the 50-yr period of meteorological observation (1931–80) appear to be representative of the last 283 yr, but significant changes in average June PDSI for Texas appear to have occurred over both 30 and ∼90-yr time intervals. Moderate or more severe June droughts (PDSI ≤ −2.0) have an estimated recurrence probability of over 90% each decade, and the risk of extreme June drought (PDSI ≤ −4.0) is estimated at over 50% every 15 yr in north Texas and every 10 yr in south Texas. The reconstructions faithfully reproduce the frequency domain properties of the actual June PDSI, and marginally significant spectral peaks are present at 2.3 yr and between 14 and 18.67 yr in both reconstructions. Significant interannual persistence of June moisture extremes apparent in the statewide June temperature, precipitation, and PDSI data from 1888 to 1982 is also present in both regional reconstructions from 1698 to 1980. The reconstructions indicate that the risk for below average June moisture conditions increases to at least 65% in north and south Texas in the summer following a June drought (PDSI ≤ −2.0). Interannual persistence is also indicated for June wetness anomalies and may have some modest value in statistical forecasts of growing season moisture conditions in Texas.

JANUARY 1988 DAVID W. STAHLE AND MALCOLM K. CLEAVELAND 59Texas Drought History Reconstructed and Analyzed from 1698 to 1980DAVID W. STAHLE AND MALCOLM K. CLEAVELANDDepartment of Geography, University of Arkansas, Fayetteville, Arkansas(Manuscript received 25 March 1987, in final form 25 September 1987)ABSTRACT A selected group of nine climate-sensitive tree-ring chronologies from old post oak trees are used to reconstructthe June Palmer Drought Severity Index (PDSI) from 1698 to 1980 for two large regions in northern andsouthern Texas. Analysis of tree growth and monthly climate variables indicate that the June PDSI is the mostrobust climate signal evident in these chronologies, and principal component analysis (PCA) reveals a northsouth geographic pattern in the relationships between the regional tree-ring time series. Serially random amplitudeseries from the first two significant eigenvectors of tree growth, which explain 65% of the total variance in thetree-ring data, were entered into stepwise multiple regression as predictors of regionally averaged June PDSI innorth and south Texas for the common interval 1931-80. The regression models explain 59% and 60% ofthe ~afiance in north and south Texas June PDSI, respectively, and both reconstructions are well verified againstindependent June PDSI data available on a statewide basis from 1888 to 1930. The weak persistence presentin the observed June PDSI series was added to the serially random tree-ring reconstructions prior to verification,using autoregressive modeling procedures. The mean and variance of June PDSI during the 50-yr period of meteorological observation (1931-80) appearto be representative of the last 283 yr, but significant changes in average June PDSI for Texas appear to haveoccurred over both 30 and ~ 90-yr time intervals. Moderate or more severe June droughts (PDSI < -2.0) havean estimated recurrence probability of over 90% each decade, and the risk of extreme June drought (PDSI~< -4.0) is estimated at over 50% every 15 yr in north Texas and every 10 yr in south Texas. The reconstructions faithfully reproduce the frequency domain properties of the actual June PDSI, and marginally significantspectral peaks are present at 2.3 yr and between 14 and 18.67 yr in both reconstructions. Significant interannualpersistence of June moisture extremes apparent in the statewide June temperature, precipitation, and PDSIdata from 1888 to 1982 is also present in both regional reconstructions from 1698 to 1980. The reconstructionsindicate that the risk for below average June moisture conditions increases to at least 65% in north and southTexas in the summer following a June drought (PDSI ~< -2.0). Interannual persistence is also indicated forJune wetness anomalies and may have some modest value in statistical forecasts of growing season moistureconditions in Texas.1. Introduction Recent analyses of the temporal and spatial characteristics of recorded drought data (Karl and Koscielny, 1982; Diaz, 1983) confirm the casual observations of many residents that Texas is one of the mostdrought-prone regions of the United States. Unfortunately, the economies of Texas and the entire countryhave become increasingly sensitive to drought due tohigher energy costs and growing pressure on limitedsurface and groundwater supplies from agricultural,industrial, and municipal interests (e.g., Karl and Koscielny, 1982). The 1980 summer heat wave, for example, cost the nation an estimated $16 billion, including some $1.5 billion in losses for Texas alone (Karland Quayle, 1981). Systematic meteorological observations in Texasbegan about 1880, but the continuity and spatial cov Corresponding author address: Dr. David W. Stahle, Departmentof Geography, University of Arkansas, Fayetteville, AR 72701.erage of the earliest data are not uniform (Griffiths andAinsworth, 1981). Only 24 moderate June droughts(or worse) are recorded on a statewide basis since 1888(Karl et al., 1983), and six of these occurred consecutively from 1951 to 1956 during the worst multiyeardrought episode in the recorded history of Texas. Thislimited sample of single and multiyear droughts foreven a drought-prone region may not be adequate todefine confidently the probability distribution ofdrought occurrence, thus limiting the application ofstatistical techniques for the characterization and longrange forecasting of drought events (Namias, 1981). Old climate-sensitive trees from undisturbed nativeforests offer one of the best means to extend annualdrought records for the past few centuries (Stockton,1975; Fritts, 1976). Analysis of these unique proxy treering series may provide insight into past climate variation and allow better estimation of drought probabilities (Wallis, 1977). In this study, nine climate-sensitive post oak (Quercus stellata) chronologies are usedto reconstruct the June Palmer Drought Severity Index(PDSI; Palmer, 1965) for two large regions in northernc 1988 American Meteorological Society60 JOURNAL OF CLIMATE VOLUME Iand southern Texas from 1698 to 1980. These longproxy drought records are then used to 1) describe animportant component of the growing season climatehistory of Texas for the past 283 years, 2) investigatethe. possible interannual persistence of moisture extremes, 3) estimate the recurrence probabilities fordrought and wetness categories of increasing severity,4) estimate the extent of possible bias in the 50 yearsof observed June PDSI data through comparisons withthe 283 yr reconstructed series, and 5) to compare thefrequency characteristics of the observed and reconstructed series and identify any long-term, quasi-periodic components that might provide practical or theoretical insight into drought recurrence in Texas.2. Previous research A number of recent studies have illustrated the useof tree-ring data for the reconstruction of regional precipitation and drought index series in both humid andsemi-arid regions of the United States (e.g., Cook andJacoby, 1977; Meko et al., 1980; Duvick and Biasing,1981; Stahle et al., 1985a). Stockton and Meko (1983)used tree-ring data to investigate drought history andthe periodic components of reconstructed annual precipitation in four regions of the Great Plains. Theirreconstruction for portions of Oklahoma and westernArkansas was based on five oak chronologies, includingdata from two of the sites used in this study. Stocktonand Meko (1983) documented a rather ill-definedrhythmic nature of drought recurrence in the GreatPlains since 1700, but low sample size (and probablyheteroscedastic tree-ring data) restricted their interpretations prior to 1750 in the Oklahoma-Arkansas sector. Biasing et al. (1988) used a larger network of tentree-ring chronologies (post oak and white oak, Q. alba,including earlier data from the four northernmost sitesused in this study) to reconstruct annual precipitationamounts for a large regional average of 12 climaticdivisions in northern Texas, southern Oklahoma, andwestern Arkansas from 1750 to 1980. They concludedthat the prolonged drought reconstructed from 1855to 1864 was the most severe regional drought event inthe 23'l-yr period, and that severe drought can be expected to return even in the absence of projected CO2warming. The present study differs from the foregoing dendroclimatic research in several fundamental respects,including (1) our use of five new tree-ring chronologies,the first available for south central Texas; (2) selectionof the nine predictor chronologies from an availablenetwork of 20 chronologies in Texas and southernOklahoma (Stahle et al., 1985b) on the basis of totallength and favorable chronology statistics (Fritts andShatz, 1975); (3) the use ofhomoscedastic and seriallyrandom principal component eigenvector amplitudesof tree growth as predictors of climate; (4) the specificfocus on Texas; (5) the reconstruction of June PDSI;(6) the time period of the reconstruction which extendsfrom 1698 to 1980; and (7) our analysis of the persistence and recurrence probabilities associated with JunePDSI.3. The data In the past 5 years, 47 new tree-ring chronologieshave been developed in the south central United States(Stahle et al., 1985b), adding to the greatly expandedtree-ring data base for North America. The networkfor the south central United States includes chronologies of five species which average from 250 to 300 yrlong, but the post oak chronologies of the prairie-foresttransition zone in Oklahoma and Texas often containthe highest quality climate signal of any tree-ring chronologies in the region, and are among the most climatesensitive tree-ring chronologies available anywhere(Stahle and Hehr, 1984). For this study, we have selected the nine best post oak chronologies available inor near Texas which date from at least 1698 (Fig. 1). The nine post oak chronologies are based on multipleincrement cores from 25 to 50 individual trees per site,which were processed and carefully cross-dated according to established procedures (Douglass, 1941;Stokes and Smiley, 1968). The exactly dated tree ringswere measured to 0.01 mm. Cross-dating and measurement accuracy were checked and confirmed withthe computer program COFECHA by correlating 35o__1~5- ~ . 9~5-__ I~G. 1. Site locations of the tree-ring chronologies used to reconstruct the June Palmer Drought Severity Index (pSDI) for the northand south Texas regions. The tree-ring data were calibrated with 50yr of actual June PDSI data (1931-80) averaged from the NorthCentral (NC) and Low Rolling Plains (LRP) climatic divisions in thenorth Texas region, and the South Central (SC) and Edwards Plateau(EP) divisions in the south. The numbered tree-ring collection sitesare 1) Quanah Mountain, Okla., 2) Mud Creek, Okla., 3) Lake Atbuckle, Okla., 4) Nichols Ranch, Tex., 5) Mason Mountain, Tex., 6)Brazos River, Tex., 7) Yegua Creek, Tex., 8) Lavaca River, Tex., and9) Coleto Creek, Tex.JANUARY 1988 DAVID W. STAHLE AND MALCOLM K. CLEAVELAND 61overlapping 50-yr segments of all measured series(Holmes, 1983). The computer program ARSTND wasused to detrend each ring measurement series and tocalculate the average site chronology as the biweightor robust mean value function of all available ringmeasurement series (Cook, 1985; Cook and Holmes,1985). Detrending and indexing of the ring measurement series are necessary to remove low-frequencyvariance associated with increasing age and circumference of trees and to eliminate differences in meangrowth rates among trees (Fritts, 1976). A trend of decreasing variance remained in some final chronologiesand was removed by fitting an inflexible spline to thevariance (Cook and Holmes, 1985). This variance trendis believed to result from declining growth vigor of theold trees sampled and from increasing sample size inthe most recent years (Stahle et al., 1985b). Divisional climate data available on a monthly basissince 1931 (Karl et al., 1983) were used to investigatethe post oak growth-climate relationship and to calibrate the tree-ring and climatic data (Fig. 1). Divisionalclimate averages avoid many problems sometimes associated with single station data such as record inhomogeneities and differing station microclimates, andusually provide better tree-ring reconstructions of climate (e.g., Blasing et al., 1981). Average monthly PDSI data are also available on astatewide basis for Texas back to 1888, although thenumber of individual reporting stations declines andthe stations become less evenly distributed within thestate during the earliest years of record (Karl et al.,1983). The statewide PDSI data from 1888 to 1930 areused for independent verification of the tree-ring reconstructions based on the 1931-80 calibration period.4. Post oak growth and climate Correlation and response function analyses (Fritts,1976) between the post oak chronologies and monthlyclimate variables have consistently demonstrated apositive growth response to precipitation and a negativeresponse to temperature throughout most of the year(Stable and Hehr, 1984; Biasing et al., 1988). The radialgrowth response to temperature and precipitation isstrongest during the growing season from Marchthrough July and appears primarily to reflect variationsin evapotranspiration demand. This suggests that ameteorological drought measure such as the PalmerIndex which integrates the combined effects of bothtemperature and precipitation would be a suitable climate variable for reconstruction. This is supported bycorrelations between the post oak chronologies andPDSI values for growing-season months, which areuniformly higher than correlations with monthly orseasonalized temperature and precipitation separately. Correlation and principal component analysis (PCA;Cooley and Lohnes, 1971) were used to determine ifthe tree-ring and climate data might be more effectivelyanalyzed as a single large-scale average or segregatedinto regional groups. All nine tree-ring chronologiesare well correlated and load highly with nearly identicalcoefficients of the same sign on the first eigenvector,which explains 47% of the total variance in the treering data and is believed to represent regional moistureanomalies. Two clear geographical areas are, nevertheless, apparent in the correlation matrix and in theloadings on the second eigenvector, which explains anadditional 18% of the variance in the tree-ring chronology network. The four northern chronologies sharethe highest intercorrelations and similar positive loadings on both eigenvectors, and the five southern chronologies share similar intercorrelations but load negatively on the second eigenvector. The observed PDSIdata for the four adjacent climatic divisions are all significantly correlated, but the two northern and southernmost divisions also share the greatest commonvariance. For these reasons, the tree-ring and climatedata were subdivided into the "north Texas" and"south Texas" regions. The monthly PDSI data for theNorth Central and Low Rolling Plains divisions wereaveraged from 1931 to 1980 for north Texas, and theSouth Central and Edwards Plateau divisions were averaged for south Texas (Fig. 1). The regionally coherent pattern in the first eigenvector of tree growth is consistent with the larger-scaleanalysis of monthly PDSI data for the continentalUnited States and is attributed to the unique synopticclimatology of the region centered on Texas and Oklahoma (Karl and Koscielny, 1982). The north-southpattern evident in the tree-ring and divisional PDSIdata for Texas is not readily apparent in the continental-scale analysis (Karl and Koscielny, 1982), but thestepwise multiple regression results reported belowprovide empirical support for the physical significanceof this pattern because the second eigenvector of treegrowth is a significant predictor of June PDSI in onlythe south Texas region. The strong regional gradientin annual potential evapotranspiration (Thornthwaite,1948) is parallel with the zero line separating oppositeloadings on the second eigenvector and may be onephysical component of the apparent difference betweennorthern and southern Texas. Correlation analysis between simple regionally averaged tree-ring chronologies and monthly PDSI datafor north Texas identified significant (P < 0.001) andhighly positive correlations during May, June, and July,with the highest correlation observed for June PDSI.Significant correlations were observed for the samemonths in south Texas, although the highest was observed during May. The slightly higher correlation withMay PDSI for the south Texas chronology probablyreflects phenological differences between northern andsouthern Texas, with an earlier growing season maximum in the south. June PDSI was selected for reconstruction in both areas because it represents the best composite of the62 JOURNAL OF CLIMATE VOLUME 1monthly drought signal strongly expressed in the regional tree-ring data. Also, the seasonal distribution ofrainfall is very similar for all four climatic divisionsunder consideration, with two broad rainfall peaks occurring from April through June, and Septemberthrough October (Griffths and Strauss, 1985). Thespring-early summer rainfall peak is most significantto regional agriculture. The June PDSI should providea good representation of the spring rainfall maximumbecause the Palmer Index represents a weighted averageof temperature and rainfall conditions for the currentand several preceding months. The identification of June PDSI as the most robustclimate signal in the Texas post oak data contrasts withearlier results reported by Biasing et al. (1988), whofound that a regional average of ten tree-ring chronologies was a better predictor of annual (July-June) precipitation than monthly PDSI values for a large areaof western Arkansas, Oklahoma, and northern Texas.The explanation for these differences may include ouruse of (1) five new tree-ring chronologies from southcentral Texas, (2) serially random tree-ring predictors:derived from PCA of the nine-chronology network,and (3) our reconstruction of smaller regional climateaverages. While the nine chronologies used in this studyare certainly well correlated with annual precipitationfor Texas, the June PDSI provides better regressionresults and may often be a better measure of agricultural drought in Texas than annual precipitation data.Late summer tropical storms may produce huge rainfallamounts that can significantly affect annual precipitation totals in Texas and obscure the actual economicimpact of droughts earlier in the growing season.5. Calibration and verification results The eigenvector amplitude series from the first twoprincipal components for the nine-chronology networkhad eigenvalues > 1.0 and were used to reconstructdrought in north and south Texas. The amplitudes wereused rather than the regional chronology averages because they consistently produced the best calibrationresults. Autoregressive (AR) modeling was used toidentify the autocorrelation structure of the tree-ringand climate data, which contain differing degrees ofpersistence. Serial persistence in the annual tree-ringdata and resulting eigenvector amplitude series is generally more pronounced than in the climate data andis largely attributed to the combined effects of climate,ecological, and physiological factors regulating thestorage and depletion of food reserves (Fritts, 1976;Meko, 1981; Cook, 1985). The first eigenvector amplitude series of regional tree growth was adequatelymodeled as an AR(I) process with a coefficient of0.171,'determined using the Akaike information criterion(Akaike, 1974). The second tree-ring amplitude serieswas modeled as an AR(2) process, with coefficients of0.282 and 0.113. The observed June PDSI for northTexas (1931-80) was modeled as an AR(0), but thenonsignificant autocorrelation at lag 1 was removedwith an AR(1)coefficient of 0.145. The south TexasJune PDSI series was modeled as an AR(1) processwith a coefficient of 0.235. The two serially random (AR modeled) tree-ringamplitude series were entered into stepwise multipleregression to predict AR modeled June PDSI for northand south Texas from 1931 to 1980 (Draper and Smith,1981; SAS Institute Inc., 1985). Only the first tree-ringamplitude series entered the regression model for northTexas June PDSI, while both amplitudes were significant predictors of June drought in south Texas(Table 1). The transfer function coefficients applied to the treering amplitudes from 1698 to 1980 are listed in Table1. The drought reconstructions were completed whenthe first-order persistence models identified for the observed June PDSI data from north and south Texaswere added into the transformed tree-ring amplitudeseries. Adding the observed climate persistence intothe serially random tree-ring estimates avoids potentialproblems related to the complicated persistence structure inherent in the tree-ring data and should providethe best possible PDSI estimates in the time and frequency domains (Meko, 1981). The explained climate variance is similar and highlysignificant for both regional reconstructions (Table 1).The reconstructions account for 59% and 60% of thevariance in June PDSI from 1931 to 1980 in north andsouth Texas, respectively (R~2aj, adjusted for loss of degrees of freedom; Draper and Smith, 1981). The Durbin-Watson test indicates that the regression residualsare not significantly autocorrelated in north Texas butare in south Texas (Table 1). The observed and reconstructed June PDSI data forthe north and south Texas regions are plotted from1931 to 1980 in Fig. 2, and the close fit between theactual and estimated time series is apparent in bothregions. Although the reconstructions explain a highlysignificant proportion of the observed climate varianceduring the calibration period, the regional reconstructions do not always reflect the moisture extremes, especially the positive extremes (Fig. 2). This is a frequently observed feature of dendroclimatic reconstructions usually attributed to the tendency forregression analysis to underestimate extreme values,and because tree growth only partially reflects climatevariation. Drought years are usually better estimatedthan wet years, because moisture deficits become moregrowth-limiting, while a host of nonclimatic growthlimiting factors such as inadequate nutrient supply,crowding, or disease may prevent a maximized growthresponse to favorable moisture conditions (Fritts, 1976;Duvick and Blasing, 1981). These considerations suggest that the June PDSI reconstructions may often represent a conservative estimate of the actual moistureextremes, particularly during abnormally wet years.JANUARY 1988 DAVID W. STAHLE AND MALCOLM K. CLEAVELAND ,, 63 TABLE 1. Transfer function coet~cients and the calibration and verification statistics derived for the tree-ring reconstructions of northand south Texas June PDSI. Average statewide June PDSI for Texas from 1888 to 1930 was used for the independent verification of bothcalibration models.Verification [ 1888-1930 (n = 43)] Calibration [1931-80 (n = 50)] Sign test t-test of Residual* 1st Dif. cross-productRegion b0 b] b: R~j autocorr. Corr. Corr. Pos. Neg. means RENorth Texas 0.03 0.88 -- 0.59*** 0.08 ns* 0.74*** 0.62*** 31'* 12 -4.10'** 4-0.54South Texas 0.04 0.92 -0.49 0.60*** 0.25* 0.66*** 0.49*** 32** 11 -3.20** +0.42 * = P < 0.05. ** = P ~< 0.01.*** = P ~< 0.001.* Durbin-Watson test for autocorrelation of residuals.* ns = not significant, P > 0.05. In spite of the underestimates of moisture extremes,the verification results for both regions are very favorable. Both reconstructions are well correlated with theindependent statewide PDSI data from 1888 to 1930(Table 1). The first difference correlations are also positive (Table 1), indicating significant coherence in thedirection of moisture anomaly changes from year toyear between the observed and reconstructed data. Thesign of the reconstructed and actual June PDSI departures from the mean are identical in over 72% of thellmO tNO ~SOUTH TEXAS FIG. 2. Observed (dashed lines) and reconstructed (solid lines) JunePDSI for the calibration period 1931-80 in north and south Texas.The Palmer Index scales drought and wetness anomalies as 0 to + 1,near normal; _+ 1 to +2, mild wetness or drought; +2 to _+3, moderatewetness or drought; _+3 to _+4, severe wetness or drought; greater than+4, extreme wetness; and less than -4, extreme drought.cases for both regions during the verification period (P< 0.01), and the t-tests on the cross-product means areboth significant (P < 0.01) indicating that reconstructedand observed values with different signs are usuallyclose to the mean. The reduction of error statistic(RE) is widely usedin the verification of dendroclimatic reconstructionsand is discussed by Gordon (I 982), Gordon and Leduc(1981), and Fritts (1976). Confidence limits are notcalculated for the RE, but Monte Carlo experimentssuggest that the approximate 95% confidence limit forn > 10 is RE >~ 0.0 (Gordon and LeDuc, 1981). Thetheoretical RE ranges from -oo to + 1.0, and a positiveRE indicates that the predicted values for the verification period are more accurate than hypothetical predictions based only on the observed data mean duringthe calibration period. The RE statistic for both regionsis strongly positive, particularly for the north Texasreconstruction. These verification tests on independentclimate data outside the calibration period demon.stratethe stability of the regression models and provide strongevidence concerning the overall accuracy of the reconstructions. The reconstructions are well verified in both regions,but the relatively lower verification statistics for southTexas may reflect certain data-specific problems unrelated to the true accuracy of the reconstruction (Table1). Both reconstructions were verified using statewideJune PDSI data available from 1888 to 1930 (Karl etal., 1983), but there was a greater concentration of reporting stations in northern Texas during the earliestyears of record (Griffiths and Ainsworth, 198 I). Consequently, the better verification results for north Texasmay simply reflect greater common variance betweenthe North Central-Low Rolling Plains average and thestatewide average prior to 1930 (Karl et al., 1983).However, the greater uncertainty concerning theregression coefficients for south Texas indicated by theautocorrelated residuals may also be reflected by thelowered verification for south Texas.64 JOURNAL OF CLIMATE VOLUME I TABLE 2. Reconstructed June PDSI from 1698 to 1980 for north and south Texas. The rank (from I to 10) of the ten driest years estimatedin north Texas is 1925, 1971, 1917, 1855, 1956, 1939, 1772, 1805, 1887, and 1790, and the ten wettest years by rank are 1833, 1740, 1919,1718, 1869, 1719, 1924, 1836, 1735, and 1867. The rank ofthe ten driest years in south Texas is 1925, 1971, 1917, 1857, 1790, 1967, 1956,1805, 1855, and 1887, and the rank of the ten wettest years is 1740, 1719, 1718, 1869, 1919, 1924, 1867, 1793, 1833, and 1900. YearDecade 0 1 2 3 4 5 6 7 8 9 North Texas June PDSI1690 -1.42 -2.211700 -0.30 0.66 2.99 - 1.31 -2.52 - 1.24 1.39 0.60 -0.71 -0.741710 -0.48 -0.50 -0.42 0.37 -2.32 -2.20 -1.32 -1.97 3.83 3.541720 2.16 2.35 -0.86 1.75 -1.64 -1.28. 0.39 -0.89 -1.87 -1.151730 -2.93 ' -1.49 1.28 0.86 -0.17 3.29 -2.09 -0.52 -1.13 1.121740 4.22 -0.33 -1.15 -1.69 0.03 -0.16 3.22 2.04 1.15 0.351750 - 1.28 - 1.04 -2.71 0.48 -0.66 -3.02 -0.18 -0.37 3.25 1.251760 2.62 0.58 1.36 -0.85 -0.96 -0.44 -0.42 -0.98 -1.41 -0.711770 0.22 0.07 -3.53 -1.39 -0.60 -1.56 -0.96 -1.82 -1.72 -0.091780 -0.93 -0.86 3.18 1.86 1.08 -2.15 -3.14 -0.72 1.89 -3.241790 . -3.31 -0.26 2.28 3.12 2.24 2.21 1.99 1.03 -1;12 2.861800 -0.02 -2.44 -0.02 0.83 -0.26 -3.40 -2.16 0.17 -2.14 1.221810 1.63 0.93 0.12 -0.06 0.78 0.32 -0.24 1.07 0.87 -0.751820 -2.14 0.15 -2.32 -0.03 -2.73 1.16 0.10 0.74 0.51 -0.391830 -0.83 -1.78 -1.11 4.52 2.55 -.1.16 3.50 0.29 0.82 -0.411840 0.48 - 1.64 -2.93 2.25 0.11 -0.91 -0.16 - 1.40 - 1.39 0.951850 1.57 1.22 -0.49 0.32 -0.13 -3.93 -0.50 -3.22 -0.01 -3.021860 -2.36 -1.30 -3.23 -1.85 -2.62 -0.26 -0.69 3.26 1.07 3.771870 0.01 1.64 -0.32 1.10 -1.10 0.90 -0.51 1.30 -0.17 -1.521880 -0.02 0.80 -0.52 0.01 1.24 2.14 -2.50 -3.38 1.63 -1.021890 2.48 1.53 -0.79 - 1.15 - 1.97 -0.99 - 1.74 0.02 0.69 0.321900 2.55 - 1.64 -2.30 2.51 -2.94 2.18 - 1.26 0.71 2.97 - 1.401910 -1.07 -1.60 0.50 -1.32 1.72 0.55 -1.25 -3.97 -2.32 3.991920 1.82 1.74 0.98 0.48 3.53 -5.67 1.34 0.11 -0.57 0.741930 0.49 0.33 1.71 0.70 -1.77 2.51 -0.18 -0.77 -0.22 -3.551940 -0.96 2.34 1.42 0.44 1.68 0.95 1.24 1.87 -0.12 0.521950 0.27 -0.90 -2.62 -3.07 - 1.61 -2.38 -3.83 0.11 0.21 - 1.381960 0.22 -0.64 0.11 -1.78 . -1.25 2.28 -0.34 -3.20 3.11 0.191970 1.22 -5.08 0.27 1.93 -0.86 2.72 0.31 0.88 -2.26 1.981980 -1.13South Texas June PDSI1690 -1.06 -2.211700 -1.26 0.17 1.70 -1.53 -2.33 -1.73 1.42 0.58 -0.16 -0.521710 0.14 -0.83 -1.13 -0.33 -3.05 -2.31 -1.66 -2.39 4.81 4.941720 3.40 3.15 -0.35 2.52 -1.06 -0.73 0.82 -0.57 -1.65 -1.851730 -3.40 -2.20 1.32 1.70 -0.11 3.38 ~ 1.94 0.02 -2.04 1.80- 1740 5.41 -0.68 -1.22 -1.79 -0.45 -0.17 2.38 2.09 0.99 0.521750 -2.09 -2.15 -2.80 0.32 - 1.54 -3.06 0.36 0.27 3.03 2.051760 2.11 0.83 1.60 -0.52 -1.53 -0.43 -0.04 -0.24 -0.65 0.141770 0.14 -0.20 -2.84 -0.79 -0.85 -2.26 -1.70 -2.12 -1.86 0.041780 -1.02 -2.13 3.12 1.34 1.30 -2.33 -2.74 -0.49 1.85 -3.561790 -4.11 -0.85 3.14 4.24 3.45 2.54 1.89 1.86 -0.78 3.321800 0.48 -.1.67 0.48 1.18 -0.42 -3.65 -2.59 0.26 -1.40 1.091810 2.05 0.83 0.83 0.31 1.50 0.60 0.01 0.25 1.70 -1.251820 -2.43 0.73 -2.54 -0.07 -2.82 0.58 -1.39 -0.04 0.14 -0.031830 - 1.38 - 1.75 -0.68 4.00 3.62 - 1.50 1.77 -0.82 0.80 -0.431840 0.03 -1.77 -2.88 1.31 0.23 -0.33 -0.36 -1.63 -1.94 0.931850 2.50 1.97 -0.50 0.02 0.09 -3.57 -0.08 -4.23 -0.25 -2.531860 -2.77 -0.40 -3.48 -1.60 -2.40 0.60 -0.34 4.26 2.44 4.641870 0.33 1.78 0.35 2.04 -0.54 1.50 -0.93 2.14 0.31 -0.941880 1.49 1.97 -0.07 -0.23 2.57 2.04 -2.04 -3:56 1.43 -1.261890 2.21 1.82 -0.98 -0.59 -2.96 0.50 -1.85 -0.75 0.54 0.421900 3.65 - 1.57 -2.93 2.39 -2.34 2.80 - 1.83 -0.60 1.74 - 1.401910 -0.44 -0.16 0.06 -!.43 1.52 -0.70 -2.32 -4.51 -1.35 4.571920 1.61 1.93 1.81 0.76 4.55 -5.96 1.50 -0.10 -0.97 1.361930 1.63 0.20 1.01 1.28 -1.44 3.02 0.88 -1.23 -0.01 -3.481940 -0.98 2.15 0.51 -0.30 2.00 0.78 1.90 1.96 0.17 0.941950 -0.08 -1.41 -2.27 -2.95 -2.67 -2.99 -3.92 -0.11 -0.24 -0.381960 -0.22 -0.95 0.04 -2.11 -0.88 2.03 1.12 -3.95 2.68 0.231970 1.84 -'5.40 0.26 2.24 -0.60 2.37 1.29 1.45 -2.48 2.391980 -0.06JANUARY 1988WET*4*2 OZ-2-4 -6DRY1698DAVID W. $TAHLE AND MALCOLM K. CLEAVELAND NORTH TEXASF1720 1740 1760 1780 1800 1820 1840 1860 1880 1900 1920 1940 1960 1980 YEAR65WET *6SOUTH TEXAS*4*2 0i&lZ-2-4 -6 DRY 1698 1720 1740 1760 1780 1800 1820 1840 1860 1880 1900 1920 1940 1960 1880 YEAR FIG. 3. Reconstructed June PDSI for north and south Texas plotted annually from 1698 to 1980, and smoothedwith a low-pass filter passing variance with a frequency of ~ ~8 yr (Fritts, 1976). The two regional reconstructionsare highly correlated over the 283-yr common period (r = 0.95, P < 0.0001).66 JOURNAL OF CLIMATE VOLUME 16. Texas climate history: 1698-1980 The June PDSI reconstructions for north and southTexas from 1698 to 1980 are listed in Table 2 andplotted in Fig. 3. The strong positive correlation between the two long reconstructions is readily apparent(Fig. 3) and suggests that the same large-scale circulation features are usually responsible for growing seasonclimate variability in both regions'. The statistical properties of both reconstructions arecompared with the observed June PDSI statistics inTable 3 and indicate the generally close agreement between the characteristics of the observed and reconstructed data. The variance statistics of the reconstructed series are consistently below the actual data,however, as was previously indicated by the calibrationresults. Because the reconstructions systematically underestimate the actual June PDSI extremes (Fig. 2; Table 3), the reconstructed values that are equivalent tothe actual levels of drought or wetness in terms of theirprobability of occurrence have been computed duringthe common period from 1931 to 1980. ReconstructedJune PDSI values of- 1.60, -2.39, and -3.11 in northTexas, and -1.58, -2.36, and -3.07 in south Texasare equivalent to the actual levels of moderate, severe,and extreme drought (observed PDSI = -2.0, -3.0,and -4.0, respectively). Reconstructed values of + 1.55,+2.32, and +3.01 in the north, and +1.52, +2.28, and+2.96 in the south are equivalent to actual levels ofmoderate, severe, and extreme wetness (observed PDSI= +2.0, +3.0, and +4.0, respectively). The relativelylarger differences between equal probabilities of observed and reconstructed wetness illustrate the asymmetry in growth response to surplus and deficit moisture. The reconstructed June PDSI series reveal severalinteresting aspects of Texas climate history (Table 2;Fig. 3). The most severe and protracted period of consecutive June droughts since 1698 appears to have occurred from 1951 to 1956 in both north and southTexas (although the drought actually began as early asJune 1947 in the South Central climatic division; Karlet al., 1983). The reconstructed June PDSI averaged-2.40 in north Texas and -2.70 in south Texas (bothequivalent to severe drought), compared to the actualregional averages of -3.38 and -3.94 over this entire6-yr period of record drought. This example illustratesboth the conservative bias of the reconstructed JunePDSI series and the need to assess the reconstructionsin probability terms equivalent to the actual June PDSI. The most severe uninterrupted sequence of Junedroughts since 1698 in Texas appears to have occurredin the 1950s, but the driest decades in both the northernand southern regions are estimated to have occurredfrom 1855-64, followed by the decades of 1950-59and 1772-81 (Table 2; Fig. 3). These dry decades wereall interrupted by some years of near normal to aboveaverage moisture conditions, but the temporarily improved conditions were probably not sufficient to mitigate the long-term environmental or economic impactof these historic drought eras. Four of the five wettest decades estimated since 1698are also identical in both regions, including the wettestdecade from 1791-1800. Most prolonged drought episodes were, in fact, preceded and/or followed by extended wet periods (Fig. 3), suggesting a weak oscillatory behavior in extended moisture anomalies in Texas. The prevalence of record droughts in recent decadesis suggested by the estimated occurrence since 1917 offive out of the six most severe June droughts over thelast 283 yr in north Texas, and five of the seven worstin south Texas (Table 2). Although record droughtsmay have been somewhat more prevalent over the lastfew decades, no significant differences in the mean orvariance of reconstructed June PDSI is apparent forthe nonoverlapping 50 yr periods running from 1731TABLE 3. Statistical characteristics of observed and reconstructed June PDSI in the north and south Texas regions.Observed PDSI Reconstructed PDSI Reconstructed PDSI1931-80 1931-80 1698-1980Statistic North Texas South Texas North Texas South Texas North Texas South TexasMean -0.18Standard deviation 2.42Maximum 5.43Minimum -4.94Range 10.37Median -0.48Serial correlation 0.16 ns*Skewness 0.21Kurtosis -0.42Normal Yes*Number of years 50-0.12 -0.17 -0.13 -0.13 -0.062.60 1.84 1.96 t .81 2.005.29 3.11 3.02 4.52 5.41-5.77 -5.09 -5.40 -5.67 -5.9611.06 8.20 8.42 10.19 11.370.61 0.15 0.02 -0.17 -0.080.24 ns 0.03 ns 0.07 ns 0.14' 0.20**-0.20 -0.50 -0.64 0.04 0.09-0.82 -0.03 -0.08 -0.11 -0.06Yes* Yes* Yes* Yes* Yes*50 50 50 283 283 * = P ~ 0.05.** =P<0.01.* ns = not significant, P > 0.05.JANUARY 1988 DAVID W. STAHLE AND MALCOLM K. CLEAVELAND 67to 1980. These observations and the statistical comparisons in Table 3 indicate that the 50 yr of observedJune PDSI data from 1931 to 1980 appear to be representative of June moisture conditions over the past283 yr. When time scales shorter and longer than 50 yr areconsidered, however, there does appear to be some evidence for climatic changes over Texas. If the reconstructed series are subdivided into 30-yr intervals, themeans of certain consecutive periods are significantlydifferent (t-tests, P < 0.10; Steel and Torfie, 1980; SASInstitute Inc., 1985). The period 1951-80 was drierthan the period 1921-50 in both north and south Texasreconstructions. The same two periods are also differentin the state average June PDSI (P < 0.10; Karl et al.,1983), indicating a high degree of short- and long-termvariability in the growing season climate of Texas during the twentieth century. These differences supportthe notion that the 30-yr "standard normal" climateperiods provide a reasonable measure of the current.growing season moisture regime, but are less suitableestimates of long-term drought conditions in Texas. Both reconstructions also indicate a very long-termpositive trend in June PDSI since the eighteenth century (Fig. 3). The mean June PDSI from 1698 to 1791was -0.265 and -0.275, compared with +0.193 and+0.364 from 1867 to 1950 in north and south Texas,respectively. These long-term differences in averageJune PDSI are statistically significant for both north(P < 0.10) and south Texas (P < 0.05). A positive trendsince the eighteenth century is also apparent in reconstructed June PDSI data for North Carolina (Stahle etal., 1988), and this tow-frequency signal is generallyconsistent with the long-term trend in annual precipitation data actually recorded for the continentalUnited States from 1851 to 1984 (Bradley et al., 1987).The strongly positive departures since the mid-1950sapparent in both the North Carolina PDSI and continental precipitation data are not evident in the Texasreconstructions, although they may be present in theactual June PDSI data for north and south Texas (Fig.2). Nevertheless, because the low-frequency signal isapparent in reconstructed June PDSI data for Texasand North Carolina since the eighteenth century, itmay be part of a regional climate trend unrelated toCO2 pollution until, perhaps, very recently. Examination of Fig. 3 reveals that drought, and toa lesser extent wetness anomalies, have been somewhatmore frequent in south Texas. In the last 283 years,66 moderate or worse droughts (PDSI equivalent~< -1.58) and 61 moderate or worse wet years (PDSIequivalent >~ +t.52) were reconstructed in south Texascompared with only 58 and 50 comparable droughtand wetness anomalies in north Te.xas (PDSI equivalents ~<-1.60 and >~ +t.55, respectively; Table 2).These intrastate comparisons are consistent with theobserved data and suggest that the summer climate ofsouthern Texas is both somewhat more variable anddrought prone than northern Texas.7. Interannual persistence 'of moisture extremes in Texas Careful examination of the reconstructed series suggests that June drought or wetness extremes tend to befollowed by similar dry or wet conditions during thenext summer (Table 2; Fig. 3). This apparent interannual persistence of growing season moisture anomalies is statistically significant, and the increased riskfor a particular moisture regime in the summer following an observed June moisture anomaly may havemodest value in statistical forecasts of Texas climate. The interannual persistence of summer moistureanomalies in Texas is tabulated in Table 4. The numberof years that were at least moderately wet or dry arelisted for both 283-yr reconstructions and for the 95yr of actual June PDSI recorded on a statewide basissince 1888 (Karl et al., 1983). The thresholds of moderately wet or dry years in the reconstructed series represent equivalent levels of moderate drought or wetnessactually recorded in the statewide data (see section 6above). The number of June moisture anomalies (PDSI~< -2.0 and >~ +2.0, or equivalents) that followedmoderate drought or wetness during the previous summer are then listed in Table 4. The statistical significance of the apparent interannual persistence was assessed using a normal approximation to the distributionof the number of joint or consecutive occurrences ofmoderate or more severe June drought or wetness (Table 4; Appendix). Finally, the increased likelihood ofa particular June moisture regime in the summer following a wet or dry June is also listed in Table 4 (i.e.,the observed percentage of at least moderately wet ordry Juries, and just above or below average June conditions for the following summer are compared withthe percentages expected to occur at random). The results in Table 4 indicate that the tendency fordrought and wetness anomalies to persist for two ormore growing seasons is statistically significant in mostcases, especially in the south Texas region. The chanceof experiencing a moderate or more severe Junedrought in south Texas increases to 39% in the summerfollowing a June drought, compared with an estimatedrandom chance of only 23% (Table 4). The potentialfor June conditions that are at least moderately wetalso increases in the summer following a wet June insouth Texas only, but the increase is not quite as large.This interannual persistence is reinforced when assessedonly in terms of above or below average moisture conditions in the year following a Jxtne moisture anomaly.The June PDSI was below average (<~0.0) in 66% and65% of all summers following a moderate drought orworse in north and south Texas, respectively (Table4). The results based on the reconstructed data in Table4 are probably conservative since several interannual68 JOURNAL OF CLIMATE VOLUME I TABLE 4. Interannual persistence of growing season moisture anomalies in north and south Texas, based on an actual statewide JunePDSI threshold of at least +2.0 and equivalent levels of reconstructed drought and wetness. The increased chance for a particular Junemoisture regime in the summer following a wet or dry June is compared with the expected chance assuming random interannua! occurrence(shown in parentheses). Percent occurrence for Years < -2.0 Standard deviation of actual and (expected)June PDSI Years following ~< -2.0 expected ~< -2.0series Time period ~< -2.0 actual and (expected) following ~< -2.0 < -2.0 <0.0DroughtActual statewide 1888-1982 24 11'* (5.8) 1.82 46 (24) 54 (51)North Texas 1698-1980 58 16' (11.7) 2.73 28 (20) 66 (54)South Texas 1698-1980 66 26*** (15.2) 2.99 39 (23) 65 (53)Wetness Percent occurrence for Years >~ +2.0 Standard deviation of actual and (expected)Years following >~ +2.0 expected >~ +2.0>~ +2.0 actual and (expected) following >~ +2.0 ~> +2.0 >~0.0Actual statewide 1888-1982 26 10' (6.8) 1.92 38 (27) 58 (49)North Texas 1698-1980 52 12 ns* (9.4) 2.51 23 (18) 60 (46)South Texas 1698-1980 65 22** .(14.7) 2.96 34 (23) 60 (47) * = P < 0.05. ** =P~<0.01.*** = P ~< 0.001.* = P~<0.10.* ns = not significant, P > 0.10.moisture anomaly pairs near the average and moderatethresholds were not included (Table 2). Significant interannual persistence of June PDSI extremes is also apparent in the actual statewide dataavailable from 1888 to 1982 (Table 4), in spite of therelatively short length of record. The ten moderatelywet Junes that followed wet conditions in the previoussummer occurred in seven separate episodes and appearto approximate the interannual persistence of wetgrowing seasons estimated for the past 283 yr in southTexas (Table 4). The interannual persistence ofdrought, based only on the 95 years of actual statewidedata'(Table 4), however, may be biased because the 11moderate or worse June droughts that followeddroughts during the previous summer occurred in justfive multiyear drought episodes (Karl et at., 1983).These multiyear drought episodes include the historicTexas drought of the 1950s when moderate to ex.tremely dry conditions in June lasted for six consecutive years. The reconstructed data do not appear to be seriouslybiased by the prolonged droughts of either the 1860sor 1950s, particularly in south Texas. The 26 moderateor worse June droughts in south Texas that followeda moderate or worse June drought during the previoussummer occurred in 15 separate episodes (Table 2),and the interannual persistence of drought in this regionremains significant when the 1860 and 1950 droughtsare omitted from the analysis-(P < 0.01). Interannualdrought persistence is not significant in north Texaswhen these two prolonged droughts are omitted, butthere is still a 60% chance for below average conditionsin the summer following a June drought. The interannual persistence of growing seasonmoisture extremes does not appear to be a spuriousresult of biological growth persistence in trees becausethe reconstructions are based on serially random principal component amplitude series of tree growth, andbecause significant interannual persistence of Junemoisture extremes is present in the actual statewidePDSI data from 1888 to 1982 (Table 4). Significantinterannual persistence of abnormal June temperatureand precipitation was also found on a statewide basisfor Texas (1888-1982). This indicates that the persistence in the PDSI data is not due simply to low interannual persistence arising from the strong month-tomonth autocorrelation built into the Palmer Index inorder to model the effects of climate on slower-changingsoil moisture conditions. [The prescribed monthly autocorrelation term in the Palmer model is +0.897(Palmer, 1965), resulting in an interannual autocorrelation that theoretically can be as high as +0.27.] Totest for persistence, the statewide temperature and precipitation data were divided into three equal classes(above average, average, and below average). These testsindicate significant interannual persistence for warmer.than average June temperatures (P < 0.001), and aboveaverage June precipitation (P < 0.05). The statewideJANUARY 1988 DAVID W. STAHLE AND MALCOLM K. CLEAVELAND 69June precipitation data are not normally distributed,but the test for persistence (Appendix) is not sensitiveto violations of normality. Finally, the look-ahead fea'tures of the Palmer Index associated with the termination of wet or dry spells (Karl, 1983) do not appearto be a serious problem because the evidence indicatespersistence over a full 12-month interval. Significant interannual persistence has been notedfor summer 700 mb height departures over the Southern Plains (Namias, 1960), and between current surfacetemperature in this region during spring-summer andthe regional 700 mb height departure of the previoussummer (Erickson, 1983). These results are based oncontinuous temperature and pressure variables representing different spring-summer averages, but nevertheless tend to substantiate the interannual persistence of summer moisture extremes reported here forTexas. The apparent persistence of midtropospheric pressure heights during summer is potentially an importantlink in the chain of physical causes that might be involved in the interannual persistence of summer moisture extremes over the Southern Plains. These physicalmechanisms may include prolonged sea surface temperature anomalies in the North Pacific and North Atlantic Oceans (Namias, 1960; Erickson, 1983), and thevarious potential feedback processes associated withlarge regions of abnormally wet or dry soil (Twomeyand Squires, 1959; Namias, 1960; Charney, 1975). Thespecific role of these and other potential physical causesof interannual persistence in summer moisture anomalies over Texas remains to be demonstrated, but theincreased potential for dry or wet conditions in thesummer following a June moisture anomaly may havemodest forecast value even in the absence of a welldocumented physical model.the reconstructions are usually longer than the intervalscalculated for the observed PDSI data (1931-80), es-'pecially at the more extreme levels. Consequently, therecurrence probability curves in Fig. 4 have been adjusted by multiplying each annual recurrence probability from 2 to 100 yr by the ratio of observed to reconstructed probabilities determined for each intervalduring the 1931-80 common period. This adjustmentexpresses the return curves based on the long reconstruction in terms of the actual June PDSI probabilitiesfor 1931-80, and is warranted in part because the 50yr common period appears to have been representativeof the drought and wetness recurrence probabilitiesover the past 283 yr. The adjusted return curves in Fig. 4 indicate a nearcertainty for the occurrence of a moderate-or-worsegrowing season drought each decade in both north andsouth Texas (June PDSI ~< -2.0, P = 0.90 and 0.92,respectively). There is a better than 50% chance for the8. Return time analysis ,.oReturn time analysis (Viessman et al., 1972) providesa longer statistical perspective on the recurrence of "drought or wetness anomalies in Texas. Recurrenceprobability curves have been calculated for six levels' ~.,of reconstructed drought or wetness severity over timeintervals from 2 to 100 yr (June PDSI = _+1.0, _+2.0, ~g,-. - +6.0). The return curves were first calculated using the reconstructed data from 1698 to 1980, and alsofor discrete 50-yr intervals from 1731 to 1980. Theseresults indicate little difference between the recurrence oprobabilities for the various levels of drought or wetnesswhen calculated for the entire 283-yr reconstructionor for any of the shorter 50-yr segments. These temporalcomparisons indicate again that the 50 years of JunePDSI observations from 1931 to 1980 appear to havebeen generally representative of the last 283 years inTexas.Because the reconstructions underestimate actualJune PDSI, however, the recurrence intervals based onNOflTH TEXASlO 2o ao 40 8o ~1o To iso Go 100 RETIJNN INTERVAL (YN)SOUTH TEXAS10 ~0 30 40 50 80 70 80 Go 100 RETURN INTERVAL (YN) FIG. 4. The adjusted return time probabilities are plotted for sixlevels of drought (solid lines) and wetness (dashed lines) in north andsouth Texas (June PDSI levels are +1.0 to +6.0 from top left tobottom fight). The estimated return curves are based on the 283 yrreconstructions, but were first multiplied by the ratio of observedto-reconstructed probabilities from 1931 to 1980 because ihe reconstructions underestimate actual moisture extremes. The + 1.0 and- 1.0 curves for south Texas are nearly identical.70 JOURNAL OF CLIMATE VOLUME 1occurrence of an extreme drought every 15 years innorth Texas and every 10 years in south Texas (JunePDSI ~<-4.0, Fig. 4). As these figures indicate, therecurrence probabilities for drought and wetnessanomalies are higher in south Texas, particularly at themost severe levels (Fig. 4). The higher risk for moistureextremes in south Texas is also suggested by the variance statistics in Table 3 and is consistent with thereturn probabilities calculated for the actual June PDSIdata used for calibration in north and south Texas.Unlike south Texas, however, the return probabilitiesfor wet anomalies are uniformly lower than the dryrisk in north Texas, and may reflect the greater distanceto moisture sources in the Gulf of Mexico. Interpretation of the return curves in Fig. 4 shouldbe qualified by several points. Return time analysisassumes independence of the June PDSI values, butthe interannual persistence discussed above suggeststhat this assumption may be violated in the case ofmoisture extremes. The high probabilities for the returnof dry or wet conditions over very short intervals maytherefore reflect in part development of interannualmoisture regimes such as the historic drought of the1950s, so the true return intervals between such regimesmight be lower. However, the return curves could beunderestimating the intervals between damagingdroughts simply because the dryness levels specified byPalmer (1965) and used in Fig. 4 are arbitrary thresholdvalues. The dry conditions indicated by PDSI valuesslightly above a given threshold may produce essentiallythe same negative impact, particularly when nonclirnatic variables such as the strength of the agriculturaleconomy or the buffering capacity of the water supplysystem are considered. Finally, the return curves are strictly applicable onlyto the large north and south Texas regions investigated.Calculations based on the individual climatic divisions,however, do not indicate any major differences compared to the return probabilities based on the largerregional averages. The South Central division has thehighest recurrence probabilities for June moisture extremes, and the North Central division has the lowestof the four Texas climate divisions investigated.9. Spectral analysis Spectral analyses (Jenkins and Watts, 1968) indicatethat the tree-ring reconstructions of June PDSI containmarginally significant spectral peaks at frequencies nearthe quasi-biennial pulse and the lunar nodal tide. Whilethese results are in general agreement with certainquasi-periodicities identified in western droughts(Stockton et al., 1983; Currie, 1981, 1984) and easternair temperatures in North America (Curde, 1984), theydo not represent an a priori hypothesis test and canonly be regarded as inconclusive statistical results (e.g.,Pittock, 1983). Cross-spectral analyses between the observed andreconstructed June PDSI series during the calibrationperiod indicate that the tree-ring reconstructions faithfully reproduce the frequency domain properties of theactual June PDSI data. The estimated power spectraof the observed and reconstructed series for the 193180 period are plotted in Fig. 5a, and broad spectralpeaks with periods between 3 and 4 yr, and 12 to 24yr are present in all series. These similarities in the lowfrequency variance of June PDSI in north and southTexas are certainly to be expected given the sharedregional climate (Pittock, 1983). The squared coherency remains well above the 95%confidence level and demonstrates the highly significantagreement between the observed and reconstructed series at all frequencies in both north and south Texas(Fig. 5b). The observed and reconstructed series in bothregions are in phase at all frequencies, and the lack ofpronounced slope in the gain spectra indicates that bothreconstructions represent unbiased estimates of theactual June PDSI series in the frequency domain(Fig. 5c). The notable consistency between the actual and reconstructed PDSI records during the calibration periodindicates that the reconstructions should provide reliable estimates of any low-frequency or quasi-periodiccomponents in growing season droughts over the past283 yr in Texas. Significant spectral peaks (P < 0.05)are present in the power spectra for both long-termJune PDSI reconstructions at 2.3 and between 14 and18.67 yr, explaining 4% and 16% of the reconstructedvariance, respectively (bandwidth = 0.045 cpy, Fig. 6).The power spectra presented in Figs. 5 and 6 werecomputed using a Hamming window (IMSL Inc.,1982). The weak red noise persistence component observed in the actual PI~SI data was added to the seriallyrandom tree-ring reconstructions, so the statistical significance of the spectral peaks identified in both typesof data were evaluated assuming a first-order autoregressive null continuum. The apparent concentration of low-frequency variance near 2.3 yr may be related to the quasi-biennialpulse often present in meteorological data (Barry andPerry, 1973), although this frequency component isnot prominent in the observed or reconstructed datafrom 1931 to 1980 (Fig. 5a). The broad spectral peakbetween 14 and 18.67 yr resolves to a significant peakbetween 17.5 and 20 yr at the narrowest bandwidth(BW = 0.18 cpy; P < 0.01 for the 17.5-yr peak in southTexas; P < 0.05 for the 20-yr peak in south Texas, andfor the 17.5 and 20-yr peaks in northern Texas). Atthis level of spectral resolution, the broad spectral peakincludes the 18.6-yr period in the lunar nodal tide(Currie, 1981; Currie and Fairbridge, 1985). Other treering studies of drought in the Great Plains have foundquasi-periodicities in this frequency range that aretemporally and spatially complex, but may be relatedJANUARY 1988 DAVID W. STAHLE AND MALCOLM K. CLEAVELAND 71.oeNORTH TEXAS SOUTH TEXASP~IW)O (y#) I~l.(X) (YII)4.0 ~.e7 2.0 m 24.0 72.0 8.0 4.0~NCy (CY~8 YR-1)FREQUENCy (CYCLES YR-~) NORTH TEXAS ~RIOO (YR)24,0 12,0 8.0 4.0 2,e7..................... !$3 _eL_. ............FREO4JENCy (CYCLES YR' 1) SOUTH TEXAS PERIOD (YR) t _-_-~----~ _ __0-~9-__~i.' _ .................... ~_s3_eL_ ............FREQUENCY (CYCLES YR-1)NORTH TEXAS KRtOO (YR) SOUTH TEXAS P~mOo (YR)24.0 12.0 e,o 4.0 2.i? 2.0FREQUENCy (CYCLES YR-1) FNECIIJENCY (CYCLES YR-1) FIG. 5. Cross-spectral analysis between observed and reconstructed June PDSI for north and south Texas, 1931-80. (a) Relativespectral density for observed (dashed lines) and reconstructed (solid lines) June PDSI calculated using a Hamming window with 12 lags,resulting in 10 deg of freedom and bandwidth of 0.105 cycles per year (asterisk represents P < 0.05); (b) squared coherency betweenobserved and reconstructed June PDSI along with the 95% and 99% confidence limits; (c) gain spectra for observed and reconstructedJune PDSI.to solar and/or lunar influences (Stockton et at., 1983;Meko et at., 1985). A t-test for differences between means of reconstructed June PDSI for the 9 years of lunar nodal maxima and nine minima since 1800 (Curfie, 1984) indicates a marginally significant difference in moistureregimes for north Texas (P < 0.10), but no significantdifference in south Texas (P < 0.23). The true significance of the apparent low-frequency signal in TexasJune PDSI remains obscure in the absence of an a priorimodel predicting periodicities at specific frequencies(Mitchell et al., 1966), and because the spectral peakis not significant when tested between 1870 and 1980.The analyses of western droughts and eastern air temperatures reviewed by Pittock (1983), however, alsoindicate a breakdown in oscillations between periodsof approximately 18.6 and 22 yr in the late nineteenthcentury. The spectral analysis of long-term June PDSI is adequate to suggest that the search for possible lunarnodal tidal influences on climate might be expandedinto Texas, which is certainly not unreasonable considering the general spatial coherence of climate acrossthe central and southwestern United States where these72 JOURNAL OF CLIMATE VOLUME I NORTH TEXAS24.0 12.0 1.0 4.0O .08 .t0 .~S .20 .26 .30 .36 .80 .4S .SO FREQUENCY (CYCLES YR - ~) SOUTH TEXAS FEmO0 (YR)24.0 t~.0 1.0 4.0 2.er 2.0FNEQ4JENCY (CYCLES YR-T) F-IG. 6. Relative spectral density estimated for the 283-yr reconstructions for north and south Texas (1698-1980). The spectral estimates were calculated using a Hamming window with 28 lags, resulting in 25 degrees of freedom and bandwidth of 0.045 cpy. The95% and 99% confidence limits were based on a tint-order autoregressive null continuum. Spectral peaks significant at the 95% levelare present in both reconstructions at 2.3 and between 14 and 18.67yr.possible effects have previously been identified. Thedetection of lunar influences over Texas would alsoappear to be generally consistent with Curde's (1981,1984) hypothesized mechanism for orographic modulation of an 18.6-yr quasi-standing atmospheric waveby the Rocky Mountains. Whatever physical or stochastic mechanisms might be responsible for the apparent low-frequency component of reconstructed JunePDSI in Texas, it is probably not due to .recurrent infestation of the sample trees by 17-yr periodical cicadas(Magicicada spp.) because the nine prairie-border postoak sites appear to be well outside of the known distribution of this species (Simon, 1979).10. Summary and conclusions Nine climate-sensitive post oak chronologies havebeen used to reconstruct the June PDSI in northern and southern Texas. The well-verified reconstructions explain some 60% of the June PDSI variance during the calibration period (1931-80) and indicate that the multiyear drought of the 1950s was the most severe continuous drought episode since 1698. The three driest decades by rank appear to have been 1855-64, 1950-59, and 1772-81. The 50 years of meteorological observations from 1931-80 are generally representative of the mean, variance, and recurrence probabilities of June PDSI over the past 283 years, but significant changes in average June PDSI appear to have occurred in Texas over time intervals shorter and longer than 50 years. Significant interannual persistence of June moisture extremes indicates that the drought hazard may increase in the summer following a moderate-or worse June drought. June wetness extremes exhibit similar persistence, but to a lesser degree. The low fre quency variation in June PDSI indicated by the broad spectral-peak between 14 and 18.67 yr in the two re constructions may be weakly related to periods of maxima and minima in the lunar nodal tide in north ern Texas. Considering the profound economic and environmental impact of the record drought from 1951 to 1956(Griffiths and Ainsworth, 1981; Karl and Quayle,1981), it is reassuring to conclude that the severity ofthis prolonged drought has evidently not been exceeded -in Texas since 1698. It is also fortunate from a longterm planning perspective that the period of meteorological observations from 1931 to 1980 appears tohave been representative of growing season climateconditions over the past 283 years, because the historical expectations of rainfall and streamflow in Texasare largely based on this time period. The estimatedoccurrence of the three driest decades in the 1770s,1860s, and 1950s very loosely suggests that the approximate frequency of these extreme episodes Of prolonged drought may be about once per century. The interannual persistence of June moisture ex tremes in Texas is evident in the actual statewide tem perature, precipitation, and PDSI data available from 1888 to 1982, but this relatively short record may be biased by the prolonged drought of the 1950s. The long tree-ring reconstructions reported here substantiate this apparent interannual persistence and suggest a near 65% risk for below average June moisture conditions in Texas for the summer following a moderate June drought. Persistence in tree-ring data has usually been treated as a problematic~al function of internal tree physiology and stand dynamics, but the autoregressive modeling approach suggested by Meko (1981) has per mitted the recovery of evidence concerning the inter annual persistence of climate in Texas that may have some modest forecast value. Annual resolution and the potential for direct quan' titative reconstruction of monthly, seasonal, or annualclimate variables are the principal virtues of proxy treering data. The lack of widespread spatial arrays of treeJANUARY 1988 DAVID W. STAHLE AND MALCOLM K. CLEAVELAND 73ring data, however, has been a severe practical limitation confronting dendroclimatic applications to largescale climatic problems. This study extends the networkof high quality, climate-sensitive tree-ring chronologiesin the southern Great Plains. The coupling of thesenew tree-ring data with the expanding geographic arrayfor the Northern Hemisphere (Stockton et at., 1985)has the potential to increase understanding of the climate system. Acknowledgments. This research was supported bythe Climate Dynamics Program of the National ScienceFoundation, grants ATM-8006964, ATM-8120615,ATM-8412912 and ATM-8612343. The expert adviceand assistance of Edward R. Cook have considerablyimproved this research, and Colm O'Cinneide helpedus assess the statistical significance of interannual persistence. We also thank Rita C. Berg, T. J. Blasing,Randall S. Cerveny, James E. Dunn, Glenn L. Evans,Thomas P. Hadan, Graham G. Hawks, John G. Hehr,and Robert Mauermann for their help, and the following landowners for permission to collect tree-ring samples from their property: Mr. Herman Leifeste, Dr.Dwight Nichols, Wichita Mountains National WildlifeRefuge, Chickasaw National Recreation Area, U.S.Army Corps of Engineers, and the Corpus ChristiPower and Light Company.APPENDIX Significance Test of Runs The statistical significance of interannual persistencewas tested by comparing it with the expected runs oftwo drought categories in a random normal distribution(O'Cinneide, personal communication, 1987) as follows: M(M- 1) Eo(T) - N (AI)where E0 is the expectation in a random normal distribution, T the number of runs of a category (PDSI<~ -2.0 following ~< -2.0, or PDSI >~ +2.0 following>~ +2.0), M the total number of occurrences of a category in a series, and N the number of years in theseries. M(M -'1)Vo( T) N x[1 +(M-N_ll)(M-2) M(5-1)] (A2)where V0 is the variance of expected occurrences inthe number of runs (T), and all other variables are asdescribed above. The significance test is T- Eo( r) Za= l[-~o(T) (A3)where za is the z-score, the variables are as describedabove, and the null hypothesis is that given the numberof occurrences of a condition in a period, the times ofoccurrence are completely random. REFERENCESAkaike, H., 1974: A new look at statistical model identification. IEEETransactions on Automatic Control, AC-19, 716-723.Barry, R. G., and A. H. Perry, 1973: Synoptic Climatology. Methuen, 555 pp.Biasing, T. J., D. N. Duvick and D, C. West, 1981: Dendroclimatic calibration and verification using regionally averaged and single station precipitation data. Tree-Ring Bull., 41, 37-43. , D. W. Stahle and D. N. Duvick, 1988: Dendroclimatic recon strnction of annual precipitation in the southcentral United States from 1750 to 1980. Water Resour. Res., (in press).Bradley, R. S., H. F. Diaz, J. K. Eischeid, P. D. Jones, P. M. Kelly and C. M. Goodess, 1987: Precipitation fluctuations over Northern Hemisphere land areas since the mid-19th century. Science, 237, 171-175.Charney, J. G., 1975: Dynamics of deserts and drought in the Sahek Quart. J. Roy. Meteor. Soc., 101, 193-202.Cook, E. R., 1985: A time series analysis approach to tree-ring stan dardization. Ph.D. dissertation, University of Arizona, 165 pp.--, and G. C. Jacoby, 1977: Tree-ring--drought relationships in the Hudson Valley, New York. Science, 198, 399-401.--, and R. L. Holmes, 1985: Users manual for program ARSTAN. Lamont-Doherty Geological Observatory, Columbia University, 29 pp.Cooley, W. W., and P. R. Lohnes, 1971: Multivariate Data Analysis. Wiley, 364 pp.Curde, R. G., 1981: Evidence for 18.6-year signal in temperature and drought conditions in North America since A.D. 1800. J. Geophys. Res., 86, 11 055-11 062.--, 1984: Periodic (18.6-year) and cyclic (1 l-year) induced drought and flood in western North America. J. Geophys. Res., 89, 7215 7230. , and R. W. Fairbridge, 1985: Periodic 18.6-year and cyclic 1 l year induced drought and flood in northeastern China and some global implications. Quaternary Science Reviews, 4, 109-134.Diaz, H. F., 1983: Some aspects of major dry and wet periods in the contiguous United States, 1895-1981. J. Climate Appl. Meteor., 22, 3-16.Douglass, A. E., 1941: Cmssdating in dendrochronology. J. Forestry, 39, 825-831.Draper, N. R., and H. Smith, 1981: Applied Regression Analysis. 2nd ed., Wiley, 709 pp.Duvick, D. N., and T. J. Biasing, 1981: A dendroclimatic reconstruc tion of annual precipitation amounts in Iowa since 1680. Water Resour. Res, 17, 1183-1189.Erickson, C. O., 1983: Hemispheric anomalies of 700 mb height and sea level pressure related to mean summer temperatures over the United States. Mon. Wea. Rev., 111, 545-561.Fritts, H. C., 1976: Tree-Rings and Climate. Academic Press, 567 Pp. , and D. J. Shatz, 1975: Selecting and characterizing tree-ring chronologies for dendrodimatic analysis. Tree-Ring Bull., 35, 31-40.Gordon, G. A., 1982: Verification of dendroclimatic reconstructions. Climate from Tree-Rings, M. K. Hughes, P. M. Kelly, J. R. Pilcher and V. C. LaMarche, Jr., Eds., Cambridge University Press, 58-61. , and S. K. LeDuc, 1981: Verification statistics for regression models. Preprints Seventh Conf. Probability and Statistics in Atmospheric Science, Monterey, Amer. Meteor. Soc., 129-133.Griffiths, J. F., and G. Ainsworth, 1981: One Hundred Years of Texas Weather 1880-1979. Office of the Texas State Climatologist, Texas A & M University, 205 pp. , and R. F. Strauss, 1985: The variety of Texas weather. Weath erwise, 38, 137-141.JOURNAL OF CLIMATE VOLUME1Holmes, R. L., 1983: Computer-assisted quality control in tree-ring dating and measurement. Tree-Ring Bull., 43, 69-78.IMSL Inc., 1982: IMSL Library Reference Manual, Edition 9, Vol. 2. IMSL, Inc., FTFREQ-1 to FTFREQ-5.Jenkins, G. M., and D. G. Watts, 1968: Spectral Analysis and Its Applications. Holden-Day, 525 pp.Karl, T. R., 1983: Some spatial characteristics of drought duration in the United States. J. Climate Appl. Meteor., 22, 1356-1366. , and R. G. Quayle, 1981: The 1980 summer heat wave and drought in historical perspective. Mon. Wen. Rev., 109, 2055 2073. , and A. J. Koscielny, 1982: Drought in the United States: 1895 1981. J. Climatol., 2, 313-329. , L. K. Metcalf, M. L. Nicodemus and R. G. Quayle, 1983: Statewide average climatic history, Texas 1888-1982. Historical Climatology Series 6-1, National Climatic Data Center, 39 pp.Meko, D. M., 1981: Applications of the Box-Jenkins methods of time series analysis to the reconstruction of drought from tree rings. Ph.D. dissertation, University of Arizona, 149 pp. . , C. W. Stockton and W. R. Boggess, 1980: A tree-ring recon struction of drought in southern California. Water Resour. Bull., 16, 594-600. , -- and T. J. Blasing, 1985: Periodicity in tree-rings from the Corn Belt. Science, 228, 381-384.Mitchell, J. M., Jr., B. Dzerdzeevskii, H. Flohn, W. L. Hofmeyr, H. H. Lamb, K. N. Rao and C. C. Wallen, 1966: Climarc change. Tech. Note No. 79, World Meteorological Organization, 79 pp.Namias, $., 1960: Factors in the initiation, perpetuation and termi nation of drought. Extract of Publ. No. 51 of the I.A.S.H. Com mission of Surface Waters, 81-94. [Available from I.A.S.H., UNESCO, Paris, France.] ,1981: Severe drought and recent history. Climate and History, R. I. Rotberg and T. K. Rabb, Eds., Princeton University Press, 117-132.Palmer, W. C., 1965: Meteorological drought. Res. Pap. No. 45, U.S. Weather Bureau, 58 pp.Pittock, A. B., 1983: Solar variability, weather and climate: An update. Quart. J. Roy. Meteor. Soc., 109, 23-55. SAS Institute Inc., 1985: SAS User~ Guide.' Statistics, Version 5 ed. SAS Institute Inc., 956 pp.Simon, C. M., 1979: Evolution of periodical cicadas: Phylogenetic inferences based on allozymic data. Systematic Zoology, 28, 22 39.Stahle, D. W., 1988: North Carolina climate variations reconstructed from tree rings: A.D. 372 to 1985. Science, (in press).--, and J. G. Hehr, 1984: Dendroclimatic relationships of post oak across a precipitation gradient in the southcentral United States. Ann. Assoc. Amer. Geogr., 74, 561-573.--, M. K. Cleaveland and J. G. Hehr, 1985a: A 450-year drought reconstruction for Arkansas, United States. Nature, 316, 530 532. , J. G. Hehr, G. G. Hawks, M. K. Cleaveland and J. R. Baldwin, 1985b: Tree-Ring Chronologies for the Southcentral United States. Tree-Ring Laboratory and Office of the State Climatol ogist, University of Arkansas, Fayetteville, 135 pp.Steel, R. G. D., and J. H. Torrie, 1980: Principles and Procedures of Statistics. 2nd ed., McGraw-Hill.Stockton, C. W., 1975: Long-term streamflow records reconstructed from tree-rings. Papers of the Laboratory of Tree-Ring Research, No. 5, University of Arizona Press, 111 pp. , and D. M. Meko, 1983: Drought recurrence in the Great Plains as reconstructed from long-term tree-ring records. J. Climate Appl. Meteor., 22~ 17-29.--, J. M. Mitchell, Jr., and D. M. Meko, 1983: A reappraisal of the 22-year drought cycle. Weather and Climate Responses to Solar Variations, B. M. McCormac, Ed., Colorado Associated University Press, 507-515.--, W. R. Boggess and D. M. Meko, 1985: Climate and tree rings. Paleoclimate Analysis and Modeling, A. D. Hecht, Ed., Wiley, 71-151.Stokes, M. A., and T. L. Smiley, 1968: Introduction to Tree-Ring Dating. University of Chicago Press, 73 pp.Thornthwalte, C. W., 1948: An approach toward a rational classifi- cation of climate. The Geographical Rev., 38, 55-96.Twomey, S., and P. Squires, 1959: The influence of cloud nucleus population on the microstructure and stability of convective clouds. Tellus, 11, 408-411.Viessman, W., T. E. Harbaugh and J. W. Knapp, 1972: Introduction to Hydrology. Intext Educational Publishers, 415 pp.Wallis, J. R., 1977: Climate, climatic change, and water supply. EOS, Trans. Amer. Geophys. Union, 58, 1012-1024.

Save