Climate Dynamics

, Volume 25, Issue 6, pp 581–609

A comparison of a GCM response to historical anthropogenic land cover change and model sensitivity to uncertainty in present-day land cover representations

Authors

    • Department of GeographyUniversity of Kansas
  • Keith Oleson
    • National Center for Atmospheric Research
  • Gordon Bonan
    • National Center for Atmospheric Research
  • Linda Mearns
    • National Center for Atmospheric Research
  • Warren Washington
    • National Center for Atmospheric Research
  • Gerald Meehl
    • National Center for Atmospheric Research
  • Douglas Nychka
    • National Center for Atmospheric Research
Article

DOI: 10.1007/s00382-005-0038-z

Cite this article as:
Feddema, J., Oleson, K., Bonan, G. et al. Climate Dynamics (2005) 25: 581. doi:10.1007/s00382-005-0038-z

Abstract

This study assesses the sensitivity of the fully coupled NCAR-DOE PCM to three different representations of present-day land cover, based on IPCC SRES land cover information. We conclude that there is significant model sensitivity to current land cover characterization, with an observed average global temperature range of 0.21 K between the simulations. Much larger contrasts (up to 5 K) are found on the regional scale; however, these changes are largely offsetting on the global scale. These results show that significant biases can be introduced when outside data sources are used to conduct anthropogenic land cover change experiments in GCMs that have been calibrated to their own representation of present-day land cover. We conclude that hybrid systems that combine the natural vegetation from the native GCM datasets combined with human land cover information from other sources are best for simulating such impacts. We also performed a prehuman simulation, which had a 0.39 K ~higher average global temperature and, perhaps of greater importance, temperature changes regionally of about 2 K. In this study, the larger regional changes coincide with large-scale agricultural areas. The initial cooling from energy balance changes appear to create feedbacks that intensify mid-latitude circulation features and weaken the summer monsoon circulation over Asia, leading to further cooling. From these results, we conclude that land cover change plays a significant role in anthropogenically forced climate change. Because these changes coincide with regions of the highest human population this climate impact could have a disproportionate impact on human systems. Therefore, it is important that land cover change be included in past and future climate change simulations.

1 Introduction

Humans have altered the surface of the Earth significantly over the last few millennia and numerous studies have demonstrated these effects and simulated their causes (e.g., Betts 2001; Bonan 1999; Chase et al. 2000; Claussen et al. 2001; Costa and Foley 2000; Eastman et al. 2001; Henderson-Sellers et al. 1993; Nobre et al. 1991; Pielke and Avissar 1990; Pielke 2001; Pitman and Zhao 2000; Tsvetsinskaya et al. 2001; Williams and Balling 1996). There is a growing awareness that these processes cannot be ignored in future climate change simulations (Hansen et al. 1998; Pielke et al. 2002; Mahoney et al. 2003; Marland et al. 2003; Bounoua et al. 2002). Yet, these effects have not been included in IPCC scenario simulations (IPCC 2001). There are several reasons why they have not. First, land surface models in GCMs have only recently matured to the point where they can effectively simulate these changes (Avissar 1995; Pielke et al. 2002). Second, documenting human land cover change adequately for such studies is very complex (Nakićenović and Swart 2000). In addition, we have little knowledge of the consequences of using different land surface characterizations in GCM simulations (Oleson et al. 2004; Myhre and Myhre 2003).

Uncertainties in GCM responses to land cover change simulations fall into two major categories. First, uncertainty arises from differences in the models themselves. For example, the Project for the Intercomparison of Land surface Parameterization Schemes (PILPS) showed that given identical land cover characteristics, different land surface models gave disparate surface energy balance outcomes (Henderson-Sellers et al. 1993; Chen et al. 1997; Boone et al. 2004). Second, uncertainty arises from differences in the way land cover is represented in different land cover classification schemes (Oleson et al. 2004; Myhre and Myhre 2003). Land cover characterizations vary widely depending on the modeling group and modeling schemes used (e.g., Cox et al. 1999; Bonan 1994; Peylin et al. 1997). Furthermore, model surface characterizations are typically different from the land cover schemes used in the IPCC Special Report on Emissions Scenarios land cover change products (Nakićenović and Swart 2000).

This study assesses the sensitivity of a fully coupled ocean–atmosphere GCM to different present-day land cover representations, and will compare this uncertainty to the simulated climate change caused by historical anthropogenic land cover modification. As part of this study, we will use readily available SRES land cover change information to evaluate these issues (Nakićenović and Swart 2000; Alcamo 1994; Alcamo et al. 1998). Simulations use the Department of Energy Parallel Climate Model (DOE-PCM; Washington et al. 2000). This model has been used to assess a number of IPCC climate change scenarios, thereby providing a valuable frame of reference for this study. Specifically, we address the following questions:
  1. 1.

    What is the sensitivity of a fully coupled general circulation model to different present-day land cover representations?

     
  2. 2.

    Given that models are calibrated with respect to specific land cover representations, what is the best method for including alternative land cover scenarios into GCM simulations?

     
  3. 3.

    Has human land cover change had a significant impact on climate? Is this signal sufficiently large to suggest the need for including land cover change as part of IPCC based climate change scenarios?

     
  4. 4.

    Are global statistics adequate to detect climate signals from land cover change? Is there a discernable spatial pattern associated with land cover induced climate change?

     

2 Methods

We use the DOE-PCM (Washington et al. 2000). This model has been used to study a number of natural and human induced climate forcings (Ammann et al. 2003; Meehl et al. 2003; Santer et al. 2003a, b). The resolution of the atmosphere is T42, or roughly 2.8×2.8°, with 18 levels in the vertical. Resolution in the ocean is roughly 2/3 degree down to 1/2 degree in the equatorial tropics, with 32 levels. Sea ice is simulated using dynamic and thermodynamic formulations (Washington et al. 2000), and the NCAR Land Surface Model (NCAR LSM) is used as the land surface component (Bonan 1998). No flux corrections are used in the model, and a relatively stable climate is simulated in terms of global-mean temperature. For example, a preindustrial 1,000-year long control integration shows only a small cooling trend of globally averaged surface air temperatures of roughly 0.03 K per century.

To test the sensitivity of the DOE-PCM to land cover variations, we devised two sets of experiments. First, we evaluated the model response to three different “present-day” land cover representations: (1) the land cover scheme used in the NCAR LSM (labeled as LSMlc hereafter); (2) the IMAGE 2.2 present-day land cover, as documented in the IPCC SRES documentation (Alcamo 1994; Alcamo et al. 1998; Nakićenović and Swart 2000; labeled as IMAlc hereafter); and (3) a hybrid land cover where the background vegetation was in accordance with the original NCAR LSM land cover types and the human land cover types were derived from the IMAGE 2.2 present-day land cover classifications (labeled HYBlc hereafter). Second, to evaluate the potential historical impact of anthropogenic land cover change on climate, we compared simulations using the IMAGE 2.2 potential vegetation, representing prehuman conditions (labeled POTlc hereafter), and the commensurate present-day IMAGE 2.2 (IMAlc) land cover. We used the IMAGE 2.2 based scenarios because the NCAR LSM land cover dataset has no equivalent potential vegetation dataset and because this will facilitate comparisons to other IMAGE 2.2 based land cover simulations, which will be discussed in future work. Land cover is fixed for each simulation, and agriculture is represented as a mix of 85 percent crop vegetation and 15% bare ground. Phenological characteristics of the crop vegetation are typical of those for a mid-latitude summer crop, with greening starting in April, LAI reaching a maximum of three in late summer, and senescence ending by late October (see Bonan 1996). In the Southern Hemisphere, the timing of phenology is offset by 6 months from the Northern Hemisphere (Bonan 1996).

Each land cover experiment consisted of a DOE-PCM simulation run for 100 years with preindustrial atmospheric conditions (280 ppm CO2 concentrations). An existing preindustrial 1,000 year long-term control integration (no variation in forcing over time), with the standard NCAR LSM land cover was used for the LSMlc climate. Simulations were started from this run in the model year 400. The simulations were allowed to reach equilibrium over the first 60 years, and the last 40 years of the simulations were used for analysis purposes.

3 Development of the land cover datasets

The IPCC special report on emissions scenarios (Nakićenović and Swart 2000) presents results from six different research groups who independently compiled and modeled data to characterize future emissions scenarios. These simulations provided data on greenhouse gas (GHG) concentrations and aerosols over time, but several of the groups also included information on land cover as part of their scenario simulations. We used IPCC compatible land cover conditions as simulated by the IMAGE 2.2 model (Alcamo 1994; Alcamo et al. 1998). All data were obtained from the IMAGE 2.2 CDROM released by the Netherlands Environmental Assessment Agency (Rijksinstituut voor Volksgezondheid en Milieu; RIVM 2002) and represent the conditions presented in the IPCC SRES report for the IMAGE 2.2 scenario simulations (Nakićenović and Swart 2000).

The IMAGE 2.2 land cover changes are based on a number of assumptions and databases. Base or control experiments are derived from datasets calibrated to 1970 conditions (Alcamo 1994; Alcamo et al. 1998; RIVM 2002). As part of the land cover scenarios a potential vegetation dataset representing prehuman conditions was created using the IMAGE 2.2 Terrestrial Vegetation Model (Leemans and van den Born 1994; Alcamo et al. 1998), which in turn uses a modified version of the BIOME model (Prentice et al. 1992).

Development of the land cover datasets required that all land cover characterizations fit the biome type land cover classifications used in the NCAR LSM. This model allows for the representation of 22 unique land cover classifications originally derived from the Olson et al. (1983) half-degree global vegetation database (Bonan 1996). The IMAGE 2.2 datasets use 18 land cover classes that differ from those used in the NCAR LSM. The present-day IMAGE 2.2 derived land cover dataset was created by finding the dominant IMAGE 2.2 land cover class in a T42 grid cell and converting this to the equivalent NCAR LSM land cover class (Table 1; IMAlc, Fig. 1). To create the Hybrid dataset (HYBlc), the IMAGE 2.2 human land cover types were aggregated to the T42 grid size and then overlaid onto the NCAR LSM natural vegetation land cover types. The IMAGE 2.2 potential vegetation (POTlc) was created by aggregating the IMAGE 2.2 potential vegetation types to the T42 grid resolution and then converted to NCAR LSM land cover classes (Table 1; Fig. 1).
Table 1

Conversion criteria to translate IMAGE 2.2 land cover classes to NCAR LSM land cover classes

IMAGE 2.2 land cover class

NCAR LSM land cover class (Bonan 1996)

Notes

1. Natural vegetation classes

 Ice

Ice (1)

 

 Tundra

Tundra (19)

 

 Wooded tundra

Evergreen forest tundra (13)

 

 Boreal forest

Needleleaf evergreen tree (3)

Same as LSM class 7

 Cool coniferous forest

Needleleaf evergreen tree (3)

 

 Temperate mixed forest

Mixed net and bdt (6)

Same as LSM class 9

 Warm mixed forest

Mixed net and bdt (6)

Same as LSM class 9

 Temperate deciduous forest

Broadleaf deciduous tree (8)

 

 Grassland and Steppe

Cool grassland (17) if the original LSM land cover classes are 3, 4, 5, 6, 7, 13, 14, 15, 17, 19, 23, 24 (Bonan 1996, Table 5) or Warm grassland (18) otherwise

 

 Hot desert

Semi-desert (22)

 

 Scrubland

Deciduous shrubland (21) if the original LSM land cover classes are 21 or 12 (Bonan 1996, Table 5) or Evergreen shrubland (20) otherwise

 

 Savanna

Savanna (12)

 

 Tropical woodland

Tropical seasonal deciduous tree (11)

 

 Tropical forest

Tropical broadleaf evergreen tree (10)

 

2. Human land cover classes

 Agricultural land

Warm crop (26)

Same as LSM class 24

 Extensive grassland

Cool grassland (17) if the original LSM land cover classes are 3, 4, 5, 6, 7, 13, 14, 15, 17, 19, 23, 24 (Bonan 1996, Table 5) or Warm grassland (18) otherwise

 

 Carbon plantation (not used)

Not applicable

 

 Regrowth forest (abandoning)

Converted from the potential vegetation type given in IMAGE 2.2

 

 Regrowth forest (timber)

Converted from the potential vegetation type given in IMAGE 2.2

 
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0038-z/MediaObjects/382_2005_38_Fig1_HTML.gif
Fig. 1

Land cover representations used for the a LSMlc, b IMAlc, c HYBlc and d POTlc land cover experiments

Treatment for the Hybrid land cover dataset (HYBlc) was slightly different and combined the IMAGE 2.2 SRES human land cover classification data (agriculture and degraded grassland classes) together with the NCAR LSM natural vegetation land cover classifications. If the aggregated IMAGE 2.2 agricultural land cover class for a T42 grid cell was dominant, i.e., greater than 50% of the area, the cell was classified as agriculture. If the IMAGE 2.2 degraded grassland was dominant, then the T42 grid cell was assigned the NCAR LSM grassland land cover class. For those locations where a human land cover type was in the minority, and where the original NCAR LSM dataset had a human land cover type, the new natural vegetation type was determined by the dominant natural vegetation type of the surrounding grid cells and checked for reasonable accuracy against the Ramankutty and Foley (1999) potential vegetation class. When required, the conversion was based on the Ramankutty and Foley (1999) potential vegetation class (Table 2).
Table 2

Conversion criteria to translate Ramankutty and Foley (1999) potential vegetation land cover classes to NCAR LSM land cover classes

R&F potential land cover class

NCAR LSM land cover class

Tropical evergreen forest/woodland

Tropical broadleaf evergreen tree (10)

Tropical deciduous forest/woodland

Tropical broadleaf evergreen tree (10)

Temperate broadleaf evergreen forest/woodland

Broadleaf deciduous tree (8)

Temperate needleleaf evergreen forest/woodland

Needleleaf evergreen tree (3)

Temperate deciduous forest/woodland

Broadleaf deciduous tree (8)

Boreal evergreen forest/woodland

Needleleaf evergreen tree (3)

Boreal deciduous forest/woodland

Cool needleleaf deciduous tree (4)

Evergreen/deciduous mixed forest

Mixed net and bdt (6)

Savanna

Savanna (12)

Grassland/steppe

Cool grassland (17)

Dense shrubland

Evergreen shrubland (20)

Open shrubland

Evergreen shrubland (20)

Tundra

Tundra (19)

Desert

Desert (2)

Polar desert/rock/ice

Desert (2)

4 Results

Results are described in two sections. The first section compares the three different present-day land cover simulations using constant preindustrial atmospheric conditions. Analysis for this section focuses on the differences in the terrestrial energy balance between the IMAlc and HYBlc land cover simulations with respect to the standard LSMlc land cover. For illustrative purposes, specific regions demonstrating a particular process will be used to illustrate specific land cover change impacts. The LSMlc climatology with present-day atmospheric conditions has been described in detail by Bonan (1998), Washington et al. (2000) and Bonan et al. (2002), and tends to have a slight cold bias. The second section of the results analyzes the difference between the POTlc and the IMAlc simulations to identify some potential impacts of historical human land cover change on climate.

Two statistical procedures are performed to evaluate the significance of the results. First, a Student’s t test at the 0.05 level of significance is applied to comparable 40-year climatologies at each grid cell and shaded in the appropriate figures. Throughout this discussion, we will refer to statistically significant change in a region when the majority of grid cells in that region are statistically different at the 0.05 confidence level. However, because there is significant spatial and temporal correlation with all the variables evaluated in this study the standard statistical significance test of the two means can be biased. For this reason, we introduce a second non-parametric procedure to evaluate significance. This bootstrap derived methodology is used in the second part of the analysis where we evaluate the IMAlc and POTlc temperature fields.

The nonparametric significance test follows from the methodology described by Livezey and Chen (1983). The main difference that we take in our approach is to re-sample from the innovation fields from a time series model. This will help to adjust the test procedure for inter-annual temporal correlation as well as spatial correlation. The test is based on the usual test statistic for comparing the population means from two independent samples and is computed separately for every model grid cell. It has the form:
$$Z(x)=\left\vert\sqrt{{40}}\left({\bar X}_{t1}-{\bar X}_{t2} \over \sqrt{{S^{2}_{t1}} + {S^{2}_{t2}}}\right)\right\vert$$
where X indexes the grid cell locations. \({\bar X}_{t \rm {1}}\) and \({\bar X}_{\rm t2}\) are the sample means for the two cases and st1 and st2 are the sample standard deviations.

The primary statistical issue is how to assess the significance of these statistics in the presence of spatial and temporal correlation among the data, given that the test will be done at many grid cells. Our approach sets the critical value by resampling using a modification of the bootstrap. We consider the distribution of the maximum of Z(x) over all grid cells under the hypothesis that the two population means are equal. Although this sets the critical value substantially higher than for a single test (i.e., much higher than 1.96 for at the 5% level) our procedure is also a very conservative test, correcting for the fact that one would like to test for significance precisely at grid cells those test statistics are found to be large.

The critical value is determined from a distribution generated by Monte Carlo sampling. To carry out this simulation initially a first-order autoregressive model is fit to the standardized time series of the model output at each grid cell and for both cases. The residuals from these fits at a particular time are spatial fields and in this case, there are a total of (40–1)×2=78 fields. These fields form the basis for the bootstrap resampling. To simulate a synthetic data set, we sample with replacement from the residual fields and use the autoregressive relationship to generate time series at each grid cell. This results in a synthetic dataset that has the same sample length as the actual output. Because the residuals were standardized to have mean zero this sampling is done from a population that has a mean of zero and therefore, the synthetic data sets come from populations with the same mean. Of course, this is precisely the hypothesis that we wish to test. In addition, the spatial correlations among the residual fields and the temporal correlation from the autoregressive model induce a dependence among the grid cells that should be similar to the actual model output. This feature justifies the application of the critical value to test the actual model output.

Based on the synthetic data, the test statistic, Z(x) is found for all grid cells and then one computes M=max{Z(x)}. This process was repeated 2,000 times, accumulating a random sample of size 2,000 for the M statistic. The critical value at the 0.05 level used for testing is just the 95th percentile of the sample of M. Critical values for the particular variables are in the range of six to eight, substantially larger than 1.96. For the model output any grid cell with a statistic greater than this percentile is deemed statistically significant at the 95% level. This follows because in the case of identical population means the maximum value of the statistic only has a 5% probability of exceeding this value. The critical values are then contoured. We note that both tests are applied to the same Z scores values just the cutoff for significance has been altered.

4.1 Comparison of present-day land cover simulations

Statistically significant differences can be found in the energy balance simulations of the three present-day land cover experiments. Starting with surface albedo, net radiation is altered dramatically with substantial seasonal variation. In turn, net radiation change and the effectiveness of different vegetation types with respect to water uptake result in additional energy balance changes due to differences in latent and sensible heat fluxes. Each of these terms will be examined in some detail below, first comparing the IMAlc and LSMlc scenarios that have large differences in natural vegetation distributions and, second the HYBlc and LSMlc simulations that differ primarily in the extent of human land cover types.

4.1.1 IMAlc compared to LSMlc

The IMAlc differs from the LSMlc both with respect to the representation of natural vegetation types and with respect to the extent of human land cover (agriculture and grazed land; Fig. 1).

Albedo

Any alteration of surface vegetation and the exposure of bare ground will have an immediate effect on the albedo of the land surface. Albedo differences between IMAlc and LSMlc are pronounced, particularly in the Asian arctic region and southwestern Alaska where disagreements are based on the distribution of evergreen and deciduous needleleaf trees between the datasets (Fig. 1). In these areas, albedo values are much lower in the IMAlc simulations (Figs. 2, 3). These changes are most pronounced in DJF and MAM, when the evergreen needleleaf trees in IMA1c reduce the effects of snow cover on albedo. Other significant changes between natural vegetation representations in the IMAlc and LSMlc datasets include the characterization of the western/southwestern US, where the LSMlc has a mixture of desert and shrub land in the Great Basin region and evergreen needleleaf trees in the coastal sections, while the IMAlc dataset has a representation of grassland over much of the area; similar changes occur in parts of Africa, southern Australia and southern South America. Generally this leads to higher albedo values in the IMAlc simulation. In the tropics, the LSMlc shows significant albedo increases where agriculture replaces savanna/grassland in Nigeria, Ethiopia and southern Brazil. IMAlc shows albedo decreases where deciduous tropical forest replaces evergreen forest in the eastern Amazon and periphery of the Congo basin, and to a lesser degree where crops and savanna in southeast Asia and China replace LSMlc forests.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0038-z/MediaObjects/382_2005_38_Fig2_HTML.gif
Fig. 2

Change in surface albedo for the IMAlc land cover simulation minus the original LSMlc land cover simulation

https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0038-z/MediaObjects/382_2005_38_Fig3_HTML.gif
Fig. 3

Energy balance component changes for Siberia. Dominant land cover types are: LSMlc and HYBlc—deciduous forest tundra; IMAlc and POTlc—needleleaf evergreen trees/tundra. Statistics averaged for all the grid cells in the specified area

In this and the following simulation comparisons, there are some significant albedo changes over the ocean areas of the North Atlantic Ocean and around the Antarctic continent due to changes in sea-ice extent. These changes are probably due to ocean and sea-ice model variability, which is very high in these regions. Other ocean albedo changes are due to changes in cloud cover and the distribution of direct and diffuse radiation, which, along with solar zenith angle are used to derive ocean albedo. Even very small open ocean albedo changes tend to be statistically significant because of the low variability in the variable, although these changes have little impact on climate.

Net radiation

Net radiation changes are primarily due to the changes in surface albedo (Fig. 4). The effect of the large wintertime (DJF) differences in albedo in northern Asia is minimized with respect to net radiation because incident radiation is minimal at this time (Fig. 3). In contrast, similar springtime (MAM) differences in albedo result in net radiation differences of 25–50 W m−2. Similarly, relatively small differences in summertime or tropical albedo have a more pronounced influence on net radiation compared with areas of lower incident radiation. Areas affected include the Amazon, western US and southern Australia with net radiation differences of about 10–15 W m−2. The Amazon region (Fig. 5) is a good illustration of how different characterizations of natural vegetation lead to significant differences in climate. This region shows strong seasonal differences in net radiation (JJA/SON) where IMAGE 2.2 tropical deciduous forest replaces NCAR LSM evergreen tropical forest. In this case, LAI is changed throughout the year (not shown), having consequences for both albedo values and Bowen ratios. There are also significant increases in net radiation where the more extensive IMAlc crop areas in the Southern Hemisphere replace LSMlc natural vegetation types and decreases where IMAlc grasses replace desert areas in the Kalahari and Southern South America (Fig. 4).
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0038-z/MediaObjects/382_2005_38_Fig4_HTML.gif
Fig. 4

Change in net radiation for the IMAlc land cover simulation minus the original LSMlc land cover simulation

https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0038-z/MediaObjects/382_2005_38_Fig5_HTML.gif
Fig. 5

Energy balance component changes for the Amazon. Dominant land cover types are: LSMlc and HYBlc—tropical evergreen trees; IMAlc and POTlc—tropical deciduous trees. Statistics averaged for all the grid cells in the specified area

Latent heat flux

The disposition of energy into sensible and latent heat fluxes is governed in part by net radiation, but also by the availability of water to drive latent heat fluxes. Therefore, latent heat flux changes observed in these experiments are partially caused by changes in net radiation due to land cover change and partly by feedbacks in the system that alter circulation features and regional precipitation patterns. Latent heat fluxes are also dependent on the vegetation type and its efficiency with respect to transpiration (Bonan 1999). On an annual basis, most statistically significant changes are over the terrestrial surface (Fig. 6 – for brevity only the annual statistics are shown, temporal changes follow logically from the seasonal characteristics of the radiation variables; Tables 3, 4). Globally the IMAlc simulation shows that this change is accompanied by a change in the distribution of water sources that drive the latent heat flux. Probably due to a reduction in global LAI and forested cover, there is a significant reduction in transpiration (about 10% of the total) and a smaller reduction in canopy transpiration. These reduction are largely offset by an increase in ground evaporation (Table 4)
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0038-z/MediaObjects/382_2005_38_Fig6_HTML.gif
Fig. 6

Change in annual average climatologies for the IMAlc land cover simulation minus the original LSMlc land cover simulation: a latent heat flux; and b sensible heat flux

Table 3

Comparison of global mean annual and seasonal climate statistics for the LSMlc control; the difference between the IMAlc and LSMlc simulations; the difference between the HYBlc and LSMlc simulations; and the difference between the IMAlc and POTlc simulations

Variable

LSMlc

IMAlc–LSMlc

HYBlc–LSMlc

IMAlc–POTlc

Annual

JJA

DJF

Annual

JJA

DJF

Annual

JJA

DJF

Annual

JJA

DJF

Albedo

0.1263

0.1163

0.1337

−0.0015

−0.0016

0.0002

0.0014

0.0008

0.0018

0.0036

0.0034

0.0040

Net radiation (W m−2)

111.183

104.179

116.737

0.127

0.256

−0.041

−0.249

−0.185

−0.133

−0.662

−0.726

−0.344

Latent heat flux (W m−2)

89.258

90.056

91.179

0.035

0.029

0.022

−0.244

−0.319

−0.209

−0.599

−0.641

−0.56

Sensible heat flux (W m−2)

22.118

23.723

21.035

−0.024

−0.018

0.097

0.034

0.087

0.162

−0.048

−0.212

0.246

Precipitation (mm day−1)

3.06

3.084

3.105

0.001

0.002

0

−0.009

−0.011

−0.008

−0.021

−0.021

−0.019

Reference height temperature (K)

285.061

286.419

283.355

0.081

0.072

0.042

−0.129

−0.129

−0.144

−0.386

−0.379

−0.369

Table 4

Comparison of terrestrial mean annual climate statistics for the LSMlc control; the difference between the IMAlc and LSMlc simulations; the difference between the HYBlc and LSMlc simulations; and the difference between the IMAlc and POTlc simulations

Variable

LSMlc

IMAlc–LSMlc

HYBlc–LSMlc

IMAlc–POTlc

Incoming solar (W m−2)

197.767

−0.667

−0.248

−0.164

Absorbed solar (W m−2)

150.045

0.162

−0.848

−1.918

Net radiation (W m−2)

79.147

0.315

−0.651

−1.460

Latent heat (W m−2)

44.919

0.328

−0.520

−0.714

 Canopy evaporation (W m−2)

5.700

−0.327

−0.465

−0.571

 Ground evaporation (W m−2)

24.239

1.850

0.752

0.195

 Transpiration (W m−2)

12.460

−1.246

−0.875

−0.413

Sensible heat (W m−2)

33.776

0.069

−0.114

−0.744

Reference height temperature (K)

278.451

0.216

−0.179

−0.540

Precipitation (mm day−1)

2.075

0.033

−0.003

−0.031

The increased net radiation values for the IMAlc simulation in northern Russia and Siberia significantly increases latent heat fluxes in the region. However, the changes make up only small portion of the change in net radiation, because of the relatively dry summer conditions. The western US experiences increases in latent heat fluxes where the IMAlc simulation has more effectively transpiring vegetations types and shows an increase in precipitation. Where agriculture replaces shrub- and grassland in northern India and Pakistan latent heat fluxes also increase. There are also a number of locations where latent heat fluxes are reduced in the IMAlc simulation. In the Amazon, the IMAlc region of deciduous tropical forest shows a significant drop in latent heat flux primarily due to changes in LAI values and transpiration rates (Figs. 5, 6). Similarly, the differences between the LSMlc tropical evergreen forest and IMAlc savanna biome classifications in southeast Asia also result in much lower latent heat fluxes for the IMAlc simulation, especially in summer. There are also significant changes in heat flux over the Indian Ocean and in areas of the Pacific Ocean, suggesting that the land cover changes over the continent may affect the Monsoon circulation and subsequently the ITCZ location.

Sensible heat flux

In Siberia, where there is a significant increase in net radiation, but little water availability in summer for latent heat fluxes, most excess energy is dissipated through sensible heat fluxes (Figs. 6, 3). In other locations where latent heat fluxes are increased due to precipitation changes and changes in vegetation transpiration efficiencies, but where net radiation has not changed significantly, there are significant decreases in sensible heat offsetting latent heat flux gains (e.g., western US, India and southwestern Australia). In southeastern Australia decreases in net radiation are primarily absorbed by latent heat in MAM when there is more water available, but in SON values are actually greater than the difference in net radiation because of drying conditions and decreases in latent heat fluxes (Fig. 7). As with latent heat fluxes, on a global scale regional gains and losses in sensible heat fluxes are largely offsetting between these simulations.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0038-z/MediaObjects/382_2005_38_Fig7_HTML.gif
Fig. 7

Energy balance component changes for SE Australia. Dominant land cover types are: LSMlc—evergreen shrub; IMAlc and HYBlc—grassland; POTlc—desert/shrub. Statistics averaged for all the grid cells in the specified area

Temperature

Temperature changes follow predictably from the energy balance changes on the surface (Fig. 8). Compared to the LSMlc simulation, the large Siberian increase in net radiation and sensible heat fluxes in the IMAlc simulation result in significant warming year round, with a maximum in excess of 5 K occurring in spring. The IMAlc deciduous forest region in the eastern Amazon is also significantly warmer compared to the LSMlc simulation with tropical evergreen forest land cover (Fig. 5). Warming also occurs where vegetated areas are changed from forest to grasslands and where vegetated areas are replaced by desert areas in the IMAlc simulation (e.g., Mesopotania region). Where LSMlc grasslands are replaced by agriculture, and deserts with vegetated areas in IMAlc, the climate tends to be cooler (e.g., Australia, and western US). These conditions come about because there are more, or equal amounts of, moisture available for evapotranspiration so that reductions in net radiation translate to reduced sensible heat fluxes and cooler temperatures. Globally, the regional temperature changes tend to cancel one another, therefore, the IMAlc simulation is only slightly warmer compared to the LSMlc simulations by 0.081 K globally and 0.216 K over the terrestrial land surface (Tables 3, 4).
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0038-z/MediaObjects/382_2005_38_Fig8_HTML.gif
Fig. 8

Change in reference height temperature for the present-day IMAlc land cover simulation minus the original LSMlc land cover simulation

Planetary boundary layer

There is a strong relationship between the land cover changes and the height of the planetary boundary layer (PBL). These changes seem to integrate the overall results of the simulations very well. Compared to the LSMlc, the IMAlc simulation significantly increases the PBL over the warmer forest areas in Siberia (about 60 m) and the warmer deciduous Amazon (about 150 m; Fig. 9). Similarly, places where forest or mixed forest has been replaced by crops show significant decreases in the PBL heights. These effects could potentially influence local convective systems.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0038-z/MediaObjects/382_2005_38_Fig9_HTML.gif
Fig. 9

Change in annual average climatologies for the IMAlc land cover simulation minus the original LSMlc land cover simulation a planetary boundary layer height; b surface pressure; c precipitation; and d total runoff

Sea level pressure

The IMAlc simulation has significantly higher sea level pressure over the northern polar regions, Australia and from southern North America extending from the Atlantic into the tropical Pacific Oceans and Australia. In general, the mid-latitudes show intensified low pressure systems as does the Middle East and northeastern Africa (Fig. 9).

Precipitation and runoff

Precipitation changes are the result of both large scale circulation changes and changes in local water vapor fluxes. Differences in IMAlc and LSMlc simulated annual precipitation distributions are largest in the tropics (Fig. 9). IMAlc’s deciduous tropical forest land cover classification in the eastern Amazon results in statistically significant lower precipitation over the region, most likely due to local water vapor flux changes and reduced local convective activity. Just to the south and in many other locations, where IMAlc has agricultural land cover types compared to natural vegetation in LSMlc, precipitation rates are higher. Southeast Asia, north central Australia and southern Africa all show drying.

The other interesting feature in both these simulation comparisons is the apparent change in precipitation regime over the Indian Ocean, suggesting that the changes observed might be the result of interference with monsoon circulation and perhaps altered frequencies of the Indian Ocean dipole circulation (e.g., Saji et al. 1999, Webster et al. 1999). The tropical pacific and ENSO features also seem to be affected suggesting a displacement of the ITZC to the north.

The precipitation changes along with changing surface properties can lead to large and significant changes in local runoff quantities (Fig. 9). There is significantly more runoff in most of the southern hemisphere, with the exception of the region in South America where IMAlc has agriculture as opposed to the LSMlc grassland. These changes are closely connected to changes in local precipitation, but the importance of land cover to runoff generation is well illustrated in southeastern Australia, where less vegetated land cover types generate significantly more runoff compared to more vegetated land cover types.

4.1.2 HYBlc compared with LSMlc

The Hybrid scenario is designed to minimize the differences between natural vegetation types when compared to the LSMlc, while retaining the human land cover types, agriculture and grazing, derived from the IMAGE 2.2 datasets. Generally agricultural areas are more expansive in the Southern Hemisphere and around the peripheries of the extensive agricultural regions of North America and Eurasia. Furthermore, IMAlc grasslands replace many of the dryland natural vegetation types of the LSMlc (Fig. 1). For brevity only the annual differences are shown. Since these human land cover changes are also included in the IMAlc to LSMlc changes, temporal trends are shown in the previous figures.

Albedo

Changes are most pronounced in higher latitudes, especially where agricultural land cover types replace evergreen forests (Fig. 10). However, in the tropics the crop cycle, especially the bare fields after harvest, can result in significant albedo variability over the year. As shown in other studies (Bonan 1997, 1998; Bounoua et al. 2002), changes from forest to agriculture tend to increase albedo (e.g., from the Baltic states into Russia), while in areas where agriculture replaces non-wooded vegetation (e.g., North American grasslands) albedo tends to decrease. Where the IMAlc degraded grasslands replace NCAR LSM shrublands or deserts, albedo generally increases (e.g., Australia, northeastern Ethiopia and Sudan, and the western US).
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0038-z/MediaObjects/382_2005_38_Fig10_HTML.gif
Fig. 10

Change in annual average climatologies for the HYBlc land cover simulation minus the original LSMlc land cover simulation a albedo; b net radiation; c latent heat flux; and d sensible heat flux

Net radiation

Net radiation change due to differences in anthropogenic land cover types follow those expected from the albedo changes (Fig. 10) and like the previous comparison are most intense in the summer hemisphere and tropical regions. Regionally, the most significant changes are in Southern Australia where grasslands replace the dryland ecosystems in LSMlc, and in the eastern US where the LSM forest-crop vegetation is replaced with crop (reduction of forest). Smaller patches of grazing land in Africa and South America experience changes similar to those in Australia. Net radiation increases are observed where agriculture replaces grasslands in the US and southern Brazil and Ethiopia.

Latent heat flux

In places like the western US, latent heat flux is increased in the IMAlc simulation due to more efficiently transpiring plants, and because of slight increases in summer water availability (Fig. 10; see Bonan 1997, 1999 for more details). Similarly, while there are statistically significant changes in net radiation over most of Australia, this does not result in large changes in latent heat fluxes because evapotranspiration is generally limited by water supply, which does not change significantly between the simulations. Due to increases in precipitation India shows a significant increase in latent heat fluxes. There is also a significant change in the North Atlantic, which is caused by a difference in winter sea-ice extent.

Sensible heat flux

Sensible heat flux change is in large part dependent on water availability and the latent heat flux changes in these simulations. Australia provides a good example of the differential responses. There is large and significant reduction in net radiation over most of the continent, but the change in disposition of the energy varies by location (Fig. 10). Comparing the HYBlc to the LSMlc, the southwest shows a slight increase in precipitation leading to a small and barely significant increase in latent heat flux. Offsetting the gains in latent heat and allowing for the reduction in net radiation, sensible heat fluxes are reduced significantly by as much as 25 W m−2. In southeastern Australia the loss in net radiation results in a statistically insignificant change in latent heat flux and significant reduction in sensible heat flux. Finally, in central Australia, the loss in net radiation is accompanied by a larger and significant reduction in latent heat flux due to a statistically significant reduction in rainfall. To offset the large latent heat loss from the region, the sensible heat flux shows a statistically significant large increase (10–25 W m−2). Similar differential responses occur in the African Sahel and South America. Globally, the increased agricultural areas in the HYBlc reduce transpiration and increase ground evaporation, although to a lesser extent compared to the IMAlc simulation.

Temperature

The primary temperature response in the HYBlc simulation is a reduction in temperatures (Fig. 11). Statistically significant changes occur in the extended agricultural areas of Russia, southern Australia and portions of Africa and South America. These changes are not nearly as large as those in the IMAlc simulation, but they show a consistent cooling over most regions. Globally this simulation is cooler by 0.129 K and by 0.179 K on the terrestrial surface, when compared to the LSMlc simulation (Tables 3, 4).
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0038-z/MediaObjects/382_2005_38_Fig11_HTML.gif
Fig. 11

Change in annual average climatologies for the HYBlc land cover simulation minus the original LSMlc land cover simulation a reference height temperature; b surface pressure; c precipitation; and d total runoff

Sea level pressure

Sea level pressure changes between the POTlc and LSMlc simulations show significant increases over Australia, and south and eastern Asia, corresponding closely with regions experiencing losses in net radiation and regional cooling (Fig. 11). There are significant decreases in pressure in northeastern Canada and over the southern oceans. The pressure responses in the southern oceans are in large part a response to sea ice extent changes in the two simulations.

Precipitation and runoff

Precipitation changes in the HYBlc simulation, when compared to the IMAlc simulation are geographically very similar to the differences between the IMAlc and LSMlc simulations. However, the level of significance is much lower for the HYBlc simulation. The runoff difference patterns are also similar with decreased runoff in portions of the North America and southeast Asia. More runoff is simulated in southern Africa and southern Australia. Significant change is also shown for several low precipitation areas, such as the Sahara desert, in these locations the change is negligible from a climate perspective, but significance is high due to the low variability in runoff. The ocean areas also respond very much like the IMAlc simulation with an apparent shift in tropical circulation systems.

4.2 Comparison of prehuman and present-day land cover

Differences in land cover types between the prehuman (POT1c) and commensurate present-day land cover (IMA1c) datasets show that change to agriculture has primarily replaced forested ecosystems in the Northern Hemisphere and savanna ecosystems in the tropics (Fig. 12). There is also some conversion of mid-latitude grasslands to agriculture. Conversion to (degraded) grassland primarily occurs in dryland areas, particularly in Australia and from Arabia into central Asia (Fig. 12).
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0038-z/MediaObjects/382_2005_38_Fig12_HTML.gif
Fig. 12

Replaced land cover types due to human activities based on the IMAGE 2.2 present-day and potential vegetation land cover land cover datasets: a natural land cover types converted to agriculture; and b natural land cover types converted to grassland

4.2.1 Albedo

Annually, there are significant changes in albedo, in particular in the northern hemisphere agricultural regions (Europe, the US, and China, Fig. 13; and using Europe as an specific example, Fig. 14). In Europe, removing the leafout of the deciduous forest in the present-day simulation increases albedo in summer, thereby reducing net radiation (Fig. 14). In other areas, in winter and spring as illustrated in the previous experiments, albedos are increased under present-day conditions because of the replacement of forest with cropland. The highest magnitude changes occur in snowy locations where evergreen trees are replaced. A second significant pattern of albedo change occurs in dryland locations, in particular, desert areas that are replaced with grassland (degraded grassland class in IMAGE 2.2) tend to have a large reduction in albedo (e.g., Saudi Arabia, the central western US, and southeastern Australia; Fig. 7).

4.2.2 Solar radiation

As shown by the European example (Fig. 14), loss of net radiation is not only a function of albedo, but also due to a reduction in incident solar radiation. Incident radiation is also significantly reduced over other Northern Hemisphere agricultural areas and to a lesser extent over the Congo basin (not statistically significant) and the agricultural region of northern Argentina and Paraguay (Fig. 13). In the Sahel and Central Australia, there are significant and large increases in incident radiation. Changes in incident radiation cannot be explained by surface energy balance differences and suggest that other model feedbacks play a role in the observed climate changes.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0038-z/MediaObjects/382_2005_38_Fig13_HTML.gif
Fig. 13

Change in annual average climatologies for the present-day IMAlc land cover simulation minus the POTlc land cover simulation: a albedo; b incident radiation; c total cloud cover (DJF); and d total cloud cover (JJA)

https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0038-z/MediaObjects/382_2005_38_Fig14_HTML.gif
Fig. 14

Energy balance component changes for Europe: dominant land cover types are: LSMlc, IMAlc and HYBlc—agriculture; POTlc—deciduous broadleaf trees. Statistics averaged for all the grid cells in the specified area

Reviewing the changes in cloud cover, it is apparent that a significant portion of the net radiation change can be explained by the cloud feedbacks in the simulations (Fig. 13). In summer (JJA), both North America and Europe experience significant change in total cloud cover (>10%). This reduces incident solar radiation to these regions and hence net radiation (Fig. 15). Similar strong increases in cloud cover are observed in the Southern Hemisphere during the winter (JJA), especially over southwestern Africa and portions of South America. Clouds increase over the northern tropical Pacific Ocean, indicating a potential shift in the ITCZ over this region. In DJF increases in cloud cover are found in the southeastern US and southwestern Australia. There are also a number of regions that experience decreases in cloud cover in the present-day (IMAlc) simulations. Most significant is north central Australia, SE Asia, the Sahel, the tropical Pacific Ocean; and Southern Africa in summer (DJF).
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0038-z/MediaObjects/382_2005_38_Fig15_HTML.gif
Fig. 15

Change in annual average climatologies for the present-day IMAlc land cover simulation minus the POTlc land cover simulation: a net radiation; b latent heat flux; c sensible heat flux

Because of the changes in albedo and cloud cover, net radiation is significantly reduced on a global scale (Fig. 15; Tables 3, 4). However, regionally and seasonally these changes are highly variable. The spring and summer seasons show very significant reductions in net radiation in the mid latitudes (not shown, but similar to the IMAlc and LSMlc simulation comparisons). The areas most affected are the agricultural areas of eastern North America, western Europe, China and eastern Australia, as well as smaller areas in eastern South Africa and eastern South America. Dryland areas converted to agriculture or grassland, mostly in the tropics and subtropics, show marked increases in net radiation during the growing season. On average, net radiation is reduced by 0.662 W m−2 globally and by 1.46 W m−2 terrestrially.

4.2.3 Latent heat flux

Changes in latent heat flux are generally greatest in the mid-latitude spring and summer seasons due to reductions in net radiation (Fig. 15; seasonal plots not shown but the effects are similar to those in the IMAlc and LSMlc comparison). Declines are less significant in the summer and especially so in fall when many locations experience reduced soil moisture conditions, resulting in little change in evapotranspiration rates. However, there are some notable exceptions in this trend. For example, in northeastern Mexico and Texas, because of increases in local precipitation in spring and summer, latent heat flux is increased in late summer. In parts of southwestern Australia and in the western US, the increased transpiration efficiency of crops also increases latent heat fluxes in these locations, particularly in the summer when there is an accompanying increase in precipitation (see discussion on precipitation). Several other areas, particularly where agriculture has been introduced in China, southeastern portions of South America (northern Argentina) and Australia similarly have increased latent heat fluxes due to precipitation increases. North central Australia, experiences significant decreases in latent heat flux as a result of decreased rainfall over this region, accompanying the reduced cloud cover conditions in the present-day simulation. Some statistically significant changes are in locations where dryland surfaces have been vegetated to degraded grasslands (e.g., desert to grassland in Saudi Arabia). However, while statistically significant, these changes are very small in magnitude.

On the terrestrial surface, latent heat flux absorbs about half the loss in net radiation (−0.714 W m−2; Table 4). Globally this value is reduced to −0.599 W m−2, which comprises the majority of the global change in net radiation (−0.662 W m−2; Table 3). This indicates that there is a significant decrease in latent heat flux over the ocean surface as a whole although there are areas with significant increases in evaporation and latent heat flux, such as the western Indian Ocean, south of Australia and, to a lesser extent, off the west coast of North America. In addition, the decreased area of forest and increased agriculture, with a harvest cycle, reduced global LAI resulting in reduced transpiration and canopy evaporation (about 1 W m−2 total) only slightly offset by about 0.2 W m−2 increase in ground evaporation (Table 4).

4.2.4 Sensible heat flux

The overall magnitude of the sensible heat flux response is similar to the latent heat flux responses on land (−0.744 W m−2), while on a global scale this change is greatly reduced (−0.048 W m−2), and makes up a much smaller proportion of the net radiation loss (Fig. 15; Tables 3, 4). Statistically, these changes are of greater significance compared to the latent heat flux changes, especially over land. As with latent heat flux, in most locations the changes are negative and are especially evident during the crop-growing season where natural land cover has been replaced with agriculture. Most notable are the extensive reductions in sensible heat flux over the agricultural areas of eastern North America, Europe and China and South America. However in the tropics there are a number of less extensive, but statistically significant, increases in sensible heat flux. Examples are over the southeastern Amazon and southeast Asia where agriculture replaces tropical forest and in the desert fringes of Africa, and central Australia where grasslands replace desert and shrublands. Several ocean areas also demonstrate statistically significant sensible heat flux responses. All the ocean areas downwind from major mid latitude agricultural regions show increased sensible heat releases, especially in winter (i.e., downwind from the eastern US, China, and to a lesser extent South America and Australia).

4.2.5 Temperature

The annual temperature differences between the IMAlc and POTlc show that the present-day climate is cooler over most the globe (Fig. 16), with a global average temperature reduction of 0.386 and 0.54 K for the terrestrial surface (Tables 3, 4). This trend is statistically significant over most of the planet in the traditional difference of the mean test. Using the bootstrapping technique to eliminate temporal and spatial correlations, the significance regions are greatly reduced and are almost exclusively limited to terrestrial areas (contours show the 95% confidence level based on the bootstrap methodology). Regionally the greatest cooling coincides with areas that have been converted from natural forest vegetation to agriculture or grasslands in mid-latitudes. Within these areas, central and east Asia stand out as statistically most significant for both statistical tests. Europe also shows statistical significance for both tests, although not the entire area is statistically significant in the annual time frame. Other small patches of land use conversion are statistically significant by the bootstrap methodology even though they do not translate into cohesive large areas of change (e.g., east Australia coast, South America and southern Africa and Ethiopia). Numerically the largest cooling is observed over the Arctic, but this is not statistically significant by the bootstrap statistics.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0038-z/MediaObjects/382_2005_38_Fig16_HTML.gif
Fig. 16

Change in reference height temperatures for the present-day IMAlc land cover simulation minus the POTlc land cover simulation: a annual; b DJF; c JJA. Contours indicate the 95% confidence level based on a nonparametric statistical comparison

Although not statistically significant by the conservative bootstrap measure, there are also a few areas where there is a systematic warming on an annual scale and significant on the standard t test significance statistic. Terrestrially, these areas are all linked to land use conversion representing either tropical forest conversion to agriculture in the Amazon and Venezuela as well as an isolated grid cell over Sumatra. In addition, the central Australian continent is warmer due to reduced water resources for latent heat fluxes. Although not statistically significant by either test, areas around the Black Sea and western North America, converted from grassland to agriculture, show a slight warming in summer. While there are large changes in temperature of the polar ocean areas, these changes are generally not significant because of the high temperature variability, and are associated with changes in sea-ice concentrations of up to 15 percent (e.g., North Atlantic and south of Cape Horn).

There are marked seasonal differences in the magnitude and location of cooling. These trends are related to the type of land cover conversion and are strongly correlated with the summer season. In JJA (Fig. 16), there is a very strong cooling signal across most of the Northern Hemisphere mid-latitudes, with most statistically significant areas in the Northern Hemisphere. Especially strong are the cooling trends over the southern and southeastern US, the Mediterranean area and Mongolia. The areas of significant change shift to the Southern Hemisphere during DJF in the agricultural regions of South America, South Africa and southeastern Australia. Compared to the Northern Hemisphere, these areas are much smaller in extent due to the limited land area. The tropics show small patches of change, also seasonally differentiated especially in Africa, southeastern Asia and northern Australia. These changes are small because there is relatively little change in land cover in the tropics in these simulations. There is also a differential seasonal response in mid-latitude dry land areas, such as the western US and the steppes of Russia, where in the winter season temperatures actually warm slightly (insignificant statistically), in contrast to the cooling in the moister adjacent agricultural areas.

Much of the observed cooling is due to a significant reduction of the mean maximum daily temperature centered over the main agricultural regions of the globe (Fig. 17). In Europe and parts of North America, this reduction is offset by a slight, but insignificant, rise in daily minimum temperatures (Fig. 17), thus reducing the diurnal temperature range in these two areas. Increased temperatures in the tropical regions are primary the result of insignificant increases in both maximum and minimum temperatures. The diurnal range in temperature increases in a few dry desert areas over the Southern Sahara extending to the Arabian Peninsula and over central Australia. All these regions are affected by the ITCZ, and all experience a drying and heating. The extents of the tropical changes are very minor, and, overall, this experiment predicts cooling over all latitude zones (Fig. 18). Zonally, these changes range from decreases of 0.2 K at the equator to about 1.0 K in the Arctic.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0038-z/MediaObjects/382_2005_38_Fig17_HTML.gif
Fig. 17

Change in a average minimum and b maximum reference height temperature for the present-day IMAlc land cover simulation minus the POTlc land cover simulation

https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0038-z/MediaObjects/382_2005_38_Fig18_HTML.gif
Fig. 18

Longitudinal temperature difference profile comparing the IMAlc minus the POTlc simulations

The difference in the bootstrap and t test significance tests is very large in the example. We believe that the t test is very liberal—it will produce false positive results for roughly 5% of the grid cells. Moreover, due to the positive spatial correlation of the fields the significant cell grid will be spatially clustered. The bootstrap test based on the maximum Z score is quite conservative. For example, at the 0.05 level 95% of the time there will be no grid cells that are identified as significant. So even a single grid cell being deemed significant based on the maximum is unusual. Having approximately 5% of the bootstrap/maximum test grid cells significant in this example is a coincidence.

4.2.6 Planetary boundary layer

The reduced temperatures and a general reduction in surface roughness as forests are replaced by agriculture and shrub and semi desert vegetation with grasslands results in a general lowering of the boundary layer height (Fig. 19). The warming over the Sahel and central Australia lead to significant increases in the PBL in those areas.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0038-z/MediaObjects/382_2005_38_Fig19_HTML.gif
Fig. 19

Change in annual average climatologies for the present-day IMAlc land cover simulation minus the POTlc land cover simulation: a planetary boundary layer height, b surface pressure, c precipitation, and d total runoff

4.2.7 Planetary boundary layer and sea level pressure

Sea level pressure is significantly affected by land cover change in the extra-tropical regions of both hemispheres (Fig. 19). High pressure areas appear to be slightly displaced toward the areas of cooling over all the large scale agricultural areas in both hemispheres. There is a significant decrease in surface pressure, and there appears to be a shift in the storm tracks, from North America extending over the Atlantic to Great Britain and Scandinavia, and to a lesser extent over Siberia. These changes suggest an intensification, and perhaps a slight shift in the centers of action of the North Atlantic Oscillation. In addition, the high pressure system over the Arctic is also intensified, however this change is relatively small and statistically not significant. A similar pattern of pressure change is evident over the Southern Hemisphere, but it is not as extensive.

Other significant changes occur in south Asia where the entire coastal areas from Arabia to China experience significant increases in surface pressure suggesting a weakening of the Asian Monsoon circulation, primarily in summer. In the equatorial zones, the pressure changes are much smaller, and statistically most significant over the eastern Pacific and western Atlantic Oceans.

4.2.8 Precipitation and runoff

There are some statistically significant changes in daily precipitation over the Polar regions, and in areas of the tropical oceans (Fig. 19). In the polar areas, the colder temperatures may have a direct effect of the ability of the atmosphere to carry moisture to these regions, reducing precipitation. The tropical changes appear primarily along the northern limit of the ENSO region and along the fringes of the regions influenced by the Indian monsoon circulation suggesting that they are related to shifts in the ITCZ. Terrestrially, areas showing precipitation decreases are mostly in southern Asia and the fringes of the ITCZ regions in Africa and Australia, and over eastern North America. These decreases are offset by increases over the equatorial Congo basin and Indonesia, and portions of the western US and western Europe.

Over the oceans, both the tropical Pacific and tropical Indian Ocean show changes in precipitation rates, which appear to be related to shifts in the ITCZ. Similar precipitation changes are apparent over these Ocean areas in all the simulations performed for this study, perhaps indicating that the model has a high sensitivity in these areas, or that there is a consistent response from southeast Asian land cover changes (all simulations have differences in the region). Changes in total runoff closely follow those in precipitation, with some effects due to changes in land cover types and the fraction of bare ground (Fig. 19).

5 Discussion

5.1 Model sensitivity and impacts of different land cover characterizations

Comparison of the three present-day simulations shows that different land cover representations can lead to significantly different simulation results. This is consistent with a study by Oleson et al. (2004) and Myhre and Myhre (2003), which also found that different characterizations of land cover significantly alter the simulated climate. Differences in these simulations can be divided into two major components: discrepancies in the classification and extent of natural land cover types; and differences in the distribution and extent of human land cover types. Comparing the simulations in this project allows us to evaluate the impacts of both of these differences.

Comparing the IMAlc and LSMlc simulations shows the combined impacts of human and natural vegetation differences. Besides other locations, the Siberian arctic and Amazon regions illustrate that there are significant differences in simulated climates because of natural land cover classification discrepancies between datasets. In the Siberian region, albedo changes result in impacts on all aspects of climate, with a net result of the IMAlc simulation having a statistically significant higher average annual temperature compared to the LSMlc simulation. The change in latent heat flux is relatively small because in summer this environment is water limited. As a result, most of the net radiation increase is dissipated as sensible heat flux, in turn leading to the temperature increase. In the Amazon, changing land cover from evergreen tropical forest to deciduous tropical forest also results in a warmer climate in the IMAlc simulation. In this case, the warming is primarily due to changes in evapotranspiration rates. The reduced disposition of energy from the surface by latent heat flux results in a warmer surface and an increase in sensible heat flux values.

In most instances, when croplands replace forests this typically leads to increased albedos and cooling, while change from natural dryland vegetation types to agriculture result in warming. Significant cooling is the result in southwestern Australia where LSMlc mixed forest-agriculture and shrub vegetation are replaced by grasslands in the IMAlc simulation; a reversal of the more dominant case of IMAGE 2.2 agriculture replacing LSM natural vegetation.

The HYBlc simulation isolates only the difference in human land cover types between two representations of present-day land cover. Thus, only locations where there are discrepancies in human land cover types, either agriculture or grassland, create the different outcomes in the simulations. In this case, the much greater extent of agriculture in the IMAlc land cover dataset results in a significantly cooler simulation for the HYBlc simulation compared to the LSMlc control. Most of the discrepancies in the agricultural extent are in the tropics, Southern Hemisphere, and, to a lesser extent, increased agricultural areas in western Canada and China. The other large change is the difference in Australian land cover where LSMlc’s shrub vegetation is largely replaced by IMAGE 2.2′s degraded grassland vegetation. These latter classification differences are not that significant ecologically and are largely based on semantics, but with respect to the definitions of these land cover classes in the climate model they result in extensive differences in land cover albedo and potential evapotranspiration rates. The net result of this increase in agricultural areas is that there is a small cooling over most of the terrestrial surface.

Overall, the IMAGE 2.2 land cover database results in a warmer global climate simulation compared to the NCAR LSM land cover dataset. Most of this warming is the result of discrepancies in the distribution of natural vegetation types. However, as shown by the Hybrid land cover simulation, cooling due to the differences in agricultural extent largely offsets this warming. These biases should be taken into consideration when selecting a particular dataset for simulating land cover change in a GCM.

5.2 Historical climate impacts of human induced land cover change

5.2.1 Global impacts

Historical human land cover change is simulated to have a 0.39 K cooling effect on global temperatures and 0.54 K over the terrestrial surface. There are several notable characteristics associated with these changes. The reduction in global average net radiation is about 0.66 W m−2, latent heat flux decreases by 0.60 W m−2 and sensible heat flux is reduced by 0.05 W m−2. These changes are much larger over the terrestrial surface, with a reduction of 1.46 W m−2 in net radiation, which is more evenly partitioned between latent and sensible heat flux reductions (0.74 W m−2 and 0.71 W m−2 respectively). These values are approximately double the maximum observed range between the different present-day land cover simulations, which in this context can be thought of as an uncertainty measure.

There are also significant differences in the seasonality of the prehuman and present-day simulations. The boreal summer (JJA) sees a much greater portion of the net radiation reduction (0.726 W m−2), leading to a 0.64 W m−2 reduction in latent heat fluxes and a 0.21 W m−2 reduction in sensible heat fluxes globally. In comparison, the boreal winter (DJF) sees a 0.34 W m−2 net radiation reduction, which, combined with a drying, creates a 0.56 W m−2 reduction in latent heat flux and a 0.246 W m−2 increase in sensible heat flux globally.

Our simulated −0.39 K globally temperature change can be compared to a number of previously reported global values. In a study with essentially the same atmospheric model, but with slab oceans, Govindasamy et al. (2001) reported a −0.25 K global temperature change. Using the HadAM3 with prescribed Ocean conditions Betts observed a −0.25 K temperature change, and Hansen et al. (1998) reported −0.14 K for the GISS model. Results from intermediate complexity models found global temperature changes of −0.35 K for the biogeophysical experiment by Brovkin et al. (1999), and −0.13 K by Matthews et al. (2003). Zonally, we find a minimum −0.2 K change over the tropics to a maximum −1.0 K over the arctic. This compares to reported zonal cooling from 0.09 K in the tropics to 0.22 K in the mid latitudes of the Northern Hemisphere by Matthews et al., (2003); and a warming of 0.8 K in the tropics and cooling of 0.7–1.1 K in the Northern Hemisphere mid latitudes reported by Bounoua et al. (2002). In all cases there is agreement on an overall cooling of global temperatures. Reported results differ in part because they use different representations of land cover and because of differences in the GCMs and the experimental set-ups. These results suggest that the NCAR LSM may be more sensitive to land cover change in comparison to other models, while the PCM is known to have a low sensitivity to greenhouse gas effects. Furthermore, our findings of a greater cooling in this fully coupled simulation, compared to for example Govindasamy et al. (2001), suggest that there are potentially significant ocean-atmosphere feedbacks that accompany the historical impacts of land cover change.

5.2.2 Regional impacts

Compared to the global statistics, regional changes between the simulations are much larger, up to 2 K, especially over areas converted to agriculture. In this context, the global statistics can be deceptive because they ignore the regional impacts of land cover change effects, which are often offsetting on a global scale. Notable in these simulations is that all the major agricultural regions of the globe (eastern US, Europe, China, mid latitudes of South America, South Africa and southeastern Australia) show significant cooling. These mid-latitude cooling trends are strongly associated with the summer season, but persist in winter. Dryland regions converted to agriculture have a slightly different response to converted wetter areas, in part because the limited water supply cannot be used to compensate for albedo induced net radiation change through latent heat flux changes.

Other regions that are significantly affected in this simulation are southeast and eastern Asia, which show a very strong cooling year round. In addition, precipitation is also decreased in most of these areas, although this trend is not as statistically significant. Also affected are the dryland areas of Africa and Australia, where the desert margins, largely controlled by ITCZ rainfall regimes, show a considerable drying and warming. Although these signals are consistent, they are generally not statistically significant. There are also very slight, insignificant, increases in Tropical precipitation over Africa and Indonesia.

5.2.3 Mechanisms of climate change due to land cover

The differences in the historical and present-day climate simulations show a distinct pattern of change. In general, the mid-latitude cooling over the agricultural regions leads to strong cooling that is initiated through changes in the surface energy balances in the region. Conversion to agriculture from forest in particular leads to significant albedo change and reductions in net radiation. However, there is also an apparent positive feedback that leads to additional cloud cover and further cooling, especially evident over Europe and eastern North America. As the pressure difference map indicates, there is an eastward shift and intensification of the zonal pressure patterns (Fig. 19). This preliminary analysis suggests there is a significant increase in zonal flow (not shown) and a shifting and perhaps intensification of regional storm tracks.

The other large scale change is in southern and eastern Asia. These areas show some of the most significant cooling all along the continental fringe. This cooling appears to have a direct impact on the monsoon circulation. Especially in the summer, the cooling of the land surface weakens the monsoon circulation leading to less onshore flow, reducing precipitation over the region. In turn the monsoon flows alter the location, and appears to reduce, the migration extent of the ITCZ, especially over eastern Africa and Australia. Both regions experience decreased cloud cover and drying.

While this study cannot specifically identify the timing of these regional climate changes, they should follow periods of land cover conversion. For Europe, this would be the period from 1,200 to about 1,700 (Goudie 1990), potentially linking land use change to little ice age conditions as proposed by Govindasamy et al. (2001). In North America the timing of deforestation is primarily from the mid 1700s to the early 1900s (Goudie 1990). Other regions would have experienced such change more recently, possibly influencing the 20th century climate record. In this study, the observed cooling is primarily through a reduction in daily maximum temperatures. Karl et al. (1993) showed that in the US there was a general rising trend in temperatures, with most warming coming from rising minimum temperatures. This observed trend is compatible with the results from this study if a land cover cooling occurred at the same time as a warming trend from other forcings (e.g., GHGs; see Bonan 2001). The greater warming effects cancel the cooling trend of the daily maximum temperatures while the minimal change in daily minimum temperatures due to land cover change does not affect the overall warming trend. Observations reported by the IPCC (2001) indicate that the daily temperature range is decreasing in many other regions of the Earth, especially southeast Asia as is simulated in this study.

6 Conclusions

Human induced land cover change is potentially an important factor in regional and global climate change, and studies of past human land cover change could improve our understanding of historical climate change. This study and most historical land cover change studies show that past land cover change has led to global cooling. The mechanisms for cooling are twofold. First there is a direct cooling effect due to albedo changes in mid-latitudes. Second, there is an indirect effect cooling due to feedbacks in mid-latitude and Asian monsoon circulation systems that result in increased cloud cover conditions. However, future land cover scenarios will not necessarily result in additional cooling because additional agricultural lands in the future would begin to replace more dryland areas and tropical forests, as compared to replacing mid-latitude forest areas in the past. It has been shown by DeFries et al. (2002) that the 2050 IMAGE land cover projections could result in warming. This may hold true for this model, but because of the small areas affected in this study it cannot be shown statistically. This signal switch between past and future effects of land cover change could be an important contributor to recent observed temperature trends.

We conclude with the answers to the questions posed at the outset of this paper:
  1. 1.

    What is the sensitivity of a fully coupled general circulation model to different present-day land cover representations?

     

The DOE-PCM shows a significant sensitivity to different representations of present-day land cover change with simulation outcomes ranging by 0.21 K in global mean annual surface air temperature. Furthermore, historical land cover change could have cooled global average temperatures by about 0.39 K and, perhaps of greater importance, regionally up to 2 K. Therefore, it is clear that land cover representation is an important consideration in GCM simulations, especially on the regional level. In addition, in order to compare land cover change studies across GCMs it is important that a consistent land cover change treatment be used.

  1. 2.

    What is the best method for including alternative land cover scenarios into GCM simulations?

     

If land cover simulations are to be used in climate change scenario simulations, the hybrid methodology used in this paper, and also used by De Fries et al. (2002), would be the most appropriate methodology for incorporating land cover change into different GCMs. Most models have been calibrated based on their native land cover classification schemes (LSMlc in this case). This methodology allows all models to use similar representation of anthropogenic land cover change, while minimizing differences between the natural vegetation classes of comparable alternative model simulations.

  1. 3.

    Within the context of the uncertainty associated with representing present-day land cover, could historical human land cover change have had a significant impact on climate? Is this signal sufficiently large to suggest the need for including land cover change as part of IPCC based future climate scenarios?

     

In this paper, the change in global temperature from prehuman vegetation conditions to present-day conditions is roughly twice the magnitude (−0.39 K) of that resulting from contrasts in present-day simulations (0.21 K). This is about one third the projected 20th century warming due to GHG, aerosol, volcanic and solar forcing in the DOE-PCM (Meehl et al. 2003, 2004). More important, however, is that the largest impacts of land cover change (~2 K) are observed on regional scales. In these areas changes due to land cover conversion are comparable to those due to other simulated forcings (Meehl et al. 2003, 2004). In many cases these significant regional changes also coincide with regions of highest human population and potential climate impacts on society. This suggests that land cover change does have a significant role in producing global and regional climate changes that could impact large sections of the human population. Therefore, land cover change should be used in historical and future SRES climate scenario simulations.

  1. 4.

    Are global statistics adequate to detect climate signals from land cover change? Is there a discernable spatial pattern associated with land cover induced climate change?

     

Global or terrestrial average climate statistics do not give a good representation of the effects of land cover change on climate. The effects of land cover change are highly regional and are largely offsetting on a global scale. This work supports the hypothesis from previous work that conversion of forested land to agriculture in the mid latitudes leads to significant cooling over these regions and a reduction in diurnal temperature ranges. In addition, feedbacks that increase cloud cover and reductions in solar radiation amplify this cooling trend. There is some evidence that tropical land cover conversion may lead to warming signals, but the small areas involved in this experiment provide insufficient evidence. This experiment also suggests that land cover change in Asia may have a significant impact on the Asian Monsoon circulation and in turn the extent of the ITCZ over the Indian and Pacific Oceans.

Acknowledgements

Special thanks to Rik Leemans for providing the SRES data, and Tom Bettge and Lawrence Buja for their assistance in running the experiments, and two anonymous reviewers for their comments and suggestions. This research was supported by the Office of Science (BER), US Department of Energy, Cooperative Agreement No. DE-FC02-97ER62402, the National Science Foundation grant numbers ATM-0107404, and ATM-0413540, the NCAR Weather and Climate Impact Assessment Science Initiative funded by NSF, and the University of Kansas, Center for Research.

Copyright information

© Springer-Verlag 2005