SEARCH

SEARCH BY CITATION

Keywords:

  • CHUAN;
  • Comprehensive Historical Upper-Air Network;
  • data intercomparisno;
  • reanalysis;
  • West Africa;
  • monsoon;
  • Twentieth Century Reanalysis;
  • NCEP/NCAR Reanalysis

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and methods
  5. 3. Results
  6. 4. Discussion and conclusions
  7. Acknowledgements
  8. References
  9. Supporting Information

The strong interannual-to-decadal variability of the West African Monsoon is subject to an active field of climate research that tries to disentangle its influencing factors and explore its predictability. Reliable observation-based data over a preferably long period are arguably the most important basis for such efforts. Here, we try to explore the quality of several data products available for the earlier period of upper-air observations (1940–1957): the Comprehensive Historical Upper-Air Network (CHUAN), the NCEP/NCAR Reanalysis (NNR), and the Twentieth Century Reanalysis (20CR). To do so, we compare wind soundings from 37 pilot balloon stations contained in CHUAN (10°S–30°N, 20°W–20°E) with the reanalyses.

The comparison with the NNR reveals seasonally and diurnally varying significant differences relative to the observations over West and Central Africa. The differences reach absolute values of several metres per second, and their spatially coherent structure strongly points to a deficiency of the reanalysis rather than observational errors. The difference fields indicate an overestimation of the strength and thickness of the low-level monsoon in all seasons and an underestimation of the Harmattan winds over the Sahel in winter. At higher levels, they point to an overestimation of the mid-tropospheric monsoon return flow and African easterly jet.

For the 20CR, the fields reveal again significant differences up to several metres per second on all levels and in all seasons. However, the direction relative to the observed monsoon flow and Harmattan/trades is opposite at the low levels. Additionally, the differences tend to be smaller and more confined to the coastal region.

Further analysis demonstrates that the observed interannual variability is only insufficiently modelled in both reanalyses. Together with the diurnal cycle of the differences, this precludes a simple correction of the reanalyses and demonstrates that, depending on the purpose of a study, one should be extremely cautious when using reanalysis products. Copyright © 2011 Royal Meteorological Society


1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and methods
  5. 3. Results
  6. 4. Discussion and conclusions
  7. Acknowledgements
  8. References
  9. Supporting Information

The West African Monsoon system is characterised by a very marked interannual as well as decadal variability. This is of fundamental importance for the predominantly agrarian societies of the Sahel region across the African continent, which depend on the strongly variable precipitation. Furthermore, the West African Monsoon is subject to a very active field of climate research that tries to quantify the multitude of factors influencing it, e.g. sea surface temperature (SSTs) in different oceanic basins and in the Mediterranean, land–surface interactions and aerosols. Also the potential predictability of Sahel rainfall is a topic under debate. Reliable observational or observation-based atmospheric data (e.g. reanalyses) over a period as long as possible are arguably the most important basis for developing the necessary theoretical understanding. In the present study, we try to explore the quality of such data products in the earlier period of upper-air observations in the region of interest.

Atmospheric reanalyses like the NCEP/NCAR 50-Year Reanalysis (NNR) (Kistler et al., 2001) are readily available to the scientific community, and therefore often used as a substitute for observational data, especially in studies on the continental to global scale. Their advantage compared to single observational time series clearly lies in their spatial and temporal completeness. This is one of the main reasons why they are in fact the most widely used datasets in atmospheric science.

However, when working with these data, one should always keep in mind that the reanalysis might not adequately represent the atmospheric state for certain regions and time periods. Not only are derived variables of the reanalyses like heat and radiative fluxes, clouds, moisture fluxes, precipitation or snow accumulation affected over large regions of the globe, as has been shown for the Tropics, the North Pacific, North America, the Caribbean, the Mediterranean or the polar regions (Higgins etal., 1996; Bony et al., 1997; Gutowski et al., 1997; Trenberth and Guillemot, 1998; Lenters et al., 2000; Poccard et al., 2000; Ladd and Bond, 2002; Betts etal., 2003a, 2003b, 2005; Kinter et al., 2004; Zhao et al., 2006; Bromwich et al., 2007; Karam and Bras, 2008; Jury, 2009; Pettenuzzo et al., 2010), but even primary (i.e. assimilated) variables such as temperature, pressure, humidity and wind. For example, besides the well-known problem of artificial jumps or trends introduced by changes in the observing system (Poccard etal., 2000; Trenberth et al., 2001; Simmons et al., 2004), the assimilation of uncorrected radiosonde data has led to a widespread warm bias in the Northern Hemisphere in the 1950s in the NNR (Grant et al., 2009), and significant temperature, pressure and humidity biases have been detected for several reanalysis products in the Antarctic (Hines et al., 2000; Nedoluha et al., 2007; Parrondo et al., 2007) as well as the Arctic (Grant et al., 2008; Tjernström and Graversen, 2009; Lüpkes et al., 2010), and in the Southern Hemisphere (Simmons et al., 2004). In a tropospheric assessment of different reanalyses Bromwich et al. (2007) actually concluded that ERA-40 (Uppala et al., 2005), JRA-25 (Onogi et al., 2005) and the NNR are unreliable during winter in the Antarctic region before the satellite era. Even the most recent generation of reanalyses is not void of such errors: e.g. Lüpkes et al. (2010) found that in the ERA-Interim reanalysis (Dee and Uppala, 2009) the melting point of snow is the most frequent near-surface temperature, while the most frequently observed value is the sea-water freezing temperature. Furthermore, their study suggests that ERA-Interim overestimates the relative humidity and temperature in the atmospheric boundary layer and the base height of the capping inversion in the Arctic, with warm biases of near-surface temperatures up to 2 K. Also, large-scale climatological features are affected by deficiencies of the reanalyses products: Van Noije et al. (2004) and Oikonomou and O'Neill (2006) noted a too intense Brewer–Dobson circulation in the ERA-40 reanalysis, and Kinter et al. (2004) detected artificial decadal regime shifts in the large-scale tropical upper tropospheric circulation and precipitation between the periods 1950–70 and 1980–2000. Further problems can be expected especially over regions of the globe that are not (or have not been) covered very well by observations, such as parts of the Tropics, the oceans and generally the Southern Hemisphere, but also for regions for which not all available atmospheric observations did enter the assimilation process of the reanalyses.

The West African Monsoon region is one example for which the latter applies. Poccard et al. (2000) identified several systematic abrupt shifts in the NNR rainfall over tropical Africa compared to observations from CRC (Centre de Recherche de Climatologie, Dijon, France) and CRU (Climatic Research Unit, Norwich, UK). They found the most remarkable shift in the year 1967, affecting nearly the whole of tropical Africa. Before this year, their analysis showed almost no correlation with observations, which the study ascribed to the scarcity of land surface and upper-air measurements assimilated into the dataset. In a follow-up study, Camberlin et al. (2001) detected abrupt shifts also in NNR geopotential height, temperature, specific humidity and zonal wind over large parts of tropical Africa in the period 1961–1993, again especially around 1967/1968. For these variables and in the lower troposphere, the shifts were found to be particularly strong over West Africa. They concluded in discouraging from the use of the NNR over tropical Africa before 1968. However, no attempt has hitherto been made to quantify these biases over tropical Africa in more detail, particularly in the early period before 1958.

During the period 1948–1957 the number of upper-air observations assimilated into the NNR over the Sahel–Guinean region is even much smaller than after the 1957 International Geophysical Year (IGY), especially before 1952/1953 (see, for example, the NNR upper-air observation count at http://nomad3.ncep.noaa.gov/cgi-bin/pdisp_r1_obs.sh). For that period, no pilot balloon wind observations have been assimilated, and only the much sparser radiosonde data have been used. Therefore, in order to detect differences of the NNR relative to observations in this region, we compare the wind fields extracted from the reanalysis with observed winds at 37 pilot balloon stations located in the domain (10°S–30°N, 20°W–20°E).

Section 2 describes the data and methods used. Section 3 presents the results of the comparison. In a first part, the wind stations are classified into different climatological regimes corresponding to respective bias regimes. Typical vertical bias profiles are shown and we demonstrate exemplarily that temporal variability of the bias profiles exists down to the diurnal scale. A more physical view of the wind bias is suggested by examining plots of the observed wind and the bias fields on different pressure levels in the lower to middle troposphere. We also present and discuss the same sort of bias fields for the Twentieth Century Reanalysis (Compo et al., 2011), a product based solely on the assimilation of surface pressure using observed SSTs as boundary condition. Finally, the interannual variability in the observations and the reanalyses is discussed. Section 4 discusses the results, draws conclusions and provides an outlook.

2. Data and methods

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and methods
  5. 3. Results
  6. 4. Discussion and conclusions
  7. Acknowledgements
  8. References
  9. Supporting Information

For this study, we have used three datasets. The first is the NCEP/NCAR 50-Year Reanalysis (NNR) which has been used in a multitude of atmospheric studies and is described in detail in Kalnay et al. (1996) and Kistler et al. (2001). The documentation is further supplemented by the additional online material given on the reanalysis websites of NOAA ESRL (http://www.esrl.noaa.gov/psd/data/reanalysis/reanalysis.shtml), NOAA CPC (http://www.cpc.ncep.noaa.gov/products/wesley/reanalysis.html) and NCAR (http://dss.ucar.edu/pub/reanalysis/). Here, we have studied horizontal wind fields from this dataset. The U and V components of the wind field on the pressure levels 925, 850, 700, 600 and 500 hPa were extracted at the exact positions of the 37 pilot balloon stations examined here (see further details below). The period considered is 1948–1957, since the observational data were available only until 1957.

The second dataset is the Twentieth Century Reanalysis (20CR) (Compo et al., 2011), a global reanalysis back to 1850 which has been produced by assimilating only sea-level pressure (SLP) or surface pressure observations and using observed SSTs and sea ice concentrations as model boundary conditions. The feasibility of such a reanalysis has been demonstrated in Compo etal. (2006), and the data can be obtained from the NOAA ESRL website (http://www.esrl.noaa.gov/psd/data/20thC_Rean/). As for the NNR, we have extracted the U and V winds at the station positions for the period 1948–1957, but also for the earlier period with available observations (1940–1947), with all pressure levels being the same except the 925 hPa level, which was substituted by the 900 hPa level available in the 20CR.

The observational data were taken from the Comprehensive Historical Upper-Air Network (CHUAN) (Stickler et al., 2010). CHUAN is a recently developed, consistent compilation of global historical upper-air data derived from different, heterogeneous datasets already available as well as from newly digitised data. It comprises 3934 station records worldwide or about 16.4 million profiles (12.6 million before 1958 and 5.3 million, primarily from pilot balloons, before 1948). For the present work, we made use of the wind data from all 37 pilot balloon stations available in that dataset for the studied region. A list of the stations with their exact geographical position, the period of available data and the total number of ascents during the considered periods are given in Table I. Since the observational wind data are given on geometrical altitude levels, we have chosen the following levels corresponding roughly to the pressure levels selected in the reanalyses: 750 (NNR)/1000 (20CR), 1500, 3000, 4250 and 5500 m above sea level (asl). The winds on the levels 750 and 4250 m asl have been linearly interpolated from the levels 500 and 1000, and 4000 and 4500 m asl, respectively. An assessment of the estimated wind errors resulting from the use of fixed geometric altitude levels as reference for fixed pressure levels (see text in the online supporting information) shows that, except in the northernmost part of the region (termed NW and W Chad in the rest of the paper) in summer, these errors can be expected to be <2 m s−1 at 600 hPa, <1.7 m s−1 at 700 hPa and <0.6 m s−1 up to 850 hPa at most. These values are significant, but clearly smaller than the differences found relative to the reanalyses and described in the present study.

Table I. List of the 37 CHUAN pilot balloon stations in the domain (10°S–30°N, 20°W–20°E) used in this study. Besides the geographical position, the periods with data coverage and the total number of observations are given (the latter with respect to the timeframes 1950–1957, 1948–1957 and 1940–1957). Stations used for the calculation of the 1940–1957 climatology are marked with an asterisk.
CHUAN #StationLatitude (°)Longitude (°)Time coverage# of obs. 1950–1957# of obs. 1948–1957# of obs 1940–1957
641Kandi11.132.933/1940–12/1942, 6/1948–7/1950, 3/1952–12/195754566 2247 028
642Tchaourou8.872.605/1942–12/1942, 1/1947–6/1950, 1/1952–12/195753436 0446 370
643*Cotonou6.352.381/1940–7/1950, 1/1951, 7/1951–12/195774859 26312 818
644*Ouagadougou12.35−1.521/1940–12/1957912810 36013 145
645*Bobo-Dioulasso11.67−4.304/1940–12/1957881210 12812 919
646Douala4.179.721/1948–12/194901 1561 156
647Bangui4.3818.571/1950–12/195792859 2859 285
648Ndjamena12.1315.031/1950–12/195793019 3019 301
649Pointe-Noire−4.8211.903/1947–12/1948, 2/1949, 6/1949–12/1949, 9/1950–12/195775628 3188 657
650Dolisie−4.1812.671/1949–12/1950, 8/1954–12/195729443 2253 225
651Bouaké7.68−5.031/1940–7/1941, 3/1944–4/1944, 1/1945 3/1945, 5/1945–10/1947, 1/1948–1/1949, 3/1949–11/1950, 5/1951–12/195770347 6469 064
652*Abidjan5.25−3.931–2/1940, 1/1941–12/195771268 83612 600
653Tabou4.42−7.371/1944, 3/1944–12/1944, 5/1946–8/1947, 1/1948–11/1950, 2/1952–2/1954, 10/1954–12/195747235 4096 365
654Port-Gentil−0.728.782/1950–12/195768936 8936 893
655Labé11.31−12.3010/1953–12/195735153 5153 515
656Mamou10.37−12.081/1949–9/195323002 7572 757
657Kankan10.38−9.3010/1944–6/1946, 9/1946–12/1957913710 44311 542
658*Conakry9.57−13.621/1940–3/1941, 5/1941, 7/1941, 9–10/1941, 12/1941–11/1944, 1/1945–12/195773989 32712 934
659Fort Gouraud22.68−12.7010–12/1941, 3–6/1942, 8–9/1942, 1/1943–12/1944, 2–10/1945, 9/1946–7/1949, 3–4/1950, 6–7/1950, 12/1951–12/195756536 3197 989
660*Nouadhibou20.93−17.051–2/1940, 5/1940–10/1950, 3/1951–12/1957909310 75514 708
661*Atar20.52−13.071–3/1940, 5/1940–12/1942, 3–5/1943, 7/1943–12/1957947410 94215 765
662Nouakchott18.12−15.938–10/1941, 1/1942–1/1943, 3–12/1943, 3–5/1944, 10/1944–9/1948, 11/1948–4/1949, 11/1949–11/1950, 3/1951–12/195771577 94010 454
663Néma16.62−7.021–2/1942, 8–12/1942, 1–4/1944, 7–9/1944, 11–12/1944, 5/1945–2/1946, 8/1946, 11/1946–12/195781688 9249 943
664Aioun El Atrouss16.73−9.6312/1952–12/195742364 2364 236
665Bilma18.6812.921/1950–12/195766876 6876 687
666Agadès16.987.982–3/1947, 5/1947, 7–12/1947, 2/1949–8/1950, 1/1952–5/1954, 11/1954–12/195760976 6816 930
667*Niamey13.502.131/1940–8/1950, 3/1951–12/195780449 87714 437
668Birni Nkonni13.805.258/1955–12/195721482 1482 148
669Zinder13.809.001/1940–2/1941, 5/1941–4/1942, 8/1942, 1/1947–12/1957887310 64212 232
670*Saint-Louis16.02−16.501/1940–7/1950, 3/1951–12/195766568 72513 478
671Dakar14.73−17.502/1940, 4/1940–10/1940, 11/1947–12/195771379 2229 506
672*Tambacounda13.77−13.681/1940–2/1947, 1/1948–11/1950, 3/1951–12/195774988 60511 266
673*Ziguinchor12.58−16.271/1940–6/1948, 11/1948–11/1950, 5/1951–12/195783329 39113 378
674Thiès14.80−17.075/1945–1/1948, 3/1952, 5/1952–6/1954, 9–10/195521382 1933 885
675Dakar14.67−17.4311/1940–1/1941, 3/1941, 5/1941, 7–8/1941, 10/1941–9/1946, 1–11/1947003 718
676Sokodé8.981.1312/1957565656
677Lomé6.171.253/1945–12//1957937211 55313 660

In the following, observations are considered ‘simultaneous’ with the respective reanalysis data whenever the time difference between the observation and the closest available standard reanalysis time (0000, 0600, 1200, 1800 UTC) is at most 2 h. For the whole analysis of differences between observations and reanalyses (including the seasonal means plotted in Figures 7–11) we selected just the times (day and hour) for which both observations and reanalyses are available to calculate the mean differences; i.e. the differences have not been calculated as differences of the monthly or seasonal means of all observations and the whole reanalyses (all four times daily values) for that month/season, but as the mean of all single differences of simultaneous observation–reanalysis events. The order of the maximum wind error resulting from this definition of simultaneity can be estimated from the study of Parker etal. (2005) (text in the online supporting information). This gives an error range for single ascents of ±1.5 m s−1, again smaller than the majority of the differences reported in the present study. Averaging over different times of day, as has been done for this study, further reduces this error.

In order to cluster the 37 stations into different climatological regions, characterised by similar vertical difference profiles, we calculated seasonal mean vertical difference profiles of the U and V winds for all seasons (DJF, MAM, JJA and SON) and stations. The uncertainty of the difference at each pressure level has been calculated as equation image with X = {U,V}, the total number of observation–reanalysis pairs for that pressure level, and the NCEP, 20CR and observed values taken at simultaneous times, respectively. We also plotted time series of the U and V difference for each station and pressure level. For the detection of a potentially existing diurnal cycle of the difference to the observed wind, an exemplary analysis was undertaken for the calendar month of July, for which the data were further stratified into four groups with respect to the time of the day of the observation (0000, 0600, 1200, 1800 UTC ±2 h).

Plots of the observed seasonal mean winds (climatology for the period 1940–1957 on the 925, 850 and 700 hPa pressure levels) and difference fields of the reanalyses relative to the observed winds (on the 925 (NNR)/900 (20CR), 850, 700 and 600 hPa pressure levels) have been produced with the purpose of finding spatially coherent structures that might point at a reanalysis deficiency as a cause of the difference rather than observational errors. For the NNR, the difference fields are calculated for the overlapping period 1948–1957. In the case of the 20CR, the difference fields have been produced for the overlapping period with the NNR (1948–1957) as well as for the earlier period 1940–1947.

To ensure that the results are representative for the season and free from an observational bias that might be introduced by a preference for days with good visibility due to the visual tracking technique used, a minimum of 20 observations on average per month of observation was required. Additionally, we made sure that no large over- or underweighting of single calendar months occurs (<20%, i.e. no calendar month is weighted by less than 80% or more than 120% compared to other calendar months) and that observational data are available for a minimum of 50% of the respective period considered when calculating seasonal means. This led to a significantly reduced number of available stations fulfilling these criteria on the 600 hPa level (7–17 (NNR) and 8–15 (20CR) of 37, depending on the season, both for the period 1948–1957). For the 20CR and the period 1940–1947, the numbers are also reduced on the lower levels due to the much smaller number of measurements in the 1940s compared to the 1950s. For the two-dimensional wind fields, the statistical significance of the differences found between the reanalyses and the observations was tested station-wise using a one-sample t-test (paired difference test; see, for example, von Storch and Zwiers, 2001, p. 113f.). The t-test is relatively robust with respect to changes in the degrees of freedom (autocorrelation) for large sample sizes (>100).

For the climatology, additional to the above-mentioned criteria, data from the longer period 1940–1957 are used, and only 11 stations having data for all years are considered (see Table I). The period 1940–1957 was chosen for the climatology because it contains a roughly equal time-span influenced by El Niño and La Niña events and should therefore not be largely biased with respect to El Niño–Southern Oscillation (ENSO), opposite to the shorter period 1948–1957. The climatology, which has been calculated as a simple mean of single ascents, is biased towards the more data-rich 1950s (cf. Table I). However, the latter period also offers a better coverage of the diurnal cycle (mostly four daily observations from 1950–1952 compared to two or three before) and is again unbiased with respect to ENSO (mean standardised Niño-3 of −0.18) and to the standardised tropical Atlantic hemispheric temperature difference between the regions (0–20°S 30°W–5°E) and (10–20°N, 50–20°W), which have both been found to be important for the West African climate. In the present study, the climatology is only used to characterise the mean seasonal horizontal circulation. The results regarding the differences between the reanalyses and the observations are therefore not dependent on the climatology.

Finally, the interannual wind variability in the reanalyses is compared to that in the observations. For this purpose, time series of the horizontal wind components at the 925/900 and 700 hPa pressure levels are compared with each other and correlation coefficients and their significance (two-sided test based on Fisher's z-transform, verified by a Spearman's rank correlation test) are calculated.

3. Results

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and methods
  5. 3. Results
  6. 4. Discussion and conclusions
  7. Acknowledgements
  8. References
  9. Supporting Information

An analysis of the vertical profiles of the NNR U and V winds relative to the observations and their seasonality indicated that seven subregions can be defined which show evidence of a relatively uniform behaviour (Figure 1): the Northwest, the Southwest, central sub-Saharan West Africa (SSWA), central and eastern Niger, western Chad, the western Central African Republic, and the southern coastal region east of the Gulf of Guinea (SCR). The stations inside these subregions are also characterised by similar climatological winds (cf. Figure 6). Figure 1 demonstrates that there is quite a good data coverage for all of the subregions except for the eastern part of SSWA, over the whole territory of Nigeria, for which no CHUAN data are available.

thumbnail image

Figure 1. Map showing the seven subregions identified over West and Central Africa by manual inspection of the vertical profiles of U and V wind differences of the NCEP/NCAR 50-Year Reanalysis (NNR; Kistler et al., 2001) relative to simultaneous CHUAN (Stickler et al., 2010) pilot balloon wind observations (time difference ≤2 h). These subregions also exhibit similar vertical profiles and seasonal behaviour of the observed winds themselves. All 37 CHUAN pilot balloon stations are named and marked with filled triangles. Numbers below the station names are CHUAN station numbers. This figure is available in colour online at wileyonlinelibrary.com/journal/qj

Download figure to PowerPoint

The U and V wind difference generally exhibits a more or less defined seasonal cycle in all subregions. As an example, Figure 2 displays the chronological series of the U wind difference relative to observations at the 925 hPa level (∼750 m asl) at Conakry, Guinea, for the years 1948–1957. This suggests that a temporally more detailed analysis should be undertaken, at least on a seasonal level.

thumbnail image

Figure 2. Time series of the difference Ureanalysis − Uobs (NNR U wind difference relative to CHUAN observations) at the 925 hPa pressure level in Conakry, Guinea, for the time period 1948–1957. The variations between positive and negative values represent the annual cycles. This figure is available in colour online at wileyonlinelibrary.com/journal/qj

Download figure to PowerPoint

3.1. NNR seasonal mean vertical difference profiles

Figures 3 and 4 show typical NNR seasonal mean vertical difference profiles for the U and V wind in SSWA compared to the climatological winds (the complete picture for all regions can be found in Figures 1 and 2 of the online supporting information). These profiles already suggest that the differences are, in many cases, statistically significant, since the latter are larger than their 1σ uncertainty, calculated as described in section 2 (for a comprehensive assessment of the statistical significance of the seasonal mean difference fields see sections 3.3 and 3.4). Differences exist throughout all seasons over the broader West and Central African Monsoon region. In all subregions, absolute differences in either the U or the V wind of >3 m s−1 are found, similar to or even larger than the climatological winds. The subregions where the largest absolute differences in the U or V wind (>5 m s−1) can be distinguished are SSWA, western Chad, the western Central African Republic, and the SCR. Furthermore, for all subregions except the western Central African Republic, the boreal summer season (JJA) turns out to be connected to a strong difference, at least for one wind component.

thumbnail image

Figure 3. Examples of seasonal mean profiles of the U wind difference (NNR - CHUAN) typical for sub-Saharan West Africa. Only the seasons with the largest magnitudes of the difference are shown. Error bars give the 1σ uncertainty of the climatology and of the difference as defined in section 2.

Download figure to PowerPoint

thumbnail image

Figure 4. The same as Figure 3 but for the V wind difference in JJA.

Download figure to PowerPoint

3.2. Diurnal cycle of the NNR difference

The calculated diurnal cycle of the NNR monthly mean vertical wind difference profiles is shown exemplarily for the V wind in the calendar month July (one of the calendar months with the strongest deviation on average for the whole region) at Ndjamena, Chad, in Figure 5. We find highly significant diurnal cycles at the station level and for single pressure levels (U or V wind), such as the one identified for Ndjamena, for a total of 30 out of 37 records (modified t-test on significance of differences of the means with unequal variances, Fisher–Behrens problem; see, for example, von Storch and Zwiers, 2001, p. 113, Si = 99.9%). All records for which the hypothesis of a highly significant diurnal cycle had to be rejected either have a low number of observations (<100 ascents) or are located in the Northwest (Néma, Nouakchott, Saint-Louis).

thumbnail image

Figure 5. Diurnal cycle of the monthly mean V wind difference of the NNR relative to CHUAN observations for the calendar month July and the period 1948–57 at Ndjamena, Chad. The monthly mean difference is shown for the times 0000, 0600, 1200 and 1800 UTC corresponding to 0100, 0700, 1300 and 1900 h local time. The horizontal bars denote 1σ error bars for the difference as defined in section 2.

Download figure to PowerPoint

A closer analysis for SSWA demonstrates that the minimum of the diurnal cycle in the SSWA U wind deviation preferentially occurs around 1200 UTC (55% of all detected cases), the maximum at 1800 UTC (80% of all cases). On the other hand, the minimum of the diurnal cycle of the SSWA V wind deviation occurs almost only at 0000 UTC (92% of all cases), while the maximum is mostly observed at 1200 UTC (50% of all cases). Here, the term ‘minimum’ is used for a minimum of the vertical average of the absolute values of the differences and vice versa for the ‘maximum’. Generally, for this region, we find a tendency for the diurnal cycle to manifest itself as a back and forth translation of the difference profile along the abscissa. This translation can therefore lead to an increase as well as a decrease of the vertically averaged difference for the ‘minimum’ compared to the diurnal mean.

3.3. NNR and 20CR seasonal mean difference fields on constant pressure levels (1948–1957)

Figures 6, 7 and 8 illustrate the climatological (period 1940–1957, from the observations) and the NNR/20CR difference fields relative to the observations (period 1948–1957) calculated for the 925/900, 850, 700 and 600 hPa pressure levels, approximately corresponding to altitudes of 750/1000, 1500, 3000 and 4250 m asl. We do not show climatological fields for the 600 hPa level due to the too small number of available observations for the 11 stations with available data for the whole climatology period. The columns in the figures correspond to the winter (DJF), spring (MAM), summer (JJA) and autumn (SON) seasons, the rows correspond to the three/four pressure levels.

thumbnail image

Figure 6. Climatology of the observed seasonal wind field (CHUAN pilot balloon wind observations) for the years 1940–1957. The rows represent the three pressure levels 925 hPa, 850 hPa and 700 hPa (from top to bottom). The four columns show the fields for the seasons DJF, MAM, JJA and SON (left to right). A minimum of 20 observations on average per month of observation is required at each station. Only the 11 stations which have data in all years (1940–1957) have been taken into account for the calculation of the climatology. For the single seasons, it has been assured that at least 17 of the 18 years have data at the respective station. No single calendar month of a season is significantly over- or underweighted relative to the rest of the calendar months (<20%). This figure is available in colour online at wileyonlinelibrary.com/journal/qj

Download figure to PowerPoint

thumbnail image

Figure 7. Seasonal mean difference fields of the NCEP/NCAR Reanalysis (NNR) wind relative to simultaneous CHUAN pilot balloon wind observations for the period 1948–1957 (blue). The rows represent the four pressure levels 925 hPa, 850 hPa, 700 hPa and 600 hPa (from top to bottom). The four columns show the fields for the seasons DJF, MAM, JJA and SON (left to right). Observations are considered ‘simultaneous’ with the respective reanalysis data whenever the time difference between the observation and the closest available standard reanalysis time (0000, 0600, 1200, 1800 UTC) is at most 2 h. As for the climatology, a minimum of 20 observations on average per month of observation is required, and no large over- or underweighting of single calendar months is allowed (<20%). Furthermore, a minimum of 5 years of observations is necessary for each season and station. For the 600 hPa level, due to the relatively low number of stations fulfilling these criteria, the rest of the difference vectors is also plotted, but with red arrows. For the 925, 850 and 700 hPa levels, the differences >0.5 m s−1 are significant at Si = 99.9% (DJF, SON), Si = 99.5% (JJA) and Si = 97.5% (MAM), and at Si > 95% for the 600 hPa level (paired difference test; see section 2).

Download figure to PowerPoint

thumbnail image

Figure 8. As Figure 7 but for the Twentieth Century Reanalysis (20CR), and showing the 900 hPa instead of the 925 hPa pressure level in the first row. For the 900 and 850 hPa levels, the differences >0.5 m s−1 are significant at Si = 99.9%, for the 700 hPa level at Si > 99.5%, and at Si > 90% for the 600 hPa level (paired difference test; see section 2).

Download figure to PowerPoint

3.3.1. 925/900 hPa pressure levels

In the case of the NNR, we find southerly (mainly at the coast in summer) to southwesterly difference vectors over the whole SSWA region and during all seasons (Figure 7, first row). Also, the SCR shows a southerly to southwesterly difference on this pressure level. The comparison with the climatology (cf. Figure 6, first row) suggests an overestimation of the low-level southwesterly monsoon winds over SSWA in all seasons and an underestimation of the northeasterly Harmattan winds over the Sahel region in winter. All differences > 0.5 m s−1 are significant at Si > 97.5%.

For the 20CR (Figure 8, first row), slight southwesterly deviations can be seen in the Northwest in DJF and MAM, while easterly deviations prevail from DJF to JJA especially along the southern coast of SSWA (more eastnortheasterly in DJF and MAM, southeasterly in JJA). The SCR displays some southwesterly (SON and DJF) to southerly deviation, similar to the NNR. The strongest difference in DJF is a northeasterly deviation over the western Central African Republic. In SON, the difference is generally weakest. In combination with the climatology (cf. Figure 6), the southwesterly deviation in the Northwest implies weakened Harmattan winds and trades in DJF and MAM, while the easterly deviation along the southern coast of SSWA means a significantly weakened (DJF/MAM) or more southerly (JJA) low-level monsoon in the 20CR compared to observations. In addition to the direction of the difference vectors relative to the observed monsoon flow and Harmattan/trades being opposite in the 20CR and the NNR, a further dissimilarity lies in the fact that the difference relative to the observed monsoon is much weaker in the 20CR and seems more confined to the coastal region. The differences >0.5 m s−1 are significant at Si = 99.9%.

3.3.2. 850 hPa pressure level

For the NNR, the overall structure of the difference does not change much over the course of the year (Figure 7, second row). The difference is largest on average over SSWA, over Niger, and over SCR in spring and summer. The direction of the difference vector is westnorthwest to westsouthwest in the first region, eastnortheast north and northeast of that region, respectively, and southeast in the latter region. The westerly difference over SSWA reaches its maximum in spring, summer and autumn. In the NW the absolute difference is relatively small over the whole year. The difference field points to an overestimation of the thickness of the low-level westerly monsoon flow in the NNR, and to an underestimation or even reversal of the climatological easterlies in the other seasons, especially in the coastal region (cf. Figure 6, second row). Here, all differences >0.5 m s−1 are again significant at Si > 97.5%.

In the case of the 20CR (Figure 8, second row), the differences are weaker than at 900 hPa in DJF and MAM, similar in JJA and slightly stronger in southern SSWA in SON, but have almost the same directions. Taken together with the climatology these differences signify weakened trades in the Northwest in all seasons in the 20CR, weakened easterlies in southern SSWA in SON, but stronger easterlies (a stronger and more southerly monsoon flow) along the southern coast of SSWA in DJF and MAM (in JJA). On average, the differences appear smallest on this pressure level, differing from the situation with the NNR. Again, the differences >0.5 m s−1 are significant at Si = 99.9%.

3.3.3. 700 hPa pressure level

On the 700 hPa level and for the NNR (Figure 7, third row), the absolute difference is relatively small except for the SCR, the spring and summer months over SSWA, and the autumn months over the eastern part of SSWA. A relatively large easterly difference shows up year-round east of the Gulf of Guinea. Apparently, the NNR seems to massively overestimate the easterly wind in the lower to mid troposphere in the equatorial region, although one has to be careful here due to the sparseness of data in that region. The large deviation might also indicate a problem with the observational data there. In spring and summer, the NNR exhibits a strong, divergent northeasterly to northwesterly deviation over SSWA. The difference is superimposed on the relatively strong climatological easterlies (cf. Figure 6, third row), indicating an overestimation of the monsoon return flow and African easterly jet (AEJ) in the reanalysis (the latter in the south and west of West Africa). All differences >0.5 m s−1 are significant at Si > 97.5%.

For the 20CR (Figure 8, third row), the differences are generally larger again than on the 850 hPa level, and also larger than on the 900 hPa level in JJA and SON. Additionally, their direction depends much more strongly on the season compared to the lower levels. DJF is characterised by a westerly deviation in the western part of West Africa, a southerly direction in the central part, and easterly or southeasterly deviations over western Chad, the western Central African Republic and the SCR. The picture looks similar in SON, except that the central part of SSWA also exhibits relatively strong westerly deviations in this season, especially towards the southern coast. On the other hand, MAM and JJA are clearly different, with more or less intense easterly deviations over most of the domain. In DJF, these differences represent stronger than observed mid-tropospheric westerlies in the Northwest and a slightly weaker than observed AEJ in the Southwest. In SON, the AEJ is additionally significantly weakened compared to observations in southern SSWA. In contrast, during MAM/JJA the AEJ appears partly much stronger than in the observations from the Southwest through to eastern SSWA, while the MAM westerlies in the Northwest have a tendency for being too weak in the 20CR. All differences >0.5 m s−1 are significant at Si > 99.5%.

3.3.4. 600 hPa pressure level

The observational data coverage falls off significantly on the 600 hPa pressure level (Figures 7 and 8, fourth row). Either the total number of observations decreases so that a fair weather bias cannot be safely excluded, or the number of years with available data is smaller than five, deemed insufficient to guarantee the representativeness of the data. Therefore, we show the respective vectors in a different colour.

In case of the NNR (Figure 7, fourth row), we find again an easterly difference in the equatorial region (SCR), although with reservations, as explained above. Over SSWA the northerly to easterly difference reaches its largest absolute values in spring and summer. In autumn, the southwesterly half of SSWA seems to be characterised by southerly to westerly deviations. In the coastal area, the maximum easterly deviation arguably emerges in spring, while the maximum northerly to easterly bias occurs in summer over the Sahel. Whereas the artificial difference flow field seems to have a more divergent character over the Sahel region, it appears rather convergent along the coast. In turn, the comparison of the difference fields with the observed winds suggests a too strong monsoon circulation in the NNR. Also, the strength of the AEJ, vertically centred between 600 and 700 hPa, seems to be overestimated, in agreement with the results found for the 700 hPa level. Despite of the smaller number of observations, all differences on this pressure level that are >0.5 m s−1 are still significant at Si > 95%.

For the 20CR (Figure 8, last row), the structure of the difference field looks similar to that on the 700 hPa level, although the strength of the differences differs regionally. The 20CR once again seems to underestimate the AEJ strength in DJF above the SW, in MAM along the southern coast of SSWA, and in SON over the Southwest and central SSWA. It overestimates its strength in MAM along the whole stretch from Lake Chad in the east to Mauritania in the west, and also in JJA in the southern Northwest and Southwest. The differences >0.5 m s−1 detected at this level are still significant at Si > 90% (except for Port-Gentil, where the difference is significant only at Si > 75%).

To sum up, the difference of the 20CR relative to the observations looks totally different from that of the NNR during the same period. Overall, the differences seem to be smaller for the 20CR than for the NNR, and the difference fields seem to be less directly connected to the climatological circulation. The station of Dolisie, Congo, stands out in both cases with respect to the magnitude of the difference on several levels and in several seasons, even though clearly smaller difference values are obtained for the 20CR than for the NNR at the 700 and 600 hPa levels. That the differences are smaller on average for the 20CR is surprising, since it assimilates only SLP and monthly mean SSTs. Generally, SLP is not estimated to be a very good indicator for circulation in the Tropics. SLP variability and spatial gradients are small due to the vanishing Coriolis acceleration in the horizontal plane close to the Equator. On the other hand, SSTs have also been assimilated into the NNR and should therefore not cause any differences. The latter might be connected to the more recent atmospheric NCEP global model version used in the 20CR compared to the NNR. It is also remarkable that the discrepancies between the 20CR and the observations do not seem to grow with altitude, a behaviour one would intuitively expect from the lack of upper-air information used. This might point to a preferably non-local determination of the circulation at the upper levels, e.g. by more remote tropical SSTs as contrasted to regional, tropical Atlantic SSTs, as has been highlighted for West Africa by a number of studies (Janicot et al., 1998; Camberlin et al., 2001; Joly and Voldoire, 2009).

3.4. 20CR seasonal mean difference fields on constant pressure levels for the period 1940–1947

Additionally to the period 1948–1957 (overlap with NNR), seasonal mean difference fields of the 20CR have been calculated for the earlier period 1940–1947 (Figure 9). Much less observational data are available during that time. Nevertheless, a comparison with the same plot for the period 1948–1957 (Figure 8) reveals a close correspondence of the spatial structure of the difference vectors at all pressure levels and in all seasons. A dissimilarity is found for the absolute magnitude of the difference, though. There is a tendency for a generally enhanced magnitude of the difference in the earlier period compared to the later one which might be caused by fewer surface observations entering the reanalysis during the earlier period. Erroneous observational upper-air data in the earlier period are unlikely to explain the larger deviations since the regional spatial structure of the difference is almost unchanged and overall not chaotic but organised, pointing more to a reanalysis deficiency, similar to what the analysis revealed for the NNR.

thumbnail image

Figure 9. As Figure 8 but for the period 1940–1947. A minimum of 20 observations on average per month of observation is required, and no large over- or underweighting of single calendar months is allowed (<20%). Furthermore, a minimum of 4 years of observations is necessary for each season and station. Due to the relatively low number of stations fulfilling these criteria in this early period, the rest of the difference vectors are also plotted on all levels, but with red arrows. Differences > 0.5 m s−1 are mostly significant at Si > 90%. Exceptions are listed in section 3.4.

Download figure to PowerPoint

In turn, significance of the differences was tested for all values > 0.5 m s−1. On the 900 hPa pressure level, the differences are generally significant at Si > 90%. On the 850, 700 and 600 hPa level, they are again mostly significant at least at Si = 90%. Exceptions here are Pointe-Noire in DJF, Kandi in JJA, and Tchaourou and Agadès in SON on the 850 hPa level, Tabou in DJF/MAM, and Agadès in JJA on the 700 hPa level. On the 600 hPa level, significance drops further for some stations/seasons, presumably due to the low number of observations. There, the differences are not significant at Si = 90% for the following stations and seasons: Pointe-Noire (DJF), Agadès (DJF), Tchaourou (MAM/JJA), Bouaké (MAM/JJA/SON), Tabou (MAM), Dakar (MAM), Cotonou (JJA), Néma (JJA), Thiès (JJA), Kandi (SON), Bobo-Dioulasso (SON) and Tambacounda (SON).

3.5. Interannual variability in the reanalyses compared to the observations

In this subsection we focus on the JJA season, the 925/900 and 700 hPa pressure levels (corresponding closely to the levels of the low-level monsoon and of the return flow/AEJ in SSWA), and stations that have enough observational data available at least for the full period 1948–1957 and for all seasons (nine stations). Figures 10, 11 and 12 display the interannual JJA U wind variability on the 700 and the V wind variability on the 900 and 925 hPa pressure levels as represented in the observations and the reanalyses. As for the other comparative analyses presented in this study, only data for simultaneous observation–reanalysis events have been selected, and the time series are plotted for the maximum overlapping time period of the reanalyses and the observations (1940–1957 for the 700 and 900 hPa levels, 1948–1957 for the 925 hPa level).

thumbnail image

Figure 10. Interannual variability of the JJA U wind at the 700 hPa pressure level for nine selected stations as represented in the observations (CHUAN, continuous) and in the reanalyses (NNR, dashed, and 20CR, dash-dotted). The linear correlation coefficients between the time series of the CHUAN observations and the time series of the reanalyses are given in the boxes.

Download figure to PowerPoint

thumbnail image

Figure 11. As Figure 9 but for the JJA 900 hPa V wind (only CHUAN and 20CR). Note the different scale of the y-axis.

Download figure to PowerPoint

thumbnail image

Figure 12. As Figure 9 but for the JJA 925 hPa V wind (only CHUAN and NNR). Note the different scale of the y-axis.

Download figure to PowerPoint

Figure 10 shows that the interannual variability of the JJA zonal wind at 700 hPa is not very well modelled in the reanalyses in general. There is quite some correspondence of the time series in Cotonou, Benin (r = 0.92, only 20CR), Nouadhibou, Mauritania (r = 0.72 for 20CR, r = 0.55 for NNR), Saint-Louis, Senegal (r = 0.57 for 20CR, r = 0.61 for NNR) and Dakar, Senegal (r = 0.44, only NNR), even though biases are clearly visible. The first station is located in the coastal region of SSWA, while the latter are all in the Northwest. Elsewhere, the correspondence of the interannual variability in the reanalyses and the observations is very weak. Additionally, at some stations (e.g. Néma, Mauritania or Ziguinchor, Senegal) the amplitude of the interannual variability seems to be underestimated by both reanalyses compared to the observations. Based on the generally low correlation coefficients in the Southwest and SSWA (except 20CR for Cotonou), we would not recommend the use of the reanalyses for studies of the interannual variability of the AEJ during the period 1940–1957.

Figure 11 shows that the correlation coefficients between the 20CR and the CHUAN observational time series are also quite low in most cases for the JJA meridional wind at 900 hPa. In this case, the correspondence is highest in Conakry, Guinea (r = 0.90) as well as Saint-Louis and Ziguinchor, Senegal (r = 0.73 and 0.58). The amplitude of the variations seems underestimated again at some stations (e.g. Nouadhibou, Bouaké, Ziguinchor).

Figure 12 demonstrates that the NNR is not doing a better job at low levels (925 hPa) than the 20CR. The correspondence between the time series is best for Conakry, Guinea (r = 0.59, but with underestimated amplitude) and Nouadhibou, Mauritania (r = 0.76), and very low elsewhere. In the studied region, V wind absolute values as well as variability usually tend to be smaller than the U winds and their interannual variability (cf. also Figures 3, 4 and 5 of the online supporting information, which give the interannual variability of the V wind at 700 and of the U winds at 900/925 hPa, and note the different y axis ranges at 700 and 900/925 hPa). The results presented in Figures 11 and 12 do not support the use of the reanalyses for studies of the interannual variability of the low-level monsoon flow in West Africa during the years 1940–1957.

Table II contains all the correlation coefficients and their significance (two-sided test based on Fisher's z-transform, verified by a Spearman's rank correlation test) for all three levels and the U and V winds in JJA.

Table II. Correlation coefficients between the time series of the JJA seasonal means of the U and V winds at the 700 and 900/925 hPa pressure levels in the observations (CHUAN) and in the reanalyses (NNR and 20CR, respectively). Only the values for stations which have enough observational data at least for the full period 1948–1957 and for all seasons are shown. Correlations that are significant at the 80% level are in italics, at the 90% level underlined and in italics, and at the 95% level bold, underlined and in italics (two-sided test using Fisher's z-transformation, equation image). Significant correlations confirmed by a distribution-free Spearman rank correlation test are additionally marked with an asterisk.
StationNNR (1948–1957)  20CR (1940–1957, available years) 
 U700V700U925V925U700V700U900V900
Northwest        
Nouadhibou0.54778*0.45120.79313*0.76016*0.7164*0.200420.45517*0.42716*
Saint-Louis0.60601*0.229180.423310.300840.56504*0.0793580.61587*0.73072*
Dakar0.436960.0022920.67054*0.2005−0.03536−0.092720.88811*0.22284
Néma0.078904−0.085760.48268*0.366760.12227−0.085760.10593−0.35968
Southwest        
Ziguinchor−0.139320.193750.79782*0.267850.235370.37492*0.5362*0.58266
Conakry−0.23283−0.084−0.024960.59278*0.237630.554480.150520.8997*
SSWA        
Cotonou−0.179460.8443*0.129420.312380.91925*0.65917*0.46628*0.2333
Bouaké0.46647−0.41314−0.31798−0.58595*0.39320.177140.004659−0.14693
Lomé0.287370.064897−0.045320.156840.165910.114720.34354−0.17222

On an interannual time-scale, if one looks at the position of the stations in Figure 1, the NNR wind components correlate generally much stronger with the observations in the Northwest than in the Southwest or in SSWA, especially north of 20°N, but even north of 15°N close to the Atlantic coast (e.g. Saint-Louis, Senegal). Nevertheless, the largest correlation coefficient in this region is 0.79, corresponding to an explained variance of only 62%. Furthermore, the correlations are only strongly significant (Si > 95%) for both zonal and meridional wind at 925 hPa in the outermost Northwest (Nouadhibou) and for the zonal wind at 925 hPa in the outermost west (Dakar). Further to the south (including the Southwest and SSWA) and towards the interior of the continent (cf. Néma), the correlation drops (especially on the 700 hPa level) or even reverses its sign (actually significantly at Si = 90% for the meridional wind on the 925 hPa level in Bouaké, Ivory Coast). The wind on the 925 hPa level in the Southwest (Ziguinchor (only zonal), Conakry (only meridional)) and the meridional wind on the 700 hPa level in Cotonou, Benin, are an exception to this rule and correlate significantly with the reanalysis (Si = 95%, Si = 90%, and Si = 95%). We conclude that the only region with clearly positive correlations for all pressure levels and wind components is the far Northwest, north of 15°N.

For the 20CR, the reanalysis wind components correlate again relatively strongly (except the 700 hPa meridional wind) and significantly (Si = 90% or 95%) in the Northwest. Further towards the south (with the exception of the 900 hPa zonal wind at Dakar) and in the continental interior, no significant correlations are found. The maximum explained variance in the Northwest is even slightly smaller (53%) than for the NNR. In contrast, the picture looks clearly better than for the NNR in the Southwest. There, both stations exhibit significant correlations for the meridional winds on the 700 and 900 hPa pressure levels. However, the explained variance is again relatively low (mean: 35%; maximum: 81%). The 20CR also seems to reflect the interannual variability in SSWA better than the NNR at first glance. Yet significantly positive correlations at Si > 90% are once again only found for Cotonou. For this station, the explained variance amounts to 46% on average (maximum 85% for the zonal wind at 700 hPa). We may conclude that the 20CR exhibits partly lower (Dakar, Nouadhibou), partly higher (Saint-Louis) correlations with observed winds in the Northwest compared to the NNR. For the other regions, the correlations with CHUAN observations tend to be higher than for the NNR (Southwest and SSWA). Nevertheless, only one station (out of three in the latter subregion) shows some sign of strongly significant correlations, especially at the higher pressure level.

Summarising the results found for the NNR and the 20CR we can state that the explained variance for the interannual variability of the JJA winds in the West African Monsoon region remains low on average for all subregions studied. In large parts of the region, no significant correlations were found at all. For one station in SSWA, even a significantly negative correlation was detected in the case of the NNR. Therefore, both reanalyses cannot be recommended for the use in studies dealing with the interannual variability of winds in the West African region during the period 1940–1957. A comparison of Figures 10–12 with Figures 3–5 (online supporting material) shows that the interannual variability encompasses the magnitude as well as the direction of the difference vectors (differences of the U and V wind between reanalyses and observations do not generally scale with each other). Finally, these results taken together with the complicated diurnal cycle of the differences (section 3.2) preclude a simple, constant in time seasonal correction of the reanalyses.

4. Discussion and conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and methods
  5. 3. Results
  6. 4. Discussion and conclusions
  7. Acknowledgements
  8. References
  9. Supporting Information

In this paper we have quantified what we would interpret mainly as the wind bias between two reanalysis products (NNR and 20CR) and observations from CHUAN, www.historicalupperair.org) for the period 1948–1957 (and additionally 1940–1947 for the 20CR), and from the lower to the mid troposphere above West and western Central Africa. We will use the term ‘bias’ in the following to express that in general the organised spatial structure of the difference vector fields of the reanalyses winds relative to the observational data points more to deficiencies of the reanalyses than to erroneous observational data.

We defined seven subregions that are characterised by similar behaviour with respect to the climatological winds as well as the NNR vertical bias profiles. We found that the NNR wind bias is clearly significant over the whole region and on all pressure levels (Si > 95% for values > 0.5 m s−1) and exhibits a clear altitude dependency and a seasonal cycle. For all subregions except the Northwest (30 out of 37 stations), our results point to a highly significant (Si = 99.9%) diurnal cycle of the bias. The bias of the NNR reaches absolute values of more than 5 m s−1, larger than the interannual variability (cf. also section 3.5). The bias fields point to an overestimation of the low-level monsoon including its mid-tropospheric return flow by the reanalysis during all seasons and a too thick monsoon layer. The Harmattan was found to be too weak in the NNR over the Sahel region in winter (DJF). The AEJ (and possibly the equatorial easterlies) appears overestimated in the NNR. Even though the findings for the uppermost pressure level for which difference fields have been presented (600 hPa) should be viewed with caution, they appear consistent with the lower levels. Furthermore, the general picture emerging from the synopsis of all pressure levels offers a ‘physical’ interpretation of the NNR bias (too strong monsoon circulation in the NNR).

The quantification of the 20CR difference fields relative to the observations yielded a totally different picture which seems less easy to interpret in a physical sense than that of the NNR. The bias is again significant on the single station level (Si > 95% in most cases during 1948–1957, Si > 90% in most cases during 1940–1947, with much less data available, again for values > 0.5 m s−1). On average, the differences were found to be smaller than in the case of the NNR during the overlapping period 1948–1957 and of similar magnitude during the earlier period 1940–1947, but they also have a completely different spatial structure. One station in the SCR (Dolisie, Congo) stood out again with respect to the magnitude of the difference on several levels and in several seasons, even though clearly smaller absolute difference values are reached than for the NNR at the 700 and 600 hPa levels. Owing to a lack of sufficient nearby station data at higher levels, it remains unclear whether the large deviations there (leading to the conclusion of possibly overestimated equatorial easterlies in the NNR) reflect a problem with the observational data or whether they are real. At the lower levels, the 20CR bias implies weakened Harmattan winds and trades in the Northwest (see Figure 1) in DJF and MAM, and a significantly weakened (DJF/MAM) or more southerly (JJA) low-level monsoon along the southern coast of SSWA compared to observations. Therefore, not only the direction of the bias of the monsoon flow and the Harmattan/trades is opposite in the 20CR and the NNR at the low levels, but additionally the bias of the monsoon is much weaker in the 20CR and seems more confined to the coastal region, at least in the overlapping period 1948–1957. Furthermore, the biases appear smallest on the 850 hPa pressure level on average for the 20CR. This differs from the NNR bias, which showed about the same magnitude at this level compared to the level below. At the higher levels, the 20CR seems to underestimate the AEJ strength in DJF above the SW (see Figure 1), in MAM along the southern coast of SSWA, and in SON over the Southwest and central SSWA. In contrast, it overestimates its strength in MAM along the whole stretch from Lake Chad in the east to the westernmost part of West Africa, and also in JJA in the southern Northwest and Southwest.

An analysis of the interannual variability in the reanalyses compared to the observations revealed that the observed interannual variability is only insufficiently modelled in the reanalyses. Furthermore, we found interannual variability of the biases identified before that encompasses the magnitude as well as the direction of the wind difference vectors. Taken together with the diurnal cycle of the bias, this precludes a simple correction of the reanalyses. Therefore, a use of the reanalyses for studies of the interannual wind variability above tropical West Africa during the period 1940–1957 cannot be recommended.

As indicated before, the biases of the NNR in the period before 1958 found over West Africa are probably connected to the small number of upper-air observations assimilated into the reanalysis. Similar problems can be expected for the early phase before 1958 in the Asian monsoon region, where a lack of assimilated upper-air data is also evident, at least before 1955 (Indochina, Malaysia, Indonesia) or even 1957 (China). Therefore, the next step will be to carry out a follow-up study for this region, for which the understanding of the monsoon mechanism and the circulation variability is at least of equal importance as for the Sahel region, due to the very large population density in that region and the vulnerability of the regional agro-economy to monsoon failure.

Acknowledgements

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and methods
  5. 3. Results
  6. 4. Discussion and conclusions
  7. Acknowledgements
  8. References
  9. Supporting Information

AS and SB were funded by the Swiss National Science Foundation (SNF PP002-102731). We also acknowledge funding through the SNF project EVALUATE. NCEP-NCAR 50-Year Reanalysis data have been downloaded from the NCAR website. Twentieth Century Reanalysis data have been downloaded from the NOAA ESRL website. Support for the Twentieth Century Reanalysis Project dataset is provided by the U.S. Department of Energy, Office of Science Innovative and Novel Computational Impact on Theory and Experiment (DOE INCITE) program, and Office of Biological and Environmental Research (BER), and by the National Oceanic and Atmospheric Administration Climate Program Office. Thanks go to the anonymous reviewers and the editor for their comments, which significantly helped improve the quality of the first draft of this paper.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and methods
  5. 3. Results
  6. 4. Discussion and conclusions
  7. Acknowledgements
  8. References
  9. Supporting Information
  • Betts AK, Ball JH, Bosilovich M, Viterbo P, Zhang Y, Rossow WB. 2003a. Intercomparison of water and energy budgets for five Mississippi subbasins between ECMWF reanalysis (ERA-40) and NASA Data Assimilation Office fvGCM for 1990–1999. J. Geophys. Res. 108: 8618, DOI: 10.1029/2002JD003127.
  • Betts AK, Ball JH, Viterbo P. 2003b. Evaluation of the ERA-40 surface water budget and surface temperature for the Mackenzie River basin. J. Hydrometeorol. 4: 11941211.
  • Betts AK, Ball JH, Viterbo P, Dai A, Marengo J, 2005. Hydrometeorology of the Amazon in ERA-40. J. Hydrometeorol. 6: 764774.
  • Bony S, Sud S, Lau KM, Susskind J, Saha S. 1997. Comparison and satellite assessment of NASA/DAO and NCEP-NCAR reanalyses over tropical ocean: atmospheric hydrology and radiation. J. Climate 10: 14411462.
  • Bromwich DH, Fogt RL, Hodges KI, Walsh JE. 2007. A tropospheric assessment of the ERA-40, NCEP, and JRA-25 global reanalyses in the polar regions. J. Geophys. Res. 112: D10111, DOI: 10.1029/2006JD007859.
  • Camberlin P, Janicot S, Poccard I. 2001. Seasonality and atmospheric dynamics of the teleconnection between African rainfall and tropical sea-surface temperature: Atlantic vs. ENSO. Int. J. Climatol. 21: 9731005.
  • Compo GP, Whitaker JS, Sardeshmukh PD. 2006. Feasibility of a 100-year reanalysis using only surface pressure data. Bull. Am. Meteorol. Soc. 87: 175190.
  • Compo GP, Whitaker JS, Sardeshmukh PD, Matsui N, Allan RJ, Yin X, Gleason BE, Vose RS, Rutledge G, Bessemoulin P, Brönnimann S, Brunet M, Crouthamel RI, Grant AN, Groisman PY, Jones PD, Kruk MC, Kruger AC, Marshall GJ, Maugeri M, Mok HY, Nordli Ø, Ross TF, Trigo RM, Wang XL, Woodruff SD, Worley SJ. 2011. The Twentieth Century Reanalysis Project. Q. J. R. Meteorol. Soc. 137: 128.
  • Dee DP, Uppala S. 2009. Variational bias correction of satellite radiance data in the ERA-Interim reanalysis. Q. J. R. Meteorol. Soc. 135: 18301841.
  • Grant AN, Brönnimann S, Haimberger L, 2008. Recent arctic warming vertical structure contested. Nature 455: E2E3, DOI: 10.1038/nature07257.
  • Grant AN, Brönnimann S, Ewen T, Nagurny A. 2009. A new look at radiosonde data prior to 1958. J. Climate 22: 32323247.
  • Gutowski WJ, Chen YB, Ötles Z. 1997. Atmospheric water vapor transport in NCEP-NCAR reanalyses: comparison with river discharge in the central United States. Bull. Am. Meteorol. Soc. 78: 19571969.
  • Higgins RW, Mo KC, Schubert SD. 1996. The moisture budget of the central United States in spring as evaluated in the NCEP/NCAR and the NASA/DAO reanalyses. Mon. Weather Rev. 124: 939963.
  • Hines KM, Bromwich DH, Marshall GJ. 2000. Artificial surface pressure trends in the NCEP-NCAR reanalysis over the southern ocean and Antarctica. J. Climate 13: 39403952.
  • Janicot S, Harzallah A, Fontaine B, Moron V. 1998. West African monsoon dynamics and eastern equatorial Atlantic and Pacific SST anomalies (1970–88). J. Climate 11: 18741882.
  • Joly M, Voldoire A. 2009. Influence of ENSO on the West African monsoon: temporal aspects and atmospheric processes. J. Climate 22: 31933210.
  • Jury MR. 2009. An intercomparison of observational, reanalysis, satellite, and coupled model data on mean rainfall in the Caribbean. J. Hydrometeorol. 10: 413430.
  • Kalnay E, Kanamitsu M, Kistler R, Collins W, Deaven D, Gandin L, Iredell M, Saha S, White G, Woollen J, Zhu Y, Chelliah M, Ebisuzaki W, Higgins W, Janowiak J, Mo KC, Ropelewski C, Wang J, Leetmaa A, Reynolds R, Jenne R, Joseph D. 1996. The NCEP/NCAR 40-Year Reanalysis Project. Bull. Am. Meteorol. Soc. 77: 437471.
  • Karam HN, Bras RL. 2008. Estimates of net atmospheric moisture flux convergence over the Amazon Basin: a comparison of reanalysis products. J. Hydrometeorol. 9: 10351047.
  • Kinter JL, Fennessy MJ, Krishnamurthy V, Marx L. 2004. An evaluation of the apparent interdecadal shift in the tropical divergent circulation in the NCEP-NCAR reanalysis. J. Climate 17: 349361.
  • Kistler R, Kalnay E, Collins W, Saha S, White G, Woollen J, Chelliah M, Ebisuzaki W, Kanamitsu M, Kousky V, van den Dool H, Jenne R, Fiorino M. 2001. The NCEP–NCAR 50-Year Reanalysis: monthly means CD-ROM and documentation. Bull. Am. Meteorol. Soc. 82: 247268.
  • Ladd C, Bond NA. 2002. Evaluation of the NCEP/NCAR reanalysis in the NE Pacific and the Bering Sea. J. Geophys. Res. 107: 31583168.
  • Lenters JD, Coe MT, Foley JA. 2000. Surface water balance of the continental United States, 1963–1995: regional evaluation of a terrestrial biosphere model and the NCEP/NCAR reanalysis. J. Geophys. Res. 105: 2239322425.
  • Lüpkes C, Vihma T, Jakobson E, König-Langlo G, Tetzlaff A. 2010. Meteorological observations from ship cruises during summer to the central Arctic: a comparison with reanalysis data. Geophys. Res. Lett. 37: L09810, DOI: 10.1029/2010GL042724.
  • Nedoluha GE, Alfred J, Benson CM, Hoppel KW, Wickert J, König-Langlo G. 2007. A comparison of radiosonde and GPS radio occultation measurements with meteorological temperature analyses in the Antarctic vortex, 1998–2004. J. Geophys. Res. 112: D16304, DOI: 10.1029/2007JD008928.
  • Oikonomou EK, O'Neill A. 2006. Evaluation of ozone and water vapor fields from the ECMWF reanalysis ERA-40 during 1991–1999 in comparison with UARS satellite and MOZAIC aircraft observations. J. Geophys. Res. 111: D14109, DOI: 10.1029/2004JD005341.
  • Onogi K, Koide H, Sakamoto M, Kobayashi S, Tsutsui J, Hatsushika H, Matsumoto T, Yamazaki N, Kamahori H, Takahashi K, Kato K, Oyama R, Ose T, Kadokura S, Wada K. 2005. JRA-25: Japanese 25-year re-analysis project: progress and status. Q. J. R. Meteorol. Soc. 131: 32593268.
  • Parker DJ, Burton RR, Diongue-Niang A, Eliis RJ, Felton M, Taylor CM, Throncroft CD, Bessemoulin P, Tompkins CM. 2005. The diurnal cycle of the West African monsoon circulation. Q. J. R. Meteorol. Soc. 131: 28392860.
  • Parrondo MC, Yeal M, Gil M, von der Gathen P, Ochoa H. 2007. Mid-winter lower stratosphere temperatures in the Antarctic vortex: comparison between observations and ECMWF and NCEP operational models. Atmos. Chem. Phys. 7: 435441.
  • Pettenuzzo D, Large WG, Pinardi N. 2010. On the corrections of ERA-40 surface flux products consistent with the Mediterranean heat and water budgets and the connection between basin surface total heat flux and NAO. J. Geophys. Res. 115: C06022, DOI: 10.1029/JC005631.
  • Poccard I, Janicot S, Camberlin P. 2000. Comparison of rainfall structures between NCEP/NCAR reanalyses and observed data over tropical Africa. Clim. Dynam. 16: 897915.
  • Simmons AJ, Jones PD, da Costa Bechtold V, Beljaars ACM, Kållberg PW, Saarinen S, Uppala SM, Viterbo P, Wedi N. 2004. Comparison of trends and low-frequency variability in CRU, ERA-40, and NCEP/NCAR analyses of surface air temperature. J. Geophys. Res. 109: D24115, DOI: 10.1029/2004JD005306.
  • Stickler A, Grant AN, Ewen T, Ross TF, Vose RS, Comeaux J, Bessemoulin P, Jylhä K, Adam WK, Jeannet P, Nagurny A, Sterin AM, Allan R, Compo GP, Griesser T, Brönnimann S. 2010. The Comprehensive Historical Upper-Air Network (CHUAN). Bull. Am. Meteorol. Soc. 91: 741751.
  • Tjernström M, Graversen RG. 2009. The vertical structure of the lower Arctic troposphere analysed from observations and the ERA-40 reanalysis. Q. J. R. Meteorol. Soc. 135: 431443.
  • Trenberth KE, Guillemot CE. 1998. Evaluation of the atmospheric moisture and hydrological cycle in the NCEP/NCAR reanalyses. Clim. Dynam. 14: 213231.
  • Trenberth KE, Stepaniak DP, Hurrell JW. 2001. Quality of reanalyses in the Tropics. J. Climate 14: 14991510.
  • Uppala SM, Kållberg PW, Simmons AJ, Andrae U, da Costa Bechtold V, Fiorino M, Gibson JK, Haseler J, Hernandez A, Kelly GA, Li X, Onogi K, Saarinen S, Sokka N, Allan RP, Andersson E, Arpe K, Balmaseda MA, Beljaars ACM, van de Berg L, Bidlot J, Bormann N, Caires S, Chevallier F, Dethof A, Dragosavac M, Fisher M, Fuentes M, Hagemann S, Hólm E, Hoskins BJ, Isaksen L, Janssen PAEM, Jenne R, McNally AP, Mahtouf J.-F., Morcrette J.-J., Rayner NA, Saunders RW, Simon P, Sterl A, Trenberth KE, Untch A, Vasiljevic D, Viterbo P, Woollen J. 2005. The ERA-40 re-analysis. Q. J. R. Meteorol. Soc. 131: 29613012.
  • van Noije TPC, Eskes HJ, van Weele M, van Velthoven PFJ. 2004. Implications of the enhanced Brewer–Dobson circulation in European Centre for Medium-Range Weather Forecasts reanalysis ERA-40 for the stratosphere–troposphere exchange of ozone in global chemistry transport models. J. Geophys. Res. 109: D19308, DOI: 10.1029/2004JD004586.
  • von Storch H, Zwiers FW. 2001. Statistical Analysis in Climate Research (1st paperback edn). Cambridge University Press: Cambridge, UK.
  • Zhao M, Running SW, Nemani RR. 2006. Sensitivity of Moderate Resolution Imaging Spectroradiometer (MODIS) terrestrial primary production to the accuracy of meteorological reanalyses. J. Geophys. Res. 111: G01002, DOI: 10.1029/2004JG000004.

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and methods
  5. 3. Results
  6. 4. Discussion and conclusions
  7. Acknowledgements
  8. References
  9. Supporting Information
FilenameFormatSizeDescription
qj_854_sm_suppinfofigs.pdf1958KSupporting Information
qj_854_sm_suppinfosimultaneity.pdf114KSupporting Information
qj_854_sm_suppinfoatmosphere.pdf128KSupporting Information

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.