Abstract

The impact of different surface vegetations on long-term surface temperature change is estimated by subtracting reanalysis trends in monthly surface temperature anomalies from observation trends over the last four decades. This is done using two reanalyses, namely, the 40-yr ECMWF (ERA-40) and NCEP–NCAR I (NNR), and two observation datasets, namely, Climatic Research Unit (CRU) and Global Historical Climate Network (GHCN). The basis of the observation minus reanalysis (OMR) approach is that the NNR reanalysis surface fields, and to a lesser extent the ERA-40, are insensitive to surface processes associated with different vegetation types and their changes because the NNR does not use surface observations over land, whereas ERA-40 only uses surface temperature observations indirectly, in order to initialize soil temperature and moisture. As a result, the OMR trends can provide an estimate of surface effects on the observed temperature trends missing in the reanalyses. The OMR trends obtained from observation minus NNR show a strong and coherent sensitivity to the independently estimated surface vegetation from normalized difference vegetation index (NDVI). The correlation between the OMR trend and the NDVI indicates that the OMR trend decreases with surface vegetation, with a correlation < −0.5, indicating that there is a stronger surface response to global warming in arid regions, whereas the OMR response is reduced in highly vegetated areas. The OMR trend averaged over the desert areas (0 < NDVI < 0.1) shows a much larger increase of temperature (∼0.4°C decade−1) than over tropical forest areas (NDVI > 0.4) where the OMR trend is nearly zero. Areas of intermediate vegetation (0.1 < NDVI < 0.4), which are mostly found over midlatitudes, reveal moderate OMR trends (approximately 0.1°–0.3°C decade−1). The OMR trends are also very sensitive to the seasonal vegetation change. While the OMR trends have little seasonal dependence over deserts and tropical forests, whose vegetation state remains rather constant throughout the year, the OMR trends over the midlatitudes, in particular Europe and North America, exhibit strong seasonal variation in response to the NDVI fluctuations associated with deciduous vegetation. The OMR trend rises up approximately to 0.2°–0.3°C decade−1 in winter and early spring when the vegetation cover is low, and is only 0.1°C decade−1 in summer and early autumn with high vegetation. However, the Asian inlands (Russia, northern China with Tibet, and Mongolia) do not show this strong OMR variation despite their midlatitude location, because of the relatively permanent aridity of these regions.

1. Introduction

Global mean surface temperature time series derived from in situ observations reveal the interdecadal global warming over the last several decades (Houghton et al. 2001). Many studies reported that this upward trend is significantly a result of primary human impacts such as greenhouse gases (Houghton et al. 2001) and land use (Pielke et al. 2002). The anthropogenic land-use impact on surface warming may become more important as the surface vegetation changes in the form of urbanization, agricultural activity, and deforestation.

The impact of surface temperature changes forced by different regional vegetation types is not well documented. Only urban impact has been assessed by comparing observations in cities with those in rural areas (Easterling et al. 1996; Hansen et al. 2001). But this approach is only applicable to urban effects, and the estimated signals vary with the criteria in classifying urban and rural areas.

The present study is motivated by the difficulty in separating the surface temperature change signals due to global and regional forcings in the observed data. The basis of this study is the fact that the surface temperature change response to land vegetation types is not present in the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis (NNR) surface data, and is only partially present in the 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis (ERA-40), while the station data include not only local surface forcings but the large-scale atmospheric warming signal resulting from greenhouse effects, natural decadal variability, and volcanoes (Pielke et al. 2002; Kalnay and Cai 2003; Zhou et al. 2004; Frauenfeld et al. 2005; Lim et al. 2005; Kalnay et al. 2006). Impact of the vegetation cover can extend to some extent into the free atmosphere and thus may influence the atmospheric and surface reanalysis (Kabat et al. 2004), but NNR is substantially insensitive to surface processes associated with different vegetation types because it does not use surface observations over land in the assimilation (Kistler et al. 2001; Kalnay and Cai 2003). Instead, NNR surface temperature fields are estimated from the upper-air information combined with model parameterizations of surface processes (Lim et al. 2005) so that the NNR provides a dynamically complete dataset of atmospheric variables. ERA-40 is somewhat more sensitive to local surface processes than NNR because the surface temperature observations are used in the initialization of soil temperature and moisture (Simmons et al. 2004).

Evaluation of reanalyzed tropical temperature time series archived from ERA-40 (Palmer et al. 1990; Betts et al. 2003; Simmons et al. 2004) and NNR (Kalnay et al. 1996; Kistler et al. 2001) indicates that the climatic trend derived from reanalysis data capture the upward surface temperature trends but that the trend is not identical to that of observed data (Chelliah and Ropelewski 2000; Hegerl and Wallace 2002; Kalnay and Cai 2003; Simmons et al. 2004; Lim et al. 2005; Pepin et al. 2005; Kalnay et al. 2006). Specifically, Simmons et al. (2004) and Lim et al. (2005) reported that the hemispheric average in two reanalyses (ERA-40 and NNR) have a smaller warming trend than that of observations. They suggest that this smaller trend arises from the fact that the reanalysis data do not adequately reproduce the long-term surface climatic trend driven by the impact of independent land-cover types.

These characteristics of the reanalysis provide us with the possibility of detecting surface temperature change signals due to regional land vegetation types by taking the difference between observed and reanalysis temperature time series [observation minus reanalysis (OMR)]. The present study, therefore, has as objectives 1) to find the relationship between OMR trend and the regional land vegetation types in terms of surface vegetation index and 2) to estimate the temperature change signal as a function of surface vegetation indexes.

It has been argued that errors such as reanalysis inhomogeneity in time, model systematic errors, including the lack of trends in the greenhouse gases, and observation biases could contaminate the true surface temperature change signal (Trenberth 2004; Vose et al. 2004; Cai and Kalnay 2004). Cai and Kalnay (2005) showed analytically that a reanalysis made with a model without anthropogenic forcing could capture the observed trends if they are present in the observations assimilated. Our OMR analysis tries to minimize the impact of those errors by 1) averaging for the relatively homogeneous reanalysis period, 2) computing the trend of the anomalies with respect to the annual cycle, and 3) choosing the most reliable observation data currently available.

There have been several studies using OMR trends for estimating the regional surface warming signal driven by different land vegetation types. Kalnay and Cai (2003) assessed the decadal surface warming trend associated with regional land uses over the eastern United States by subtracting the reanalysis trend from the observed one. Kalnay et al. (2006) found regions of OMR warming and cooling, in good agreement with the regional trends obtained by Hansen et al. (2001) when using nightlights to identify rural and urban stations. Zhou et al. (2004), Frauenfeld et al. (2005), and Lim et al. (2005), using the same method, estimated reasonable values for surface warming trends caused by Chinese urbanization, Tibetan Plateau land uses, and the Northern Hemispheric land vegetation types, respectively. All these studies except Lim et al. (2005) concentrated on regional geographical areas, and thus are not sufficient to draw an overall conclusion for the globe.

In summary, the advantage of the OMR approach is that the removal of the reanalysis estimates from the surface observations makes it possible to isolate the near-surface warming signals associated with the regional surface vegetation types from the warming signals resulting from large-scale atmospheric forcings (e.g., greenhouse gases and volcanoes). As a result, it is expected that the OMR trends can give an estimate of the surface temperature change signal arising from the different types of regional surface vegetation. In the present study, we will attempt to find the relationship between OMR surface warming patterns and land vegetation types in terms of surface vegetation status using normalized difference vegetation index (NDVI; Sellers 1985; James and Kalluri 1994) made from satellite-derived greenness values.

Section 2 introduces the observation and the reanalysis temperature data, and the vegetation index data used in this study. Surface temperature time series of observation, reanalysis, and the OMR averaged over major landmasses are shown in section 3. Section 4 delineates the geographical distribution of the surface vegetation index, the relationship between OMR trends and the surface vegetation index, and estimates the surface temperature change signal associated with NDVI. The variation of the OMR in response to the seasonal NDVI change is described in section 5, followed by a summary and discussion given in section 6.

2. Data

For this study we use monthly surface temperature from two reanalyses (NNR and ERA-40) and from two gridded datasets based on surface observations [Global Historical Climatology Network (GHCN) (http://www.ncdc.noaa.gov; Peterson and Vose 1997) and Climatic Research Unit (CRU) (http://www.cru.uea.ac.uk; Jones and Moberg 2003)], all covering the period 1960–99. We downloaded observational (GHCN and CRU) data available on a 5° × 5° grid. For consistency, the reanalyses data have also been linearly interpolated to the same 5° × 5° GHCN and CRU grid.

Since the station coverage declined during the 1990s in the CRU measurements, CRU anomalies were calculated with respect to normals for 1961–90. For consistency, anomalies of other data have been calculated with respect to their own climatic normals for 1961–90.

Like the reanalyses, surface observational datasets may also have some limitations, because the observations themselves have some biases. The quality of surface measurement variables tends to depend on the coverage of the surface observation network, homogeneity, and the accuracy of surface measurements. In the present study, we use the CRU and GHCN surface temperatures as observed datasets. CRU and GHCN data values are similar, since they draw from over 90% common data, and differ mostly in their processing. Jones and Moberg (2003) indicate that the gridded (5° × 5°) database available on the CRU Web site (http://www.cru.uea.ac.uk) comprises 5159 station records, which are more densely distributed over midlatitudes. The data are also corrected by newly homogenized series with adjustment of the variance of individual gridbox series to remove the effects of changing station numbers through time. As to GHCN (Peterson and Vose 1997), the quality of surface temperature values are enhanced by including a century-scale dataset with monthly surface observations from ∼7000 stations from around the world, which makes it possible to improve regional-scale analyses, particularly in previously data-sparse areas. Rigorous and objective homogeneity adjustments are performed to decrease the effect of nonclimatic factors on the time series. Therefore, these two observation datasets are accepted as a reliable basis for investigating and assessing surface temperature change signal associated with different vegetation types from the OMR time series. However, we note that both CRU and GHCN have fewer station data per grid point in the tropics than in the midlatitudes, which could reduce the reliability of the gridded observation data in the tropical region.

NDVI data (Sellers 1985; James and Kalluri 1994), downloaded from http://daac.gsfc.nasa.gov/, are used to find out if there is a relationship between the distribution of the surface vegetation and its seasonal changes, and the decadal OMR trends. The NDVI satellite-derived surface greenness values (Bounoua et al. 2000) are produced using the measurements from the Advanced Very High Resolution Radiometer (AVHRR) on board the National Oceanic and Atmospheric Administration (NOAA) polar-orbiting meteorological satellites. The dataset contains the global monthly composites of the NDVI at 1° resolution covering the period from 1981 to 1994. The reflectance measured from channel 1 (visible: 0.58–0.68 μm) and channel 2 (near-infrared: 0.725–1.0 μm) are used to calculate the index. The NDVI value is defined as the ratio of the difference to the total reflectance: (channel 2 − channel 1)/(channel 2 + channel 1). Green leaves commonly have larger reflectances in the near-infrared than in the visible range. Clouds, water, and snow have larger reflectances in the visible than in the near-infrared, so that negative values of the vegetation index may correspond to snow or ice cover, whereas the difference in reflectance is almost zero for bare soils such as deserts. As a result, NDVI values can range from −1.0 to 1.0, but typical ranges are from 0.1 up to 0.7, with higher values associated with greater density and greenness of plant canopies. We have also made comparisons with another NDVI dataset derived from Global Inventory Modeling and Mapping Studies (GIMMS; Tucker et al. 2005) and the results remained similar.

3. Regional surface temperature time series of observation and reanalysis

Surface temperature anomalies averaged over major land regions in the Northern Hemisphere (NH) derived from the two reanalyses and two observations are plotted in Fig. 1. Anomalies are further adjusted to have zero mean over the last 10 yr (1993–2002) since the biases of reanalyses are smallest for the most recent years (Simmons et al. 2004). The key features in Fig. 1 are as follows:

  • (i) The two surface observation datasets (CRU and GHCN) in each panel on the left are nearly indistinguishable (Figs. 1a–d), showing a gradual upward trend of surface temperature over Asia, Europe, North and South America, and the Africa region. The two reanalyses are also in remarkable agreement with the observations in terms of capturing the interannual variability and the long-term trends.

  • (ii) Nevertheless, it is evident that the reanalyses exhibit a smaller warming trend than observations, as reported in Kalnay and Cai (2003) and Lim et al. (2005). This feature is found in all regions (Figs. 1a–d). Because of this, the OMR time series (Figs. 1e–h) obtained by subtracting reanalyses from observations show a positive trend in every region.

  • (iii) The right panels in Fig. 1 indicate that overall, the OMR time series with ERA-40 yields a smaller warming trend than that derived from the NNR (Figs. 1e–h). This is to be expected because ERA-40 uses to some extent the surface observation in their assimilation system, while the NNR does not use the surface information. Land surface temperature and soil moisture in the ERA-40 are estimated by assimilating the CRU surface observations (Jones and Moberg 2003) in an offline mode. Therefore, a portion of the surface warming trend associated with regional characteristics above the surface (land-cover types) may be reflected in the ERA-40, resulting in a smaller OMR trend than that derived from NNR (Simmons et al. 2004; Lim et al. 2005) (Figs. 1e–h), although ERA-40 could also contain bias passed from the in situ surface measurements. From Fig. 1 we conclude that both reanalyses could serve as a dataset for surface temperature trend analysis because they do not use the surface temperature observations directly, but for the purpose of assessing the surface temperature trend associated with surface vegetation characteristics, the NNR has the advantage of not using surface observations to initialize the soil temperature and moisture.

Fig. 1.

Time series (10-yr running mean) of the surface temperature anomalies (°C) averaged over the (a) east Asia, (b) Europe and west Asia, (c) North and South America, and (d) Africa area. Anomaly values are obtained by removing the 30-yr mean from 1961 to 1990 and they are further adjusted to have zero mean over the last 10 yr (1993–2002). (e)–(h) As in (a)–(d) but for their OMR (CRU–ERA-40, CRU–NNR, GHCN–ERA-40, and GHCN–NNR) time series.

Fig. 1.

Time series (10-yr running mean) of the surface temperature anomalies (°C) averaged over the (a) east Asia, (b) Europe and west Asia, (c) North and South America, and (d) Africa area. Anomaly values are obtained by removing the 30-yr mean from 1961 to 1990 and they are further adjusted to have zero mean over the last 10 yr (1993–2002). (e)–(h) As in (a)–(d) but for their OMR (CRU–ERA-40, CRU–NNR, GHCN–ERA-40, and GHCN–NNR) time series.

4. The OMR trends associated with surface vegetation

a. Geographical surface vegetation field and its annual range

Figure 2 represents the geographical distribution of the NDVI for (a) all 12 months, (b) June–August (JJA), (c) December–February (DJF), and (d) the seasonal difference (JJA–DJF). The maps depict the pattern of global greenness, along with their annual ranges. The well-known desert areas such as Sahara, Middle East, western China, and Mongolia are classified in the least vegetated region (Figs. 2a–c). These areas show little seasonal vegetation change (Fig. 2d). Arctic areas are also characterized by small NDVI in summer and negative NDVI (frozen water) in winter. On the other hand, tropical evergreen forest regions including the equatorial Africa, Southeast Asia, and the Maritime Continent, central America, and Amazon areas show a large vegetation index with little seasonal change (Figs. 2a–c). Note that there are several grids over India and Indochina peninsula where the wintertime NDVI is greater than that in summer. This is because seasonal vegetation growth is a few months out of phase with monsoon precipitation over those regions. Monthly NDVI fields (not shown here) reveal that the leaf area index in the Indian region has its minima in the premonsoon season (April, May) and begins to increase slowly with the onset of the Indian summer monsoon. The summer monsoon gets to central India only by mid-June, and the period prior to that is very stressful for vegetation given the scorching heat and dryness of the premonsoon period. Rainfall in June–July is a relief, and vegetation begins to come back, reaching the largest NDVI values in the postmonsoon season. The main crop cycle, wheat, called “rabi crops” begins with planting in fall and is nurtured by the milder winter monsoon rainfall; harvesting occurs in spring. Therefore, although initially surprising, the slightly higher NDVI in winter than in summer observed over India is representative of the actual vegetation characteristics. We have found that the same spatial NDVI features are reproduced in the other NDVI dataset derived from GIMMS (http://gimms.gsfc.nasa.gov/) group (Tucker et al. 2005).

Fig. 2.

Surface vegetation map derived from NDVI. Vegetation index is averaged over (a) 1981–94, (b) summer (JJA), and (c) winter (DJF). (d) The seasonal NDVI difference (JJA–DJF).

Fig. 2.

Surface vegetation map derived from NDVI. Vegetation index is averaged over (a) 1981–94, (b) summer (JJA), and (c) winter (DJF). (d) The seasonal NDVI difference (JJA–DJF).

Midlatitude regions, which generally comprise cropland, mixed (broadleaf and needleleaf) forests, shrubland, grass, and needleleaf tree forests, have a much more conspicuous seasonal change (Figs. 2b–d). In those regions the large vegetation index in the JJA period drops drastically during the DJF period. While the seasonal vegetation change is strong in the European countries and North American region, midlatitude central Asia (Russia, northern China, and Mongolia) shows a relatively weak seasonal NDVI change due to the annually consistent aridity (e.g., Gobi Desert) over the region (Fig. 2d).

b. Relationship between the OMR trend and the vegetation index

We now relate the surface temperature change signals estimated by OMR to the different surface vegetation indexes. To this end, the decadal OMR trend at each grid is scatterplotted as a function of annual mean NDVI (Bounoua et al. 2000). This kind of approach is also found in Hanamean et al. (2003) in relating NDVI to the 850–700-hPa-layer mean temperature change derived from the NNR for the Colorado area.

As in Kalnay and Cai (2003), the OMR trend per decade at each grid point is obtained by taking the average of two decadal mean differences, that is, 1990s − 1980s and 1970s − 1960s, to avoid the small jumps in the reanalysis associated with the major addition of satellite observations in 1979. The decadal observation, reanalysis, and OMR trends are scatterplotted with the NDVI values for the 20°S–50°N latitudinal band and all longitudes, an area where most of landmass is found (Hurrell and Trenberth 1998). NH high latitudes, including the Arctic zone, are not included in this analysis due to their different climate response mechanism to surface vegetation (Robock 1983; Wang and Key 2005) compared with mid- and low-latitude regions. Analysis of high-latitude regions is not within the scope of this study and will be investigated later.

As shown in Fig. 3a, decadal trends in GHCN observations show no significant relationship with the NDVI (r = −0.07), presumably because they reflect all climate change signals. However, the trend in NNR reanalysis (Fig. 3d) is significantly proportional to the vegetation index (r = 0.56), indicating that it is missing the relationship demonstrated in modeling experiments (Xue and Shukla 1993; Dai et al. 2004; Hales et al. 2004) showing stronger surface warming in arid areas with low vegetation. This lack of reproduction of the surface temperature change signal associated with the impact of vegetation types is also present to a lesser extent in the ERA-40 reanalysis (r = 0.17), as shown in Fig. 3b. Partial inclusion of surface information in ERA-40 makes the correlation with NDVI less positive than NNR. The outliers in the scatterplots, that is, large negative ERA-40 trends and large NDVI, were all in the tropics, within 20° latitude (Fig. 3b), whereas no such outliers were found for NNR (Fig. 3d).

Fig. 3.

Scatter diagram between the NDVI and the decadal surface temperature trend of (a) GHCN, (b) ERA-40, (c) GHCN–ERA-40, (d) NNR, and (e) GHCN–NNR over the 20°S–50°N, 0°–360°E region. Data have been spatially smoothed to remove the extreme outliers. Here r is the correlation coefficient of all the data points.

Fig. 3.

Scatter diagram between the NDVI and the decadal surface temperature trend of (a) GHCN, (b) ERA-40, (c) GHCN–ERA-40, (d) NNR, and (e) GHCN–NNR over the 20°S–50°N, 0°–360°E region. Data have been spatially smoothed to remove the extreme outliers. Here r is the correlation coefficient of all the data points.

Because of these characteristics of reanalysis data, the decadal OMR trend obtained from GHCN–ERA-40 and GHCN–NNR, respectively, is negatively correlated with the surface vegetation index (r = −0.32, r = −0.67), demonstrating that the strong (weak) surface warming response to the surface aridity (greenness) is adequately represented by OMR, especially for “observation minus NNR.” As shown in Figs. 3c and 3e, the inverse proportionality of decadal OMR trend to vegetation index is clearer for GHCN–NNR than GHCN–ERA-40 because the former better extracts the surface temperature change signal associated with vegetation types.

Figure 4 is the same as Fig. 3 but with the CRU instead of the GHCN observations. The key features in Fig. 4 are identical to those delineated in Fig. 3. While CRU observation trends are not correlated with the NDVI (r = 0.11), the OMR trends have strong negative correlation with surface greenness, especially for CRU − NNR (r = −0.58).

Fig. 4.

As in Fig. 3 but for (a) CRU, (c) CRU–ERA-40, and (e) CRU–NNR.

Fig. 4.

As in Fig. 3 but for (a) CRU, (c) CRU–ERA-40, and (e) CRU–NNR.

c. Surface temperature change as a function of vegetation index

Decadal OMR trends with respect to the surface vegetation index are assessed. Figure 5 depicts the annual mean OMR trends as a function of NDVI values with 0.1 intervals. The OMR trend values at each grid point are averaged for the same surface vegetation index values. Trend values are represented by closed circles, along with the error range at 95% significance level by cross marks. The number of stations used for calculation of these trends is given in Table 1.

Fig. 5.

Assessment of annual mean OMR trend (°C decade−1) (lines with filled circles) as a function of annual mean NDVI. The OMR trend for the (a), (b) NNR reanalysis and (c), (d) ERA-40 reanalysis. Lines with cross mark represent the error range at 95% significance level.

Fig. 5.

Assessment of annual mean OMR trend (°C decade−1) (lines with filled circles) as a function of annual mean NDVI. The OMR trend for the (a), (b) NNR reanalysis and (c), (d) ERA-40 reanalysis. Lines with cross mark represent the error range at 95% significance level.

Table 1.

The number of grid points used for calculation of OMR trends in Fig. 5.

The number of grid points used for calculation of OMR trends in Fig. 5.
The number of grid points used for calculation of OMR trends in Fig. 5.

The two independent reanalyses appear to have a very similar dependence of the OMR trends with respect to surface vegetation as well as their statistical significance levels. The key features in the OMR trends for the approximately 20°S–50°N latitudes are

  • (i) The OMR trend decreases with the surface vegetation index. The quantitative estimate from the NNR reanalysis reveals that for vegetation index greater than 0.4 there is near-zero additional near-surface contribution to temperature change (Figs. 5a,b). Error ranges at 95% significance level support the statistical confidence of this assessment. The highest vegetation index area generally comprises the broadleaf forest over tropical forest regions. As discussed in modeling works by Shukla et al. (1990), Xue and Shukla (1993), and Giambelluca et al. (1997), this area is characterized by the strong transpiration and evaporative cooling feedback from the leaves, resulting in the suppression of surface warming.

  • (ii) The OMR trends over the moderate surface vegetation index (approximately 0.2–0.4) area are in the range of approximately 0.1°–0.2°C decade−1 (CRU–NNR and GHCN–NNR) (Figs. 5a,b). These areas are mainly composed of midlatitudinal crop and grass land, deciduous broadleaf or needleaf trees, and shrubs. Because they are less green than the tropical forest areas, the cooling feedbacks from leaves are weaker than those in tropical forest areas. This contributes to the moderation of surface warming over these areas.

  • (iii) Areas of less than 0.1 greenness with sparse vegetation show the largest warming associated with surface effects, ∼0.3°C decade−1 (CRU–NNR and GHCN–NNR) (Figs. 5a,b). In areas where soil moisture is very limited and the land surface is characterized by bare soil, the evaporation negative feedback would be negligible, explaining a larger regional surface warming response under the same amount of radiative forcings, as discussed in Shukla et al. (1990), Bounoua et al. (2000), Hoffmann and Jackson (2000), Dai et al. (2004), Hales et al. (2004), Diffenbaugh (2005), and Saito et al. (2006).

  • (iv) The OMR trends for the ERA-40 reanalysis (Figs. 5c,d) are similar but less pronounced than the trends for the NNR due to the partial inclusion of the surface processes associated with land vegetation types in the ERA-40, which are not included in the NNR.

5. Surface temperature trend response to the seasonal vegetation change

a. Monthly variation of correlation between OMR trend and vegetation

As shown in Fig. 2, the vegetation index has a seasonal variation, especially for midlatitudes. The spatial correlations between the NDVI and the OMR trend for each month is now calculated to understand the seasonal variation of OMR trends associated with seasonal vegetation changes. Since the NDVI shows seasonal changes at fixed location, we expect that the OMR trend would also exhibit a temporal variation in response to the seasonal vegetation change. Month-to-month variation of correlations between the NDVI and the OMR trend identifies that they range from −0.6 to −0.35 (Fig. 6, lines with cross mark). The observed negative correlations indicate that the OMR trend satisfies the negative relationship with the NDVI throughout the year; that is, there is an increase (decrease) in OMR trend with decreasing (increasing) vegetation. Seasonal difference in the OMR trend also indicates that the trend is more negatively correlated with NDVI when the overall vegetation is low (in the winter) than when it is high (the summer). On the other hand, the observed decadal trends (GHCN and CRU) show weak correlations with NDVI seasonal variations (less than 0.2), indicating little significant relationship with surface vegetation types throughout the year, because, as observed before, the observed trends are dominated by combined atmospheric and surface warming effects, not only by vegetation effects (Fig. 6, lines with open circles).

Fig. 6.

Month-to-month variation of the correlation between NDVI and the decadal OMR trend over the 20°S–50°N, 0°–360°E region. The trend values have been temporally smoothed by 5-month running mean to remove the extreme outliers. Lines with open circle denote the CRU (long dash) and the GHCN (solid), and the plus signs denote the CRU–NNR (long dash) and GHCN–NNR (solid).

Fig. 6.

Month-to-month variation of the correlation between NDVI and the decadal OMR trend over the 20°S–50°N, 0°–360°E region. The trend values have been temporally smoothed by 5-month running mean to remove the extreme outliers. Lines with open circle denote the CRU (long dash) and the GHCN (solid), and the plus signs denote the CRU–NNR (long dash) and GHCN–NNR (solid).

b. Variation of the OMR trend with the seasonal vegetation change

Based on the understanding that the OMR trend varies with the temporal vegetation change, the seasonal variation of the OMR trends is estimated for several major regions over the NH. Five different regions are chosen here to investigate the response of the OMR trends for the NNR to the variation of the surface vegetation. In Fig. 7 we consider two extremes with low amplitude in the NDVI seasonal cycle. First, desert areas (Sahel and Middle East) are selected as areas of little vegetation, limited soil moisture, and little seasonal vegetation change. Second, low-latitude broadleaf forests (equatorial Africa, Amazon, Indochina peninsula, and Maritime Continent) are chosen as representative of high vegetation, abundant soil moisture, and evergreenness. In Fig. 8, by contrast, we select areas of strong seasonal vegetation change between summer and winter in midlatitudes, for example, Europe and United States, and the arid midlatitude central Asia.

Fig. 7.

Seasonal variation of the OMR trend (°C decade−1) in response to the seasonal vegetation change over (a) desertic area (Sahara and Middle East) and (b) tropical evergreen forest (equatorial Africa and Asia, and Amazon). Solid lines with closed circle and closed square denote the seasonal variation of OMR trend of GHCN–NNR and CRU–NNR, respectively. Dashed lines are plotted by switching NNR to ERA-40 reanalysis. Seasonal NDVI change is denoted by solid line with open circles.

Fig. 7.

Seasonal variation of the OMR trend (°C decade−1) in response to the seasonal vegetation change over (a) desertic area (Sahara and Middle East) and (b) tropical evergreen forest (equatorial Africa and Asia, and Amazon). Solid lines with closed circle and closed square denote the seasonal variation of OMR trend of GHCN–NNR and CRU–NNR, respectively. Dashed lines are plotted by switching NNR to ERA-40 reanalysis. Seasonal NDVI change is denoted by solid line with open circles.

Fig. 8.

As in Fig. 7 but for the midlatitude landmass: (a) Europe, (b) United States, and (c) central Asia with aridity covering Russia, northern China, and Mongolia.

Fig. 8.

As in Fig. 7 but for the midlatitude landmass: (a) Europe, (b) United States, and (c) central Asia with aridity covering Russia, northern China, and Mongolia.

The seasonal variation of the OMR trend for NNR clearly shows that strong surface warming (∼0.4°C decade−1) beyond what would be expected from atmospheric warming is observed over the desert area throughout the year (see solid lines with closed circles and squares in Fig. 7a) as the NDVI remains close to zero (solid line with open circles). Limited soil moisture and little evaporative cooling feedback year-round appear to drive the consistent strong warming over this area (Dai et al. 2004; Hales et al. 2004).

In contrast, tropical evergreen forest area in Fig. 7b shows little surface warming throughout the year (solid lines with closed circles and squares) due to the evergreenness (solid line with open circles). Surface cooling by evaporation, transpiration, and soil moisture associated with surface greenness and tropical humid climate remains effective throughout the year, yielding little surface warming in every month (Xue and Shukla 1993).

In midlatitudes such as Europe (Fig. 8a) and the United States (Fig. 8b), OMR trends exhibit a strong annual cycle as the NDVI has strong seasonal changes, with high vegetation in the summer and low vegetation in the winter. The corresponding OMR trend fluctuates seasonally almost out of phase with the seasonal NDVI oscillation. Therefore, we can conclude that the surface warming response to the regional vegetation status over the midlatitude tends to be strong in winter and early spring (approximately 0.2°–0.3°C decade−1) when the vegetation is low but weak in summer and early autumn (0.1°C decade−1) with high vegetation.

The OMR trend over the midlatitude central Asia (Russia, northern China, and Mongolia) does not show a strong seasonal variation, as shown in Fig. 8c. As seen in the NDVI, this area is characterized by being arid throughout the year. Therefore, the OMR trend remains high with little seasonal change despite being in a midlatitude geographical location.

We clearly demonstrated how sensitively the OMR trend for NNR responds to the seasonal vegetation change. However, the variation of OMR trend for ERA-40 does not show any significant relationship with the seasonal vegetation change in Figs. 7 and 8 (see dashed line). As discussed before, the partial inclusion of surface temperature information in the ERA-40 makes the OMR temperature change signal associated with seasonal vegetation change weaker.

6. Summary and discussions

Anomalies in monthly surface temperature time series derived from two reanalyses (NNR and ERA-40) and two observational datasets (GHCN and CRU) are analyzed by the OMR approach (observation minus reanalysis) to 1) investigate the relationship between the OMR trend and land vegetation types and 2) assess the surface temperature change signal by the impact of independent land vegetation types from the OMR. The rationale for the OMR approach is that while reanalyses contain the large-scale temperature change signals that could be forced by greenhouse gases, volcanoes, and natural decadal variability, the NNR and (to a lesser extent) the ERA-40 are insensitive to regional surface processes associated with different land vegetation types because little surface data or information were used in the data assimilation process. Pronounced features identified from analysis are as follows:

  1. The long-term trends for both observations and reanalyses averaged over several areas show a gradual warming, with greater upward trend in observations over the last 4 decades. This is caused by the poor reproduction of surface temperature change signal associated with impacts of regional vegetation types in the reanalyses data. As a result, the difference time series between observation and reanalyses (OMR) grows with time.

  2. The positive OMR trend is larger for NNR than for ERA-40 because of the different data assimilations used in the two reanalyses. The NNR used no surface observations over land, whereas ERA-40 used surface temperature observations to initialize soil temperature and moisture. This makes the NNR reanalysis more insensitive to surface processes than the ERA-40. As a result of this lack of surface information in the NNR, more surface temperature change signal resulting from the impact of different land vegetation types are captured in the OMR time series using NNR than ERA-40 (Kalnay and Cai 2003; Zhou et al. 2004; Frauenfeld et al. 2005; Lim et al. 2005). This fact gives an indication that the OMR time series for NNR would provide useful information to assess the multidecadal surface temperature change signal with regard to different land vegetation types. Nevertheless, the general dependence of OMR on vegetation type is clearly similar between the two reanalyses.

  3. For a clearer demonstration of the relationship, the decadal OMR trends at each grid were correlated with NDVI. The results prove that the decadal OMR trend is inversely proportional to NDVI with statistical significance (correlation < −0.5). The OMR trend for the NNR yields a more negative correlation with NDVI [correlation = −0.58 (CRU–NNR) and −0.67 (GHCN–NNR)] than the OMR trend for the ERA-40 does [correlation = −0.26 (CRU–ERA-40) and correlation = −0.32 (GHCN–ERA-40)]. This suggests that the decadal trend of the “observation–NNR” substantially accounts for the impact on surface warming effects associated with different land vegetation types, absent in the NNR. This feature is robust with respect to seasonal changes in the NDVI. Correlations range from −0.6 to −0.35 throughout the year, indicating that for the NNR the decadal OMR trend varies with seasonal NDVI change at fixed locations to maintain the negative correlation.

  4. The surface temperature change signal inferred from the OMR trend is also assessed as a function of surface vegetation index. It shows that the trend is very sensitive to the different surface greenness. The strongest warming trend due to surface processes is found over the desert areas [Sahara, Middle East, western China (Tibet), and Mongolia] (∼0.3°–0.4°C decade−1) whereas tropical forest areas (equatorial Africa, Southeast Asia, Maritime Continent, and Amazon), are associated with negligible warming or slight cooling (∼0°C decade−1). These results are consistent with Dai et al. (2004), Hales et al. (2004), and Diffenbaugh (2005), who argue that there is stronger (weaker) warming in barren (vegetated) areas due to the response of surface albedo, soil moisture, and evaporative cooling feedback to solar radiative forcing. Several midlatitude areas where the vegetation density lies in between desert and tropical forest show a moderate decadal warming (∼0.2°C decade−1).

  5. The OMR trend patterns are strongly dependent on seasonal vegetation change for midlatitude areas such as Europe and North America, where the strong seasonal vegetation that takes place shows a weak warming trend (∼0.1°C decade−1) in summer and early autumn and strong warming (approximately 0.2°–0.3°C decade−1) in winter and early spring. This is in agreement with the studies of Shukla et al. (1990) and Xue and Shukla (1993) who have shown that vegetation changes by forestation (deforestation) suppresses (enhances) the surface warming effect.

In contrast, tropical forest areas and desert areas (Sahara and Middle East) tend to show a constant OMR trend throughout the year, as their regions exert little vegetation change with season. Despite the geographical midlatitude location, the OMR trend in the central Asia region remains high without any noticeable seasonal change because of the stationary aridity throughout the year.

The findings obtained in this study support our argument that the surface temperature change signal associated with different land vegetation types are reasonably well captured by the OMR approach. The analysis demonstrates that lower long-term trends in the reanalysis surface temperature than observations are attributable to the absence in surface data in the data assimilation procedure. As a result, the OMR approach facilitates isolating the impact of vegetation types on long-term surface temperature trend by removing large-scale global warming signal as recorded in the reanalysis from the surface observation.

These findings could be affected by other errors that might arise from the reanalysis inhomogeneity in time, from model systematic errors and observation biases. Our OMR analysis tried to minimize the impact of those errors by 1) averaging the trends for relatively homogeneous reanalysis periods, 2) calculating the trends for the anomalies with respect to a 30-yr annual cycle, and 3) choosing the most reliable observation data currently being used. The clear relationship between the OMR trend and NDVI suggests that the effect of these errors is relatively small in comparison with the temperature change signal that we tried to isolate.

Note also that the NDVI is applied under the assumption that the global distribution of the surface vegetation for each calendar month is approximately constant. The real NDVI time series for each calendar month exhibit some interannual variation (not shown), but the amplitude of the changes is small, especially on a decadal scale. We believe that this assumption is reasonable since there have been no large interannual NDVI changes (e.g., a switch between forest and desertic bare soil) on a 5° × 5° horizontal grid scale that would be needed to have a noticeable effect on the OMR trend values.

Acknowledgments

This study was supported by the National Science Foundation (ATM-0403211 and ATM-0403518) and by the Climate Change Data and Detection program of the NOAA Office of Global Programs (GC04-259). ERA-40 data were obtained from the ECMWF public Web server, http://data.ecmwf.int/data. We thank three anonymous reviewers for very helpful and constructive comments. We also express our gratitude to Profs. S. Nigam, J. Shukla, J. Srinivasan, and S. Gadgil for their insightful comments on why the NDVI seems to be out of phase with precipitation in India.

REFERENCES

REFERENCES
Betts
,
A. K.
,
J. H.
Ball
, and
P.
Viterbo
,
2003
:
Evaluation of the ERA-40 surface water budget and surface temperature for the Mackenzie River basin.
J. Hydrometeor.
,
4
,
1194
1211
.
Bounoua
,
L.
,
G. J.
Collatz
,
S. O.
Los
,
P. J.
Sellers
,
D. A.
Dazlich
,
C. J.
Tucker
, and
D. A.
Randall
,
2000
:
Sensitivity of climate to changes in NDVI.
J. Climate
,
13
,
2277
2292
.
Cai
,
M.
, and
E.
Kalnay
,
2004
:
Impact of land-use change on climate: Response to the comments by Rose et al. and Trenberth.
Nature
,
427
,
214
.
Cai
,
M.
, and
E.
Kalnay
,
2005
:
Can reanalysis have anthropogenic climate trends without model forcing?
J. Climate
,
18
,
1844
1849
.
Chelliah
,
M.
, and
C. F.
Ropelewski
,
2000
:
Reanalyses-based tropospheric temperature estimates: Uncertainties in the context of global climate change detection.
J. Climate
,
13
,
3187
3205
.
Dai
,
A.
,
K. E.
Trenberth
, and
T.
Qian
,
2004
:
A global dataset of Palmer drought severity index for 1870–2002: Relationship with soil moisture and effects of surface warming.
J. Hydrometeor.
,
5
,
1117
1130
.
Diffenbaugh
,
N. S.
,
2005
:
Atmosphere–land cover feedbacks alter the response of surface temperature to CO2 forcing in the western United States.
Climate Dyn.
,
24
,
237
251
.
Easterling
,
D. R.
,
T. C.
Peterson
, and
T. R.
Karl
,
1996
:
On the development and use of homogenized climate datasets.
J. Climate
,
9
,
1429
1434
.
Frauenfeld
,
O. W.
,
T.
Zhang
, and
M. C.
Serreze
,
2005
:
Climate change and variability using European Centre for Medium-Range Weather Forecasts reanalysis (ERA-40) temperatures on the Tibetan Plateau.
J. Geophys. Res.
,
110
.
D02101, doi:10.1029/2004JD005230
.
Giambelluca
,
T. W.
,
D.
Hölscher
,
T. X.
Bastos
,
R. R.
Frazão
,
M. A.
Nullet
, and
A. D.
Ziegler
,
1997
:
Observations of albedo and radiation balance over postforest land surfaces in the eastern Amazon basin.
J. Climate
,
10
,
919
928
.
Hales
,
K.
,
J. D.
Neelin
, and
N.
Zeng
,
2004
:
Sensitivity of tropical land climate to leaf area index: Role of surface conductance versus albedo.
J. Climate
,
17
,
1459
1473
.
Hanamean
,
J. R.
,
R. A.
Pielke
,
C. L.
Castro
,
D. S.
Ojima
,
B. C.
Reed
, and
Z.
Gao
,
2003
:
Vegetation greenness impacts on maximum and minimum temperatures in northeast Colorado.
Meteor. Appl.
,
10
,
203
215
.
Hansen
,
J.
,
R.
Ruedy
,
M.
Sato
,
M.
Imhoff
,
W.
Lawrence
,
D.
Easterling
,
T.
Peterson
, and
T.
Karl
,
2001
:
A closer look at United States and global surface temperature change.
J. Geophys. Res.
,
106
,
23947
23964
.
Hegerl
,
G. C.
, and
J. M.
Wallace
,
2002
:
Influence of patterns of climate variability on the difference between satellite and surface temperature trends.
J. Climate
,
15
,
2412
2428
.
Hoffmann
,
W. A.
, and
R. B.
Jackson
,
2000
:
Vegetation–climate feedbacks in the conversion of tropical savanna to grassland.
J. Climate
,
13
,
1593
1602
.
Houghton
,
J. T.
,
Y.
Ding
,
D. J.
Griggs
,
M.
Noguer
,
P. J.
van der Linden
,
X.
Dai
,
K.
Maskell
, and
C. A.
Johnson
,
2001
:
Climate Change 2001: The Scientific Basis.
Cambridge University Press, 881 pp
.
Hurrell
,
J. W.
, and
K. E.
Trenberth
,
1998
:
Difficulties in obtaining reliable temperature trends: Reconciling the surface and satellite microwave sounding unit records.
J. Climate
,
11
,
945
967
.
James
,
M. E.
, and
S. N. V.
Kalluri
,
1994
:
The Pathfinder AVHRR land data set: An improved coarse-resolution data set for terrestrial monitoring.
Int. J. Remote Sens.
,
15
,
3347
3363
.
Jones
,
P. D.
, and
A.
Moberg
,
2003
:
Hemispheric and large-scale surface air temperature variations: An extensive revision and an update to 2001.
J. Climate
,
16
,
206
223
.
Kabat
,
P.
, and
Coeditors
,
2004
:
Vegetation, Water, Humans and the Climate: A New Perspective on an Interactive System.
Global Change—The IGBP Series, Springer, 566 pp
.
Kalnay
,
E.
, and
M.
Cai
,
2003
:
Impact of urbanization and land-use change on climate.
Nature
,
423
,
528
531
.
Kalnay
,
E.
, and
Coauthors
,
1996
:
The NCEP/NCAR 40-Year Reanalysis Project.
Bull. Amer. Meteor. Soc.
,
77
,
437
471
.
Kalnay
,
E.
,
M.
Cai
,
H.
Li
, and
J.
Tobin
,
2006
:
Estimation of the impact of land-surface forcings on temperature trends in eastern Unites States.
J. Geophys. Res.
,
111
.
D06106, doi:10.1029/2005JD006555
.
Kistler
,
R.
, and
Coauthors
,
2001
:
The NCEP–NCAR 50-Year Reanalysis: Monthly means CD-ROM and documentation.
Bull. Amer. Meteor. Soc.
,
82
,
247
267
.
Lim
,
Y-K.
,
M.
Cai
,
E.
Kalnay
, and
L.
Zhou
,
2005
:
Observational evidence of sensitivity of surface climate changes to land types and urbanization.
Geophys. Res. Lett.
,
32
.
L22712, doi:10.1029/2005GL024267
.
Palmer
,
T. N.
,
C.
Brankovic
,
F.
Molteni
,
S.
Tibaldi
,
L.
Ferranti
,
A.
Hollingsworth
,
U.
Cubasch
, and
E.
Klinker
,
1990
:
The European Centre for Medium-Range Weather Forecasts (ECMWF) program on extended-range prediction.
Bull. Amer. Meteor. Soc.
,
71
,
1317
1330
.
Pepin
,
N. C.
,
M.
Losleben
,
M.
Hartman
, and
K.
Chowanski
,
2005
:
A comparison of SNOTEL and GHCN/CRU surface temperatures with free-air temperatures at high elevations in the western United States: Data compatibility and trends.
J. Climate
,
18
,
1967
1985
.
Peterson
,
T. C.
, and
R. S.
Vose
,
1997
:
An overview of the Global Historical Climatology Network temperature database.
Bull. Amer. Meteor. Soc.
,
78
,
2837
2849
.
Pielke
Sr.,
R. A.
,
G.
Marland
,
R. A.
Betts
,
T. N.
Chase
,
J. L.
Eastman
,
J. O.
Niles
,
D. D. S.
Niyogi
, and
S. W.
Running
,
2002
:
The influence of land-use change and landscape dynamics on the climate system: Relevance to climate-change policy beyond the radiative effect of greenhouse gases.
Philos. Trans. Roy. Soc. London
,
360
,
1705
1719
.
Robock
,
A.
,
1983
:
Ice and snow feedbacks and the latitudinal and seasonal distribution of climate sensitivity.
J. Atmos. Sci.
,
40
,
986
997
.
Saito
,
K.
,
T.
Yasunari
, and
K.
Takata
,
2006
:
Relative roles of large-scale orography and land surface processes in the global hydroclimate. Part II: Impacts on hydroclimate over Eurasia.
J. Hydrometeor.
,
7
,
642
659
.
Sellers
,
P. J.
,
1985
:
Canopy reflectance, photosynthesis and transpiration.
Int. J. Remote Sens.
,
6
,
1335
1372
.
Shukla
,
J.
,
C.
Nobre
, and
P.
Sellers
,
1990
:
Amazon deforestation and climate change.
Science
,
247
,
1322
1325
.
Simmons
,
A. J.
, and
Coauthors
,
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
.
Trenberth
,
K. E.
,
2004
:
Climatology (communication arising): Rural land-use change and climate.
Nature
,
427
,
213
.
Tucker
,
C. J.
,
J. E.
Pinzon
,
M. E.
Brown
,
D. A.
Slayback
,
E. W.
Pak
,
R.
Mahoney
,
E. F.
Vermote
, and
N.
El Saleous
,
2005
:
An extended AVHRR 8-km NDVI dataset compatible with MODIS and SPOT vegetation NDVI data.
Int. J. Remote Sens.
,
26
,
4485
4498
.
Vose
,
R. S.
,
T. R.
Karl
,
D. R.
Easterling
,
C. N.
Williams
, and
M. J.
Menne
,
2004
:
Climate (communication arising): Impact of land-use change on climate.
Nature
,
427
,
213
214
.
Wang
,
X.
, and
J. R.
Key
,
2005
:
Arctic surface, cloud, and radiation properties based on the AVHRR polar pathfinder dataset. Part II: Recent trends.
J. Climate
,
18
,
2575
2593
.
Xue
,
Y.
, and
J.
Shukla
,
1993
:
The influence of land surface properties on Sahel climate. Part I: Desertification.
J. Climate
,
6
,
2232
2245
.
Zhou
,
L.
,
R. E.
Dickinson
,
Y.
Tian
,
J.
Fang
,
Q.
Li
,
R. K.
Kaufmann
,
C. J.
Tucker
, and
R. B.
Myneni
,
2004
:
Evidence for a significant urbanization effect on climate in China.
Proc. Natl. Acad. Sci. USA
,
101
,
9540
9544
.

Footnotes

Corresponding author address: Young-Kwon Lim, Center for Ocean–Atmospheric Prediction Studies, The Florida State University, Tallahassee, FL 32306-2840. Email: lim@coaps.fsu.edu

This article included in the Impacts of Land Use and Land Cover Change on the Atmosphere special collection.