The Consistency of Modeled and Observed Temperature Trends in the Tropical Troposphere: A Comment on Santer et al (2008)

SHORT COMMUN I CAT ION The Consistency of Modeled and Observed Temperature T rends in the Tropical Troposphere: A Comment on Santer et al (2008) Stephen McInt yr e Climate Audit Toronto, Ontario Ross McKitrick University of Guelph Guelph Ontario ABSTRACT Santer et al (2008) (S08) compared climate model s and observations in the tropical troposphere and reported that “there is no longer a serious d iscrepancy between modeled and observed trends in tr opical lapse rates.” The y found no statisticall y significant differences between modeled (ens emble mean) trends and observed tr ends at the T2LT and T2 la y ers, and the y found no significant difference between observed a nd modeled surface-minus-troposphere lapse rates. How ever they onl y used data over t he 1979-1999 period. Using the S08 metho dology on up-to-date data, we find a statistically significant discrepanc y between observations and models with respect to trends in the UAH data, as well as lapse rate trends comparin g either RSS or UAH to the HADCR UT3v land-ocean surface trend. KEY WORDS Tropospheric temperature changes, cli mate model evaluation, statis tical significance of trend differences, tropical l apse rates, differential warming of surface and t roposphere. The Consistency of Mod eled and Observed Temperature Trends in the T ropical Troposphere: A Comment on Santer et al (2008) Introduction Santer et al (2008) (“S08 ”) analyzed what the y termed a “potential inconsistenc y” between modelled and o bserved trends in tropical lapse rates using “new observational estimates of surface and troposph eric temperature changes”, concludin g that “there is no longer a serious discrepanc y between modelled and observed trends in tropical lapse rates”. Their analyses involv ed both satellite and radiosonde data. Ou r comment pertains to the satellite data, which has been a source of ongoing controv ersy and which is worth examining in its own ri ght. In respect to the satellite data, S08 car ried out separate analyses comparing obse rved trends to the mean of an ensemble o f 49 climate model runs and to the overall populati on of models. We restrict our comm ent to a re-assessment of their conclusions that were based on results involving the model ens emble mean. Similar exami nation of conclusions for the individual model runs would al so be worthwhile. S08 placed considerable i mportance on results for the “new ” RSS satellite data set, of which the y stated that “the surface w arming is invariabl y a mplified in the tr opical troposphere, consistent wit h model results.” For the UAH satell ite data set, they reported both that the observed trop ical lapse rate trends are not significantl y different from models and that there wa s no statisticall y significant di fference between the trends of the model ensemble mean an d observations at the two t ropospheric levels (T2LT, T2). These results have been relied on several rec ent assessment reports (e.g. CCS P 2009). Although S08 stated that they used “new observational esti mates,” and criticized Douglass et al (2007) for their use of older datasets, S 08’s own analysis was based on observations ending in 1999, even thou gh data were available to them up t o the end of 2007. We show herein th at with the addition of a decade of observational data, ke y S08 claims do not hold up, including the important comparison involvin g the RSS- HADCRUTv2 lapse rate. Methodology S08 claimed that there were no statist ically significant differences betw een the observed trends in the tropical T2LT and T2 levels of the two major satellit e data sets: UAH (Spencer and Christ y 1990, Ch risty et al. 2007) and RS S (Mears et al 2003, Mears and Wentz 2005) and th e corresponding ensemble means from their 49-run ensemble of models. Their test was ca rried out using data ending in 1999. (All di scussions in t his article are restricted to tropical data.) S08 used a modified t-tes t (their d 1 * ) to test the statisti cal significance of the difference between the model ensem ble trend, denoted >> << m b , and the observed trend, denot ed 0 b , defined in their equation (12) as follows: 2 0 2 1 0 1 ) ( ) ( * b s b s b b d m n m m + > < − >> << = (1) where 2 0 ) ( b s is the estimated variance of 0 b , 2 ) ( > < m b s is the square of the ‘inter-mod el standard devi ation of ens emble-mean t rends’ and m n is the number of models (19 i n S08). Values for each quantit y were given in their Table 1. In their calculation of the standard error of the observed trend, ) ( 0 b s , S08 adjusted the number of de grees of freedom of the trend residuals to take account of first-order (AR1) autocorrelation, usi ng a method described in Karl et al (2006) (“CCS P”), the method itself dating back at le ast to Quenouille (1952). S08 co nsidered significance using a two-si ded t-test at 10, 5 and 1% levels (see their Tables II I and V). We successfully emulat ed S08 results in their Table III using data ending in 1999 (see Supporting Information). S08 (see their Supplemen tary Information) observed that it was reasonable to ext rapolate model trends for compari son with post-1999 observations, a proto col that we adopt here using our implementatio n of their algorithm to com pare observational data ending in 1999, 2007 and Ju ne 2009 to the model ensemble t rends reported in S08. S08 also compared the mo del ensemble surface-mi nus-T2LT lapse rate trend to the corresponding observed l apse rate, using the two major satellite data sets a nd each of four surface data sets: one lan d-and-ocean series (Brohan et al 2006 - H adCRUT3v) and three sea surface temperature (S ST) series (Rayner et al 2003 - Had ISST; Smith and Reynolds 2005 - ERSSTv2; Smith et al 2008 - ERSSTv3). T2 comparisons we re not reported. We applied our emulation of their m ethodology to the comparison with upd ated surface data. The ERSSTv3 versi on used in S08 was disconti nued shortly after publication of S08 and replaced b y a new version (ERS STv3b) that did not incorporate as much satellite data (Smith and Re ynolds 2008). We use ERSSTv3 b here. ERSSTv2 is not updated as promptly as ERSSTv3b a nd thus our results for this comparison include data onl y to April 2009. S08 did not in clude a comparison over land, a comparison which we mak e here because of the differences betwee n land-and-ocean and ocean-onl y results. We examined three major tropical land data sets: C RUTEM3v (Brohan et al. 2006), GISS wit h 250 km smoothing (Hansen et al 2001) and NOAA-GHCN (Peterson and Vose 1997). Results T2LT and T2 Trends S08 reported that the mo del ensemble trend for the T2 LT level was 0.215 deg C (T2: 0.199). Using data to J une 2009, the trend for RSS T2LT was 0.140 deg C/decad e (T2: 0.104) and for UAH T2LT was 0.051 deg C/decade (T2: 0.021) and. Th e model-versus- observation trend differe nce was significant at the 5% l evel under a two-sided t-test for two of the four tests (thos e involving UAH) - see Table 1. The mod el-versus-observed trend difference for one RSS test (T2) was significant under a one-sided t-test at a 10% level. Trends in Su rface-Troposphere Lapse Rates S08 observed that trend t ests involving surface-T2LT difference series are more “stringent” than simple tr end tests, because differencing removes much of t he common variability, thereb y reducing AR1 autocorrelation in the trend residuals. Combined Land-and-Ocean: S 08 reported that the model ensemble trend fo r the surface- T2LT difference series over land- and-ocean was -0.069 deg C/decade (T2: -0.053 ) i.e. troposphere warming more than surf ace. For their comparison of obse rvations over land- and-ocean, S08 compare d satellite data to HADCR UT3v surface data. Using data to J une 2009 (see Table 2), all four lapse rat e trends exhibit statis tically significant differences from models under a two-sided t-test at a 5% level (at a 1% level i n 3 of the 4 cases). Thus an additional decad e of data resulted in a subst antial reversal of results reported in S08 for RSS comp arisons. Ocean: S08 reported that t he model ensemble trend for the surface-T2 LT difference series over ocean was -0.085 deg C/de cade. Observed trends in surface-T2 LT difference series involving RSS ran ge from -0.058 to -0.083 deg C/decade, with the differences from the model ensemble m ean not being significant (see Table 3) in l ine with reported S08 reports. The correspond ing trends involving UAH data w ere positive and all differences with models were statistically significant (again in li ne with S08). (Corresponding results for T2 data w ere similar to T2 LT results.) Land: S08 did no t report a model ensemble mean for the surface- T2LT difference series over land; the implied va lue from the reported land-and-ocean and oce an values was - 0.012 deg C/decade. Obs erved trends in surface-T2 LT difference series inv olving RSS range from +0.015 to +0.02 5 deg C/decade, having an opposite si gn to the model ensemble, with the differ ences being significant at a 5% level unde r a two-sided t-test for comparisons to CRUTEM3 and NOAA-GHCN surface data (see T able 4). Differences with UAH satellite data were even greater. Discussion and Con clusions S08 stated that “when the R SS-derived tropospheric temperatu re trend is compared with four different observed est imates of surface temperature change, the surfa ce warming is invariably amplified in the tropical tropos phere, consistent with model results.” Using updated data, this statem ent is no longer true for the combined land- and-ocean comparison (using HADC RUT3v). While the statement continues to hold with updated data for the three SST dat a sets used in S08, it does not hold for the three major land data sets. S08 stated that “even” fo r UAH data, observed trop ical lapse rate trends are not significantly different fro m models and that the observed UAH tropical trop ospheric trends were not significa ntly different from the ensemble m ean. Again, this does not hol d true using updated data. In this c ase, by re-doing the tests herein with data endi ng in 2007 we can show that the chang e from non-significance to significance do es not result from end-point effects associated with 2008 temperatures. Instead it result s almost entirely from a nearl y 50% increase in the number of observations, thereb y materially increasing the number of d egrees of freedom in the stati stical calculation even with autocorrelation. Overall, the conclusion of S08 that “there is no longer a serious di screpancy between modeled and observed tr ends in tropical lapse rates” must be reconsidered in light of up- to-date data. The “potenti al inconsistenc y ” between models and observatio ns in the tropical region, as reported b y Karl et al (2006), remains an issue. The results reported here only pertain to an AR1 autocorrelation mo del as used in S08. Consideration of the impact of other autocorrelation specifications or oth er sources of error (e.g. Thorne et al. 2007 ) may well be of interest (both for thi s and other studies) and might contribute to a res olution of the “potential inconsist ency” that concerned Karl et al (2006). However even if a recon ciliation proves possib le on alternative grounds, that does not affect the specif ic results presented here, n amely that key S08 results change based on use of updated data. REFERENCES: Brohan, P., J. J . Kennedy, I. Harris, S. F. B. Tett, and P. D. Jones. 2006. Uncertaint y estimates in regional and global obse rved temperature changes: a ne w dataset from 1850. J. Geophys. Res 111, no. D12. http://hadobs.m etoffice.com/crutem3/HadCRUT3_accepted.pdf. Christy, J. R., W . B. Norris, R. W. Spencer, and J . J. Hnilo. 2007. T ropospheric temperature change since 197 9 from tropical radiosonde and satellite measurements. Journal of Geophysi cal Research 112, no. D6. Climate Change Science P rogram (U.S.). 2009. Global C limate Change Impacts in the United States . http:/ /www.globalchange.gov/publications/reports/scientifi c- assessments/us-impacts. Hansen, J., R. Ru edy, M. Sato, M. Imhoff, W. Lawrence, D. Easte rling, T. Peterson, and T. Karl. 2001. A closer l ook at United States and global surface temperature change. Journal of Geop hysical Research-Atmosph eres 106, no. D20. Karl, T. R., Susan J . Hassol, Christopher D. Miller, and W illieam L. Murray. 2006. Temperature Trends in the L ower Atmosphere: Steps for Und erstanding and Reconciling Differences. Synthesis and Assessment P roduct. Climate Change Science Program and the S ubcommittee on Global Ch ange Research. http://www.climatescience.gov/ Library/sap/sap1-1/finalreport/sap1-1-final -all.pdf. Mears, C. A., M. C. Schab el, and F. J. Wentz. 2003. A reanalysis of the MSU channel 2 tropospheric temperature record. Journal of Climate 16, no. 22: 3 650-3664. Mears, C. A., and F. J. Wentz . 2005. The effect of diurnal correction o n satellite-derived lower tropospheric temperature. S cience 309: 1548-1551. Peterson, T. C., and R. S. Vose. 1997. An Overview of the Global Hi storical Clim atology Network Temperature Database. Bulleti n of the American Met eorological Society 78, no. 12: 2837-2849. Quenouille, M. H. 1952. Associated Measurement s . Butterworths Scientific Publ ications. Rayner, N. A., D. E. Park er, E. B. Horton, C. K. Folland, L. V. Alexander, D. P. Rowell , E. C. Kent, and A. Kapla n. 2003. Global anal yses of sea surface temperature, sea ice, and ni ght m arine air temperature since the late ni neteenth centur y. J. Geophys. Res 108, no. D14: 4407. Santer, B. D., P. W. Thor ne, L. Haimberger, K. E. Taylor, T. M. L. Wigley, J. R . Lanzante, S. Solomon, M. Free, P . J. Gleckler, and P. D. J ones. 2008. Consistenc y of Modelled and Observed Temp erature Trends in the Tropical Troposph ere. Int. J. Climatol . Smith, T. M., and R. W. Reynolds. 2005. A Global Merged Land–Air–Sea Surface Temperature Reconstruct ion Based on Historical Observations (1880–1997). Journal of Climat e 18, no. 12: 2021-2036. Smith, T. M., R. W. Reynolds, T. C. Peterson, and J. L awrimore. 2008. Improvements to NOAA’s Historical Merged Land–Ocean Surface Temperature Anal ysis (1880– 2006). Journal of Cli mate 21, no. 10: 2283-2296. Spencer, R. W., and J . R. Christy. 1990. Precise monitoring of global tempe rature trends from satellites. Science 2 47, no. 4950: 1558-1562. Thorne, P. W., D. E. Par ker, B. D. Santer, M. P. McCarth y, D. M. H. Sexton, M. J . W ebb, J. M. Murph y , M. Collins, H. A. Ti tchner, and G. S. J ones. 2007. Tropical vertical temperature trends: A real discrepancy? Geophysical Research Lett ers 34, no. 16: L16702. Satellite Ensemble Trend Observed Trend d 1 * up to 1999:12 (from S08) d 1 * up to 2009:6 (t-test percentile) Significance under two- sided t-test (one-sided) RSS T2LT 0.215 0.140 0.37 1.08 (85.5) - (-) UAH T2LT 0.215 0.051 1.11 2.42 (98.8) ** (**) RSS T2 0.199 0.104 0.44 1.41 (91.4) - (*) UAH T2 0.199 0.021 1.19 2.72 (99.4) ** (***) TABLE 1: T2 LT and T2 Trend Comparisons. Trends in de g C/decade. One, two, and three asterisks indicate m odel-versus-observed trend d ifferences that are significant at the 10, 5, and 1% levels resp ectively. Surface Satellite Ensemble Lapse Trend Observed Lapse Trend d 1 * up to 1999:12 (from S08) d 1 * up to 2009:6 (t-test percentile) Significance under two- sided t-test (one-sided HadCRUT3v RSS T2LT -0.069 -0.028 -0.67 -2.34 ( 1.1) ** (**) HadCRUT3v UAH T2LT -0.069 +0.061 -3.50 -7.00 (0.0) *** (***) HadCRUT3v RSS T2 -0.053 +0.008 n.r. -3.30 (0.0) *** (***) HadCRUT3v UAH T2 -0.053 +0.091 n.r. -7.73 (0.0) *** (***) TABLE 2: La pse Rates (Land-and-Ocean). Trends in deg C/decade. One, two, and three asterisks indicate m odel-versus-observed trend d ifferences that are significant at the 10, 5, and 1% levels resp ectively; (two-tailed tests). All UAH lapse trends are strongl y significant. nr: not report ed . S08 showed absolute value of d 1 *. Surface Satellite Surface-minus- Satellite d 1 * up to 1999:12 from S08 d 1 * up to 2009:6 (t-test percentile) Significance under two- sided t-test Ensemble Observed (one-sided) HadISST1 RSS T2LT -0.085 -0.083 -0.75 -0.07 (47.1) - (-) ERSST-v2 RSS T2LT -0.085 -0.061 -0.48 -1.06 (14.8) - (-) ERSST-v3 RSS T2LT -0.085 -0.058 -0.12 -1.22 (11.5) - (-) HadISST1 UAH T2LT -0.085 +0.01 8 - 3.52 -4.29 (0.0) *** (***) ERSST-v2 UAH T2LT -0.085 +0.03 9 - 3.04 -5.04 (0.0) *** (***) ERSST-v3 UAH T2LT -0.085 +0.04 4 - 2.68 -5.25 (0.0) *** (***) TABLE 3: La pse Rates (Ocean). Trends in deg C /decade. One, two, and three asterisks indicate model-versus-ob served trend differences that are significant at the 10, 5, and 1% levels respectively; (two- tailed tests). S08 showed absolute value of d 1 *. Surface Satellite S urface-minus-T2LT d 1 * up to 1999:12 from S08 d 1 * up to 2009:6 (t-test percentile) Significance under two- sided t-test Ensemble Observed (one-sided) CRUTEM3 RSS T2LT -0.012 +0.039 Nr -2.72 (0.5) ***(***) GISS-250 RSS T2LT -0.012 +0.015 Nr -1.46 (7.5) - (*) NOAA RSS T2LT -0.012 +0.025. Nr -2.08 (2.1) ** (**) CRUTEM3 UAH T2LT -0.012 +0.10 4 Nr -6.66 (0.0) *** (***) GISS-250 UAH T2LT -0.012 +0.07 9 Nr -4.82 (0.0) *** (***) NOAA UAH T2LT -0.012 +0.089 Nr -6.21 (0.0) *** (***) TABLE 4: La pse Rates (Land ) One, two, and three asterisks i ndicate model-versus- observed trend difference s that are significant at the 10, 5, and 1% lev els respectivel y ; (two-tailed tests).