1. Introduction
The land surface is a dynamic and complex component of the global climate system. It acts as a lower boundary for approximately 30% of the atmosphere, exchanging with it moisture, momentum, and heat. In comparison to the oceans, it has much less storage of thermal energy and negligible horizontal transport. Nevertheless, the land surface is more variable in terms of many important coupling processes. The land surface, when wet, can exchange water with the atmosphere rapidly because of its high surface roughness; but when dry, it provides little water to the atmosphere. Because of the relatively low heat capacity of the land surface, local thermal conditions are much more responsive to net radiation. The land surface is also characterized by its heterogeneity with the distribution of soil properties and vegetation cover being almost fractal in nature (Dickinson 1992). The land surface has many features (soil, vegetation, topography), many characteristics (soil porosity, vegetation greenness, roughness, etc.), and the ability to store water in slow and fast reservoirs (soil moisture, storage in plants, snow, groundwater, ice, rivers, lakes, and reservoirs). All of these factors interact with the atmospheric circulation, temperature, humidity, and precipitation to play important roles in the variability and predictability of the climate.
Over the past 20 yr there has been considerable interest in the role of the land surface in contributing to climate variability. General circulation models (GCMs) have been used to explore the land surface’s role in climate. Examples include land evaporation feedbacks (e.g., Shukla and Mintz 1982; Sud and Molod 1988; Rowell and Blondin 1990), desertification in the Sahel region of Africa (e.g., Charney 1975; Xue and Shukla 1993; Xue 1997; Nicholson et al. 1998), tropical deforestation (e.g., Dickinson and Henderson-Sellers 1988; Shukla et al. 1990; Henderson-Sellers et al. 1993; Dirmeyer and Shukla 1994; Lean and Rowntree 1997), and the role of the Maritime Continent (e.g., Polcher and Laval 1994; Delire et al. 2001). Some studies of land–climate interactions are focused on the feedback between land and atmosphere, identifying sensitivities and modes that are potentially predictable. For example, the initial state of the land surface has been found to have a strong control on the subsequent evolution of the climate (Fennessy and Shukla 1999; Koster et al. 2004; Dirmeyer 2005). These modeling studies demonstrate the existence of potential impacts on climate by land surface properties such as surface albedo, roughness length, and soil moisture. Much progress has been made in understanding how these properties influence climate by modifying surface budgets of energy, water, and momentum. However, there exist other land characteristics whose importance and impacts on climate are still inadequately understood. Key vegetation properties are among them, including leaf area index (LAI), the fraction of photosynthetically active radiation absorbed by the vegetation canopy (fPAR), and greenness fraction of the total plant matter.
Vegetation has a significant control over the exchange processes between land and atmosphere. A primary parameter to characterize vegetation canopy is LAI. It quantifies the overall density of the vegetation canopy and determines the absorption of solar radiation by vegetation, the amount of precipitation intercepted by vegetation, the surface area for transpiration and sensible heat exchange, and the emissivity of the canopy. Over large areas the vegetation cover fraction (the percentage of the soil obscured by vegetation when viewed from the zenith) describes heterogeneities in terms of relative fractions of different surfaces and their spatial scales. Greenness fraction is a simplified one-parameter metric to describe the spectral shortwave absorption properties of the vegetation. Within the context of land surface parameterizations (LSPs) for GCMs, such as the simplified Simple Biosphere model (SSiB; Xue et al. 1991), LAI is usually used to derive many important intermediate variables, while vegetation cover fraction, derived from fPAR and greenness fraction, provides an integrated quantification of land surface fluxes and state variables over a heterogeneous model grid. In other words, three key vegetation properties (LAI, fPAR, and greenness), as initial input variables to LSPs, are directly or indirectly involved in all major exchange processes—radiative transfer (albedo), turbulent transfer (roughness), and heat/water vapor exchange (stomatal resistance) in land surface–atmosphere interactions. Therefore, it is important to assess the direct impacts of these vegetation properties, as any of their changes will modulate fluxes of moisture, heat, and momentum, which in turn largely determine the climate and its variability.
Most of the previous sensitivity studies concerning the impacts of surface albedo, roughness length, or soil wetness are generally designed in an idealized framework, assuming a fixed change in a particular surface property of interest. Therefore the magnitude of the change is somewhat arbitrary and sometimes extreme. In addition, the relationship among the various interactive land surface parameters or the relation between the synergistic changes in all the relevant parameters is usually neglected in these studies. For instance, a decrease of greenness fraction due to senescence or drought is accompanied necessarily by decreases in fPAR and LAI. Several recent studies have attempted to investigate the sensitivity of general circulation dynamics to “quasi-natural” changes (in contrast to arbitrary changes) in the vegetation properties. Pitman et al. (1999) investigated the sensitivity of land surface simulation of evaporation and soil wetness to different ways of specifying spatial and temporal LAI variability within the Global Soil Wetness Project (GSWP) framework. Chase et al. (1996) compared climate simulations using a distribution of maximum LAI derived from satellite data as a control and a distribution of potential LAI assumed to be in equilibrium with current climate hydrology. They also accounted for the simultaneous variation of vegetation fraction with different LAI scenarios. Bounoua et al. (2000) examined the sensitivity of climate to changes in vegetation density—exemplified by the difference between the maximum and minimum normalized difference vegetation index (NDVI) value observed over the 1982–90 period. Although NDVI data for the two scenarios were processed to produce the corresponding fields of fPAR, LAI, greenness fraction, snow-free albedo, and roughness for use as boundary conditions, such scenarios were more extreme than the interannual variability actually observed during the 9-yr period. Hales et al. (2004) studied the sensitivity of tropical land climate to LAI through surface conductance and albedo, with magnitudes of percent change in LAI estimated from 9-yr interannual variations of rainy season LAI for different vegetation types.
In this study, we attempt to examine the sensitivity of land surface climate to observed changes in combinations of key vegetation properties (LAI, fPAR, and greenness fraction), derived from multiyear satellite observations. The observed changes in the magnitude of vegetation properties and their spatial distribution result from two different treatments of these properties when specified in the LSPs as boundary conditions, with one based on natural interannual variability of vegetation properties and the other using their multiyear mean climatology (the approach typically employed by most climate models). Our experience with uncoupled land surface modeling (without atmosphere feedback) suggests that there exist sensitivities in simulations of surface fluxes and state variables to the choice of mean seasonal cycle versus time-varying vegetation properties (Guo et al. 2006). Discovery of these sensitivities helps to motivate the coupled studies and raises several questions: How do such sensitivities present themselves in the coupled climate system? What role does land–atmosphere feedback play in the response of the land surface–vegetation phenology variability? Is the impact of interannual variations of vegetation on a global, regional, or only a local scale? Is there skill to be gained in climate forecasts from including observed vegetation anomalies?
The structure of the paper is as follows. In section 2, we briefly describe the model and experiment design. The results from the coupled GCM simulations are discussed in section 3 with reference to parallel offline model experiments when relevant. This includes an assessment of model skill with and without interannually varying vegetation. In section 4, the ability of the model with different specifications of vegetation parameters to reproduce observed regional climate anomalies is assessed. Conclusions are summarized in section 5.
2. Models and setup
The global atmosphere and land surface models are briefly described in this section. Also, the production and application of the essential datasets are reviewed and the framework of the numerical integrations is detailed.
a. Models
A new atmospheric component of the Center for Ocean–Land–Atmosphere Studies (COLA) climate model (version 3) recently has been developed, which is based on the previous version of the COLA atmospheric GCM. The V2.2 was a research version of the global spectral model described by Sela (1980) and is similar to that described in detail by Kinter et al. (1997). The V3.2 is the current version comprising the National Center for Atmospheric Research (NCAR) Community Climate Model version 3.6.6 (CCM3.6.6) dynamical core (Kiehl et al. 1998), but it integrates a set of revised subgrid-scale physical parameterizations compared to V2.2. These include the longwave radiation parameterization of Collins et al. (2002) replacing that of Harshvardhan et al. (1987), the relaxed Arakawa–Schubert scheme of Bacmeister et al. (2000) substituting for that of Moorthi and Suarez (1992) for convective precipitation, and the nonlocal scheme of Hong and Pan (1996) in place of the local-K theory (Mellor and Yamada 1982) for the planetary boundary layer (PBL) parameterization. The shortwave radiation parameterization (Briegleb 1992) and shallow convection scheme (Tiedtke 1984) are essentially the same between two versions. The V3.2 is run at a spectral resolution of T62 (1.875° longitude and latitude on the corresponding Gaussian grid) and 28 vertical levels, the same resolution as the National Centers for Environmental Prediction (NCEP) reanalysis models (Kalnay et al. 1996; Kanamitsu et al. 2002) and with identical topography. The V2.2 was run at a spectral resolution of T63 and 18 vertical levels. Such modifications were made to ease the execution of seasonal hindcasts with V3.2 AGCM where initial conditions now can be taken directly from the NCEP reanalysis without the need for interpolation.
The LSP coupled to the AGCM is an updated version of the SSiB model described by Dirmeyer and Zeng (1999). SSiB is a complex land surface scheme (LSS) that represents the canopy explicitly. SSiB has been very extensively tested, evaluated, and used in a wide range of studies. The principal difference in V3.2 is that the root zone layer (one layer in V2.2) is further divided into four sublayers, which, in combination with one surface layer and one deep recharge zone, results in six soil layers. The four root layers are partitioned to contain an equal mass of roots, based on the profiles of Zeng (2001). Such treatment provides a more realistic simulation of the vertical profiles of soil moisture, soil temperature, and water movement between soil layers. As a result, the decrease of transpiration with soil moisture stress is better characterized, preventing the so-called “stomatal suicide” that abruptly shuts down transpiration in previous versions of SSiB.
In SSiB, vegetation parameters come from two sets of characteristics. The one that is time invariant is the vegetation map. There exist 11 types of vegetation cover plus bare soil and permanent ice. Each type of vegetation has a set of parameters that define properties such as the rooting depth, height of the vegetation, critical temperatures for stomatal stress, and other radiative and hydrologic characteristics. There is also a set of vegetation properties (LAI, fPAR, and greenness) that vary with the seasons. Monthly data are used and values are linearly interpolated between the midpoints of each month. The combination of vegetation type and properties, along with an independent map of soil attributes, determines the exact values of the parameters used in the equations that calculate radiative exchanges, moisture fluxes, and thermal properties of the land surface.
Other new features in V3.2 include a uniform calculation of saturation vapor pressure throughout all subroutines of the GCM (Marx 2002), a variation of the latent heat of phase change with temperature (Bohren and Albrecht 1998), the reduction of the fourth-order horizontal diffusion coefficient by two orders of magnitude, and the introduction of the second-order horizontal diffusion in the top two layers of the atmosphere to achieve numerical stability. For a more detailed description of the V3.2 AGCM, please refer to Misra et al. (2006).
b. Datasets and experiment design
The same LSP described above was used for COLA’s participation in the Second Global Soil Wetness Project (GSWP-2; Dirmeyer et al. 2006b). The GSWP approach is to integrate one-way uncoupled LSSs driven by externally specified near-surface meteorology. This provides the land models with some of the most accurate forcing data available, but deprives the system of the feedbacks of the land surface on the atmosphere. GSWP-2 is closely linked to the International Satellite Land Surface Climatology Project (ISLSCP) Initiative-II data effort (Hall et al. 2006), and the LSS simulations in GSWP-2 encompass the same core 10-yr period as the ISLSCP Initiative II (1986–95). The model simulations for GSWP-2 are conducted globally over land (excluding Antarctica) on a regular 1° × 1° grid. The 3-hourly near-surface meteorological forcing datasets are derived from the regridding of the NCEP/Department of Energy (DOE) reanalyses (Kanamitsu et al. 2002) for ISLSCP Initiative II, with corrections to the systematic biases in the reanalysis fields made by hybridization of the 3-hourly analysis with global observationally based gridded datasets at lower temporal resolution (Zhao and Dirmeyer 2003). The bulk of the GSWP-2 output data are reported at a daily interval, which have been averaged to monthly for this study.
For the COLA GCM–LSP coupled integrations, ensembles of 10 members are produced for the same 10-yr period as GSWP-2. We choose a decadal integration instead of a series of seasonal predictions to isolate the role of the vegetation boundary conditions in the absence of effects from initial conditions. The 10 ensemble members are initialized from the 0000 UTC NCEP global reanalyses (Kalnay et al. 1996) on the last 10 days of December of 1985. Over ocean, boundary conditions of sea surface temperature (SST) are specified from the weekly analysis of Reynolds and Smith (1994). Monthly output data on the Gaussian grid corresponding to the T62 spectral resolution were used for this study.
The spatially varying fields of soil properties (texture, slope, albedo, and depth) and vegetation distribution, as well as monthly varying vegetation properties (LAI, fPAR, and greenness) come directly from the ISLSCP Initiative-II dataset at a resolution of 1° × 1°. Please refer to Hall et al. (2006) for more information about these biophysical vegetation properties and their derivations from the Fourier-adjusted, sensor- and solar zenith angle–corrected, interpolated, and reconstructed (FASIR) NDVI monthly time series. These fields are spatially interpolated to the GCM grid prior to serving as boundary conditions to the COLA GCM.
We notice that there appear widespread strong positive anomalies in all the vegetation properties over the large portion of the globe from July 1993 to August 1994. This is quite suspicious and likely unrealistic, but the reason behind this feature is not known. In addition, there may exist errors in data from July to October of 1991 as a result of insufficient correction for volcanic aerosols. Hall et al. (2006) pointed out that FASIR NDVI is neither completely atmospherically corrected (no explicit corrections for water vapor or tropospheric aerosols) nor calibrated for use in an absolute sense for carbon, water, and energy or climate analyses. Despite the potential errors described above, we can still assess the sensitivities of land and atmosphere to the perturbations in magnitudes of vegetation properties as well as the geographical dependences of the sensitivities. In such a context, our analyses and interpretations of the model results should still hold valid, although they may not reflect real conditions during portions of the 10-yr period.
The same pair of experiments is conducted within the GSWP-2 framework in an uncoupled mode and with the coupled COLA climate model to examine the impacts of interannual variations of vegetation properties (LAI, fPAR, and greenness). The mean annual cycle integrations (MAC) involve the use of 10-yr mean climatology of vegetation properties throughout the simulation, while for the integrations with the interannual cycle (IAC), year-to-year phenology variability in vegetation properties are specified instead. The results are presented as the difference between IAC and MAC simulations. We also attempt to validate whether the GCM hindcasts are improved with interannually varying vegetation, although the problems in the vegetation dataset described above compromise this effort somewhat.
3. Coupled sensitivity studies
We expect that a positive anomaly in vegetation properties should typically result in lower surface albedo, greater absorption of shortwave radiation, and an increase in evapotranspiration (contributed by increased transpiration and canopy interception partially offset by decreased evaporation from the soil). This should lead to a higher equivalent potential temperature near the surface (more total energy from thermal and latent heat content), which should be conducive to increased rainfall. Likewise, we expect the opposite effects during periods of negative anomalies. We draw these hypotheses from lessons learned from many modeling studies on the effects of land-cover change conducted over the last 20 yr. However, we expect that the strength of the response will depend on many factors, including the background state of the atmosphere and the relative importance of the general circulation and remote forcings in determining the local climate.
a. Mean fluxes
Figure 1 shows the 10-yr mean differences between the IAC and MAC cases of energy fluxes and cloud cover in the coupled model, averaged across all 10 ensemble members. In most regions the differences in the specification of vegetation properties has little effect on surface latent heat flux (Fig. 1d), but areas of pronounced decrease do exist over southeastern China, the subtropical fringes of South America and Africa, the Nordeste, and eastern Australia. There are a few small regions of increased terrestrial latent heat flux, most prominently over the northern Amazon, the western and southern U.S. Great Plains, Mexico, western Asia, and northern China. The change in latent heat flux over these regions appears to have a local connection to increased moisture, as indicated by the change in cloud cover (Fig. 1e). The 10-yr mean change in sensible heat flux (Fig. 1c) strongly mirrors that of latent heat flux. This suggests that much of the impact of vegetation variations is manifested in the partitioning of available surface energy between latent and sensible heat fluxes. Figures 1a,b show the impact on the net longwave and net shortwave radiation (downward minus upward), which mimics the latent and sensible heat flux changes, respectively. The equal but opposite changes between net shortwave and longwave suggest the underlying cloud changes as well. Table 1 summarizes the spatial correlations among pairs of the 10-yr mean differences between the IAC and MAC cases of these fluxes and cloud cover. As we expect, there exist strong anticorrelations between net shortwave and net longwave radiation as well as between sensible and latent heat fluxes (r = −0.82), whereas net longwave radiation and cloud cover is strongly positively correlated (r = 0.81). The correlations among other pairs also seem to be consistent with what we observed in Fig. 1. For instance, net shortwave and longwave radiation are positively correlated with the sensible and latent heat flux, respectively, and net shortwave radiation is negatively correlated with cloud cover. Note that all the changes in time-mean surface fluxes are small with respect to typical mean values.
Changes in surface fluxes can occur directly as a result of the changing surface properties, and not necessarily through a feedback with the atmosphere. This is evident in the Southern Hemisphere in Fig. 1. Here, the increases in latent heating over southern Africa and South America are not accompanied by increases in cloud cover. Other mechanisms may be in effect. To understand what may be happening, we must take a foray into the uncoupled versions of the LSP and examine the separate components that contribute to the latent heat flux.
Offline uncoupled simulations show that differences in vegetation transpiration appear to correspond spatially to those in soil evaporation but with opposite sign. The spatial correlation between the 10-yr mean differences of vegetation transpiration and soil evaporation is strong with an r value of −0.80. The same feature of spatial correspondence between vegetation transpiration and soil evaporation can also be found in individual GSWP-2 models such as the National Aeronautics and Space Administration (NASA) Seasonal-to-Interannual Prediction Project (NSIPP; −0.74) catchment model and the Soil–Water–Atmosphere–Plants (SWAP; −0.91) model. These two models are the only models other than SSiB that performed the sensitivity study in GSWP-2 concerning different treatments of vegetation properties.
The feature of compensation between vegetation transpiration and soil evaporation is further exemplified in Fig. 2. Scatter diagrams of the fPAR anomaly versus differences in transpiration and soil evaporation between uncoupled IAC and MAC cases are shown for different vegetation types over Southern Africa (within the box bounded by 60°S–0°, 0°–60°E) in the beginning (November), the peak (February), and the end (May) of the rainy season. Strong anticorrelations between vegetation transpiration and soil evaporation are immediately evident in all the plots. As expected, transpiration is positively correlated to fPAR, whereas soil evaporation is negatively correlated. However, the responses are not “equal and opposite” on either side of the origin for each vegetation type. There exist very strong linear relationships with little seasonal variation for tropical rain forest. Over savanna, slightly asymmetric relationships are apparent at the beginning and the peak of the rainy season, but become more asymmetric by the end of the rainy season. Compared to tropical forest and savanna, agriculture shows curvilinear relationships with more seasonal variation. The relationships are strongest at the peak of rainy season and taper off toward both ends. Both grassland and broadleaf shrub (steppe) tend to show strongly asymmetric relationships with much stronger sensitivity to positive vegetation changes. Such relationships also exhibit the greatest variability across the seasons, with the peak of the rainy season demonstrating the greatest sensitivities of vegetation transpiration and soil evaporation to fPAR. The sensitivities are weaker at the beginning and the end of the rainy season. Note that broadleaf shrub (Fig. 2d) shows the smallest differences in the water fluxes between IAC and MAC cases over the 10 yr with only −0.2 to 0.2 mm day−1 at the peak of the rainy season. For these more sparsely vegetated categories, the response of fluxes to positive anomalies of fPAR is much stronger than for negative anomalies.
Besides the different strength of response to positive and negative fPAR anomalies, there is another characteristic that explains the features in the Southern Hemisphere in Fig. 1d. Note that for savannah, grassland, and cropland the change in transpiration for a given fPAR anomaly is not completely offset by the change in soil evaporation. The net result is that the resulting anomaly in total evapotranspiration (and thus latent heat flux) is driven mainly by the transpiration response, and is larger for positive anomalies than for negative (given the same magnitude of fPAR anomaly). Thus, a net change in latent heat flux can occur in IAC even though the 10-yr mean values of the vegetation properties are unchanged. In the Northern Hemisphere, the response in the coupled model is more complex, engaging more strongly the atmospheric branch of the hydrologic cycle as evidenced by the change in cloud cover. In the uncoupled runs downward radiation is fixed for both cases, and similar responses are seen over these vegetation types north of the equator, suggesting this mechanism may operate there as well.
We further examine the impact of anomalous vegetation in the coupled system by inspecting the variance in the ensemble run. Using the monthly mean values from 10-yr simulations with ensembles of 10 members each, we calculate the mean of the intraensemble variances from each year. Using the intraensemble standard deviation (hereafter σens) as a measure of uncertainty or noise, we quantify the relative impact of anomalous vegetation as the ratio of the 10-yr mean changes of surface fluxes (shown in Fig. 1) to σens. The results indicate that the σens of these energy terms in the MAC case are greater than 5 W m−2 almost everywhere (outside of the driest deserts), and greater than 15 W m−2 for net shortwave, sensible, and latent heat flux in the world’s monsoon regions. Overall, the changes of all the energy fluxes account for less than 10% of the σens over most of the globe with only a small areas of the land surface achieving 20%. This suggests that the changes seen in Fig. 1 are not large compared to the internal variability of the climate system.
b. Variability
Given that the time mean of the vegetation properties are the same in both IAC and MAC, it is not surprising that there is little difference between the time-mean values of the surface fluxes. Yet, we expect that the increased temporal variability of vegetation properties in IAC would lead to a similar increase in the variability of surface fluxes and other climate variables when compared to MAC. We find that this expected response is not universal. Table 2 shows the fraction of land points where there is an increase in the total temporal standard deviation (σtotal) of various energy and water flux terms between the IAC and MAC simulations. Probabilities are based on a Student’s t test and a conservative estimate of about 235 degrees of freedom for GCM precipitation over all land points excluding Antarctica [based on the method of Dirmeyer (1999); here we simply approximate the degrees of freedom for all surface fluxes with the degrees of freedom for precipitation]. The null expectation is that half of the points would show an increase, and half a decrease ( f = 0.50). The fraction of the global land area with positive values is significant at the 99% confidence level for net longwave radiation, soil evaporation, canopy interception loss, latent heat flux, and evaporation. For net shortwave radiation, f is significant at a 93% confidence level. Such results indicate that the use of interannually varying vegetation properties tends to increase the temporal variability of these fluxes more often than not. Nevertheless, there are many areas with a decrease in variability, either because of a stabilization effect of variable vegetation, or because of a large noise component in the variability. There appears to be no dominant vegetation impact on increasing temporal variabilities of sensible heat flux or vegetation transpiration.
The same statistical test is applied to the global mean of changes in temporal standard deviation, shown in the last two columns of Table 2. The global means of differences in σtotal between IAC and MAC are not discernable from chance variations for the sensible and latent heat fluxes, vegetation transpiration, and evaporation. However, the differences for the other fluxes are significant at the 99% confidence level (except for net shortwave radiation, which is significant at the 96% confidence level), suggesting that the IAC simulation (i.e., the naturally varying vegetation parameters) tends to increase the temporal variability for these fluxes. Note that the results from these two kinds of significant tests are not interchangeable: one (the fraction of the global land area with positive values) examines the distribution of proportions (in terms of frequency) and the other measures the distribution of mean (in terms of magnitude).
c. Skill of simulations
Although the purpose of this paper is to assess the sensitivity of the COLA GCM to vegetation anomalies, it is useful to compare the model simulations to what was actually observed. The skill of precipitation and near-surface air temperature simulations, measured in terms of local temporal correlations and root-mean-square error (RMSE) across the 10-yr period, is detailed in this section. To quantify the skill of the coupled land–atmosphere model in simulating the temporal variability and the ability of the model to capture climate anomalies over land, a standard anomaly correlation coefficient (ACC) between observations and model simulations is calculated at each terrestrial model grid point across the 10-yr sample. Anomalies for observations are relative to the mean annual cycle for the 10-yr period 1986–95, and anomalies for the model are calculated for the ensemble mean of each month relative to the model’s mean annual cycle over the same 10 yr.
Figure 3 shows the global distribution of the temporal ACC of precipitation for MAC and IAC simulations and their differences in significance level (calculated separately at each point) as well as the skill of precipitation simulations in terms of RMSE. Also shown in each panel is the fraction of the global land area with positive values. Comparison is made with monthly gridded observational estimates from the Climate Prediction Center (CPC) Merged Analysis of Precipitation (CMAP; Xie and Arkin 1997). The significance level at each grid point is calculated using the Student’s t test with the degrees of freedom at each grid point derived on the basis of the lag autocorrelation (the time scale where correlations drop below 1/e). When all 10 yr are considered, there appears to be no obvious impact from different specifications of vegetation parameters. The temporal ACC of precipitation from IAC and MAC simulations exhibit similar spatial patterns. Regions with moderate positive correlations (0.2–0.4) are concentrated in the northern Amazon, central and South Africa, the southwestern United States, Australia, New Guinea, and the Middle East. These are mainly regions where ENSO influence on precipitation is known to exist. There is only a regional indication of skill in the model simulations of precipitation over land. In terms of differences in the significance levels between IAC and MAC, most of the globe remains almost unchanged with the positive and negative changes close to an even split. The pattern of RMSE largely reflects the distribution of precipitation—highest in the tropics and lowest over deserts. Overall, there is no apparent impact of vegetation on the 10-yr time series of precipitation as the RMSE of IAC and MAC simulations remaining essentially the same (i.e., less than 0.05 mm day−1 differences) over most of the globe.
Figure 4 shows the ensemble mean anomaly correlation coefficients for near-surface air temperature, in the same manner as Fig. 3. Comparison here is made to the Climate Anomaly Monitoring System (CAMS) monthly mean gridded temperature data (Ropelewski et al. 1985) corrected for the difference in elevation between the model grid boxes and the observing stations. The skill over land is much better for near-surface air temperature than for precipitation in both runs. There are large spatially coherent regions of high correlation. The ACC is greater than 0.3 over most of the globe but can be as high as 0.9 over several regions, such as Mexico, the west coast of South America, Ethiopia, Madagascar, southern Asia, New Guinea, and the adjacent Indonesian islands. Much of the globe has model correlations with observations that exceed the 90% local significance level. There is little skill evident in the high latitudes, the Guinea coast of Africa, the Pampas of South America, and central Australia. Correlation improvement of IAC over MAC is strongest over Europe, the Great Plains of North America, and southeastern Russia—in regions mostly aligning with grassland and cropland. There are also regions where correlations decrease, particularly Alaska and the Yukon. High-latitude regions are most susceptible to bogus significance due to the baroclinic nature of the atmosphere and large seasonal and synoptic temperature variations. Overall, there is no net global vegetation impact with significance levels remaining unchanged between IAC and MAC over most of the globe (i.e., the fraction of the global land area with positive values is 0.50). Monthly RMSE for anomaly fields are generally less than 2 K south of 45°N, but can attain 3–4 K in the high latitudes (Fig. 4). This is the case for both IAC and MAC simulations. Their similar performances are further verified by their RMSE differences. Over most of the globe, the differences in RMSE between IAC and MAC simulations fall within the range of ±0.1 K, suggesting that there is no evident vegetation impact on skill of near-surface air temperature simulations when all 10 yr are considered together. The statistical test for the fraction of the global land area with the positive differences in the significance level of the temporal ACC (or negative differences in RMSE) between IAC and MAC gives consistent results.
It is helpful to also examine the impact of vegetation in terms of the spatial anomaly correlation coefficients (SACCs) and RMSE of the monthly anomalies of precipitation and temperature simulations. Figure 5 shows the monthly progression of the ensemble mean spatial ACC (pattern correlation) and RMSE over land points (60°S–90°N), averaged across 10 yr for the IAC and MAC cases. Note that these are not the same quantities as shown in Figs. 3 and 4, which are temporal ACC and RMSE. The SACC of precipitation in both cases fluctuates in the same fashion across the annual cycle. Correlations start high in January, drop in February, and gradually increase to a peak in May before declining again. There is a dramatic overall drop in July with correlations at a minimum and approaching 0.0 for reasons that are not clear but may be due to timing errors in the onset of the boreal monsoons. The IAC case shows an improvement in the otherwise weak skill for all months except July, August, and December. The greatest improvements occur in January and May. Although small, the improvements exist in nearly all months, suggesting they are not the result of chance. The overall skill of surface temperature in both cases also evolves similarly from month to month with two peaks in May and November. The surface temperature shows a systematically higher level of skill and stronger seasonal variations than precipitation, but there is not a strong systematic improvement in spatial correlations for the IAC case. On average, monthly SACC for anomaly fields is less than 0.1 for precipitation and 0.25 for temperature. There appears to be no systematic improvement in RMSE for precipitation or surface temperature for the IAC case. Monthly RMSE for the anomaly fields are about 1.1 mm day−1 for precipitation and 2.5 K for temperature. Except for precipitation SACC, there appears to be little favorable impact on climate simulations by the use of realistic vegetation parameters in this model configuration.
In summary, there is little evidence of a clear and consistent impact of vegetation variations on the climate in this model when the entire 10-yr period is considered. However, the impact when seasonal vegetation anomalies are large may be detectable but lost in the previous analyses, just as the influence of tropical sea surface temperature anomalies on climate and its predictability is strong during the peak of El Niño and La Niña events, but difficult to detect otherwise. Some case studies are examined in the next section to see if there are strong episodic impacts of realistic vegetation anomalies on climate simulations.
4. Regional cases
There are a number of significant regional climate events that occurred during the years simulated. The late 1980s and early 1990s were a period of considerable variability in rainfall over many regions that rely heavily on seasonal precipitation regimes. How well did the models reproduce these anomalies? We have seen that the climate model already does rather well in simulating temperature anomalies. Here we focus on the impact of vegetation properties on the precipitation response. All anomalies for observations and model simulations analyzed hereafter are calculated relative to the mean of the 10-yr GSWP-2 period (1986–95). We did not choose cases on the basis of precipitation anomalies. In fact, we first look for large-scale persistent anomalies in vegetation, as described later in this section. Several of those cases correspond to well-known hydrologic events.
a. North America
The 1988 summer drought and the 1993 summer flood over the central United States are obvious cases to examine first. We investigate how each model simulation performs in depicting the evolution of the flood with the anomalies of May, June, and July over North America (Fig. 6). These 3 months are selected as the floods of 1993 were documented to be largely concentrated during a 4–5-week span from mid-June to mid-July. The May anomaly in the IAC represents the initiation of the flood with comparable magnitude to observations, but the center is shifted slightly to the west and lacks the dry anomalies to the east and south. However, both IAC and MAC simulations give a much more widespread distribution of wet anomalies than observed. There appear to be some vegetation impacts—IAC simulates a better magnitude for the wet anomaly over the Midwest and a better pattern over eastern Canada as well as a better pattern of the dry anomaly over southern Canada. A somewhat stronger picture emerges when the June anomalies are considered. There is greater resemblance between IAC and the observation in terms of pattern and magnitude. MAC shows an unrealistic distribution, ranging from wet in the west to very dry over the Ohio River valley and along the East Coast. The rainfall signal in the Midwest is very well captured in the IAC simulation, indicating that the vegetation impact is significant in simulating the development of the flood. However, the vegetation impact starts to disappear and the signal in IAC becomes weak in July. As a result, both model simulations begin to behave similarly; neither does well in simulating the later portion of the flood when strong moisture advection from the Caribbean drove flooding (Dirmeyer and Brubaker 1999).
Figure 7 shows the seasonal precipitation anomaly [June–August (JJA)] during the 1988 summer drought over North America in the same manner as Fig. 6. The dry anomaly is very prominent in the observations over a broad area of the contiguous United States, stretching from the West Coast to the Great Lakes. Each of the model simulations produces a dry anomaly that is weaker by at least a factor of 4, and displaced well to the northwest relative to observations. The IAC and MAC simulations are very similar to each other. Neither model simulation represents the dry anomaly well over the southern United States and along the southeast coast. The anomalies from individual months show similar problems (figures not shown). There seems to be little vegetation impact on simulating the 1988 summer rainfall signal over the United States. In fact, it should be noted that the drought of 1988 was documented to be concentrated mostly during spring and early summer, and had began to abate by July. June–August is examined here because the anomalies in the vegetation properties were greatest during these months.
b. Tropics
There are many contrasting years for surface hydrology over southern Asia and the tropics. For instance, 1986 and 1987 were consecutive years of poor monsoon rainfall over India, whereas 1988 was a very wet year. There are smaller but still relevant variations during the 1990s. Over the Sahel region of Africa, the decade-long drought that began during the 1960s persisted through the GSWP-2 period. The drought was particularly severe in 1986, 1987, and 1990 with the only truly wet year in the period being 1994 (1988 rainfall was near normal when compared to the rest of the twentieth century). Figure 7 shows the seasonal anomalies over India and Africa during the dry year of 1986 and the wet year of 1994. Both the IAC and MAC simulations do a credible job of resolving the dry anomaly pattern of 1986 over India and the Sahel region of Africa. The performances of two model simulations are quite similar except that the dry anomaly in the IAC case seems to be drawn farther into the Congo–Rift Valley area—a pattern that is also evident in the observations. In the summer of 1994, IAC appears to simulate the flood signal over India better than MAC, but with errors in position (slightly displaced to the east). Yet both model simulations do a fairly good job of capturing the observed drought signal in south India. Over the Sahel, the MAC simulation appears to resolve the observed wet signal to some extent. The IAC simulation does not improve over MAC and gives a diffuse and mixed response. Overall, the performances of the model simulations are poor over Africa in 1994.
The final example, shown in Fig. 7, are the anomalies of the wet season (February–May) over South America during 1990 and 1995. Note that this region generally shows a strong maritime influence where SST anomalies can affect the climate. During 1990, IAC appears to do a very good job of capturing the observed dry signal over the Nordeste region and the observed wet signal over northern Amazon, whereas MAC does not appear to capture the signals well over these two areas. However, the wet signal over southern Brazil and northern Argentina is simulated slightly better in MAC than in IAC. During 1995, both model simulations exhibit the same strengths and weaknesses over the tropical band—MAC and IAC do rather well in simulating the wet signal over the Nordeste region. Both fail to capture the observed dry conditions over northern Amazon. The improvement of IAC over MAC is mainly over southern Brazil and northern Argentina where the observed dry signal is better simulated in IAC.
c. Statistics
These regional cases show mixed results. Previous examination of global maps of the simulated vegetation impact on the various flux terms reveals that the sensitivity may be limited to certain times and locations with significant vegetation anomalies. Here we summarize the cohesive regions across the globe where the fPAR anomalies of at least 2 months from the 10-yr mean annual cycle maintain 1.5 or more standard deviations. The standard deviation is calculated for each month from the 1986–95 time series. Such a criterion sometimes results in relatively small areas of 10 pixels or so, but results in about 2 dozen cases over each continent. Nevertheless, by selecting the study regions in this fashion we hope to identify the impact of significant regional vegetation differences. The selected regions are grouped by continent; North America (29 cases), South America (29 cases), Africa (31 cases), Europe (28 cases), Asia (20 cases), and Australia (23 cases). Each region is then individually analyzed for precipitation, near-surface air temperature, sensible and latent heat flux, evaporation, cloud cover, and shortwave and longwave radiation.
The region-based statistics seem to show the same surface impacts that are generally true across the globe. Decrease in vegetation density causes upward shortwave radiation to increase, while often precipitation and evapotranspiration decrease. As a result, canopy interception loss and vegetation transpiration also decreases. The decrease in evapotranspiration further reduces the formation of cloud cover and maintains a positive feedback between the land surface and the atmosphere. The impact of vegetation density on surface temperature is more complicated. On the one hand, surface temperature tends to increase as a result of decreased evapotranspiration (less latent heat) and decreased cloud cover (more incoming shortwave radiation). On the other hand, surface temperature can decrease as a result of higher albedo (absorbing less solar radiation). Such opposite feedbacks tend to cancel each other and result in relatively small net change in surface temperature. These results are in good agreement with previous studies. Among all the cases extracted, about one-third show the expected response for IAC, half of the cases have weak response, and the remaining cases show unexpected changes in the IAC ensemble. Table 3 shows the average of all the extracted regions composited by positive and negative fPAR anomalies, respectively. As can be seen, the average fPAR anomaly for the selected events is about ±20%, which seems to result in a precipitation response of about ±6%. Although there is a tremendous amount of noise compared to the signal across individual regions, by averaging together all of the extreme cases, the global statistics are sound and robust.
Investigation of the model’s ability to simulate regional patterns of precipitation anomalies reveals inconstant model performance. For example, anomalous vegetation yields a much better representation of the observed wet anomalies in the 1993 U.S. summer flood, but shows the same weakness as climatological vegetation in simulating the observed dry anomalies during the 1988 U.S. drought. What causes the difference? The regional time series of area-average model-simulated soil wetness indicates similar annual cycles in two cases across the study period (Fig. 8). The annual cycle for the GCM simulations is generally smaller than for GSWP-2. This is partially explained by the use of an ensemble mean for the GCM, but that does not account for the different ranges of variations in many regions, such as India (too dry during the monsoon) or Nordeste (too wet in the dry season). Those differences reflect the balance of errors in the fluxes affecting the water balance, particularly precipitation and net radiation (which affects evapotranspiration).
When the near-surface soil moisture remains in extreme states or fails to leave a narrow range, small perturbations have no lasting impact on evapotranspiration or the partitioning of heat fluxes (Dirmeyer 2001). This insensitivity could either hamper the skill for the climate model in simulating the response of the climate to the observed SST variability, or be large enough to overcome the establishment of any reinforcing feedbacks between the land and atmosphere by the use of realistic (interannually varying) vegetation.
Others have found that biases in the hydrologic cycle impact GCM sensitivity to vegetation-induced feedbacks. Kim and Wang (2007) found a similar response over North America in the NCAR climate model, with greater sensitivity and precipitation feedbacks for positive vegetation anomalies than for negative ones. However, in our experiment the model performance is rather different over India and Africa. Examination of all relevant years in the GSWP-2 period indicates that vegetation impacts sometimes show better performance in cases of dry anomalies than in wet anomalies in these areas. The regional time series of soil wetness in Fig. 8 show that India remains very dry, which should put the soil moisture in a perpetually insensitive range. In such cases, vegetation variations may have little direct impact on the water cycle although they may still affect precipitation through changes in the local radiation or momentum balances. The tropical band in South America is the only region where both model simulations perform well in dry and wet conditions, although more improvement is found during dry conditions. Given the wet bias evident in this area, sensitivity may be greater for dry anomalies.
5. Conclusions
A global climate model has been integrated in the ensemble mode for 1986–95 with two different treatments of specified vegetation parameters—observed interannually varying vegetation versus a climatological annual cycle. The impact of realistically varying vegetation on land surface climate and its potential role in improving the hindcasts are examined in this modeling framework. Parallel integrations are also implemented and analyzed in an uncoupled mode within the GSWP-2 framework for the same 10-yr period to isolate the behavior of specific components in the response of the land surface to vegetation phenology variability. Our main objective is to determine the significance of near-real-time continuous global monitoring of vegetation (by satellite) for subseasonal-to-seasonal climate prediction and the potential utility of long-term datasets of global vegetation properties for understanding past and future climate variability. Positive results would also motivate the inclusion of predictive phenology in forecast models.
In the coupled simulations, several mechanisms seem to be responsible for the sensitivities of surface fluxes to vegetation anomalies. In some regions, changes in some surface flux fields (e.g., latent heat flux) occur directly as a result of the changing surface vegetation properties, or more accurately, due to its component fluxes (vegetation transpiration and soil evaporation) exhibiting asymmetric strength of response to positive and negative vegetation anomalies. Over other regions, the responses could involve more strongly the atmospheric branch of the hydrological cycle. There appear to be the compensating changes in the 10-yr mean of sensible and latent heat flux, which suggests that much of the impact of vegetation variations is manifested in the partitioning of available surface energy between latent and sensible heat fluxes. Spatially more often than not, interannual variations of vegetation tend to increase temporal variability for net longwave radiation, latent heat flux, and its component fluxes (except transpiration) when compared to climatological vegetation, but lead to increasing the global means of temporal variability only for net shortwave and longwave radiation, soil evaporation, and canopy interception loss. Overall, signals are weak when the entire 10-yr time series is used for statistical evaluations.
We find moderate skill on the regional scale, measured in terms of local temporal ACC, in the model simulations of precipitation over land. However, interannual variations of vegetation often do not seem to help improve the skill appreciably. The baseline skill is much better for surface temperature than for precipitation with large spatially coherent regions of high correlation (>0.5). Specification of interannually varying vegetation properties leads to regional improvements, mostly in areas of grassland and cropland. As far as simulations of spatial patterns (measured by spatial ACC) are concerned, surface temperature also shows a systematically higher level of skill and stronger seasonal variations than precipitation. Unlike the simulation of temporal variability, interannually varying vegetation tends to improve simulation of spatial patterns of precipitation for most of the months, but does not result in any strong systematic improvement for surface temperature.
Overall, favorable impacts are episodic in this model configuration, as indicated by several skill measures. Thus, we focus on cases where vegetation anomalies are large and persistent over cohesive regions to determine if there is sensitivity of rainfall in specific episodes. Results indicate that surface fluxes, temperature, and precipitation respond to vegetation anomalies with variable strength among all the extracted cases. When a significant response occurs, it is often not in the direction we might expect, suggesting the reaction of climate to vegetation anomalies is more complex than we hypothesized. Despite a large amount of scatter, the average of all the extracted cases composited by positive and negative fPAR anomalies indicates that a fPAR anomaly of ±20% results in a precipitation response of about ±6%. There are suspicious variations in the vegetation data that appear unrealistic, which cause us to view this more as a sensitivity experiment than a validation effort.
In summary, the impact of vegetation variations on the hydrological cycle in this model is not evident when the entire 10-yr period is considered. However, when seasonal vegetation anomalies are large, the vegetation impact is detectable but not overwhelming. This is analogous to the influence of tropical SST anomalies on climate—predictability is clearly enhanced during the peak of strong El Niño and La Niña events, but is often difficult to detect otherwise. In addition, it might be that the signal would have been much stronger in monthly or seasonal hindcasts than that for the same seasons from this 10-yr-long simulation. Also, Dirmeyer et al. (2006a) showed that this GCM and all other GCMs that participated in the Global Land–Atmosphere Coupling Experiment (GLACE) have fundamental problems simulating key aspects of land–atmosphere interactions. These shortcomings may hinder the ability of this model to respond to vegetation anomalies.
It should be pointed out that extreme climate anomalies in the real world may be the result of a favorable superposition of both a boundary-forced response and internal atmospheric variability. Climate models generally do a poor job of simulating precipitation anomalies outside of those regions strongly affected by tropical SST anomalies. The ensemble averaging improves a model’s ability to capture anomaly patterns by suppressing the internal noise and thus enhancing the boundary-forced signal. However, at the same time, the capacity of the model to reproduce extreme anomalies may be curtailed as a result of reduced variance of the ensemble compared to that of a single integration. This may explain the tendency of the model simulations to underforecast the relative strength of anomalies. Likewise, the more that extremes in climate episodes are actually the product of chance, rather than the result of anomalies in the land or ocean, the less hope we have in simulating and predicting specific events.
Returning to our original motivation for this study, we reconsider the questions posed in the introduction. The sensitivities to vegetation anomalies that we see appear to be strictly contained in the land surface fluxes in some cases, and part of a larger feedback with the atmosphere and greater water cycle in others. The response to realistic vegetation anomalies is largely local, highly episodic in nature, and most likely for large anomalies. However, there is no guarantee that a strong vegetation anomaly will result in significant precipitation or temperature anomalies. Is vegetation a useful element of the land surface for enhancing seasonal predictability? Our results suggest the answer is probably yes, but only marginally. It is possible that there is much more predictability to be harvested from interannual variations of vegetation than we realized from this experiment. It should be noted that the specific details of the results of this investigation are almost certainly dependent on the choice of the model. Although mechanisms exist in most physically based LSPs through which changes in the seasonal or spatial distribution of vegetation parameters could affect the land surface, the actual influence of vegetation parameters will depend on which processes a land surface model includes, and how they are parameterized (e.g., a model might or might not use LAI in the canopy albedo parameterization). It may well be that other sources of error in this climate model prevent all of the potential improvement from realistic vegetation from being realized. Alternatively, it could be that there is genuinely little skill to be gained from specification of realistic vegetation variability in this model. Care should be taken in the generalization of these results to other models. At the very least, this study suggests a lower bound for the degree of impact of anomalies in vegetation phenology on climate.
Acknowledgments
This work was supported partially under the National Aeronautics and Space Administration Grant NAG5-11579, and as part of omnibus research at the Center for Ocean–Land–Atmosphere Studies, supported by NSF Grant ATM-0122850, NOAA Grant NA16-GP2248, and NASA Grant NAG5-11656.
REFERENCES
Bacmeister, J., Pegion P J. , Schubert S D. , and Suarez M J. , 2000: Atlas of Seasonal Means Simulated by the NSIPP 1 Atmospheric GCM. Vol. 17, NASA Tech. Memo. 2000-104606, NASA, 194 pp.
Bohren, C F., and Albrecht B A. , 1998: Atmospheric Thermodynamics. Oxford University Press, 402 pp.
Bounoua, L., Collatz G J. , Los S O. , Sellers P J. , Dazlich D A. , Tucker C J. , and Randall D A. , 2000: Sensitivity of climate to changes in NDVI. J. Climate, 13 , 2277–2292.
Briegleb, B P., 1992: Delta-Eddington approximation for solar radiation in the NCAR community climate model. J. Geophys. Res., 97 , 7603–7612.
Charney, J G., 1975: Dynamics of deserts and drought in the Sahel. Quart. J. Roy. Meteor. Soc., 101 , 193–202.
Chase, T N., Pielke R A. , and Kittel T. G. F. , 1996: Sensitivity of a general circulation model to global changes in leaf area index. J. Geophys. Res., 101 , 7393–7408.
Collins, W D., Hackney J K. , and Edwards D P. , 2002: An updated parameterization for infrared emission and absorption by water vapor in the National Center for Atmospheric Research Community Atmosphere Model. J. Geophys. Res., 107 .4664, doi:10.1029/2001JD001365.
Delire, C., Behling P. , Coe M T. , Foley J A. , Jacob R. , Kutzbacin J. , Liu Z. , and Yarrus S. , 2001: Simulated response of the atmosphere-ocean system to deforestation in the Indonesian Archipelago. Geophys. Res. Lett., 28 , 2081–2084.
Dickinson, R E., 1992: Land surface. Climate System Modeling, K. E. Trenberth, Ed., Cambridge University Press, 149–171.
Dickinson, R E., and Henderson-Sellers A. , 1988: Modeling tropical deforestation: A study of GCM land-surface parameterizations. Quart. J. Roy. Meteor. Soc., 114 , 439–462.
Dirmeyer, P A., 1999: Assessing GCM sensitivity to soil wetness using GSWP data. J. Meteor. Soc. Japan, 77 , 367–385.
Dirmeyer, P A., 2001: Climate drift in a coupled land–atmosphere model. J. Hydrometeor., 2 , 89–100.
Dirmeyer, P A., 2005: The land surface contribution to the potential predictability of boreal summer season climate. J. Hydrometeor., 6 , 618–632.
Dirmeyer, P A., and Shukla J. , 1994: Albedo as a modulator of climate response to tropical deforestation. J. Geophys. Res., 99 , 20863–20877.
Dirmeyer, P A., and Brubaker K L. , 1999: Contrasting evaporative moisture sources during the drought of 1988 and the flood of 1993. J. Geophys. Res., 104 , 19383–19397.
Dirmeyer, P A., and Zeng F J. , 1999: An update to the distribution and treatment of vegetation and soil properties in SSiB. COLA Tech. Rep. 78, 25 pp. [Available from the Center for Ocean–Land–Atmosphere Studies, 4041 Powder Mill Rd., Suite 302, Calverton, MD 20705.].
Dirmeyer, P A., Koster R D. , and Guo Z. , 2006a: Do global models properly represent the feedback between land and atmosphere? J. Hydrometeor., 7 , 1177–1198.
Dirmeyer, P A., Gao X. , Zhao M. , Guo Z. , Oki T. , and Hanasaki N. , 2006b: The Second Global Soil Wetness Project (GSWP-2): Multimodel analysis and implications for our perception of the land surface. Bull. Amer. Meteor. Soc., 87 , 1381–1397.
Fennessy, M J., and Shukla J. , 1999: Impact of initial soil wetness on seasonal atmospheric prediction. J. Climate, 12 , 3167–3179.
Guo, Z., Dirmeyer P A. , Hu Z-Z. , Gao X. , and Zhao M. , 2006: Evaluation of GSWP-2 Soil Moisture Simulations. Part 2: Sensitivity to external meteorological forcing. J. Geophys. Res., 111 .D22S03, doi:10.1029/2006JD007845.
Hales, K., Neelin J D. , and Zeng N. , 2004: Sensitivity of tropical land climate to leaf area index: Role of surface conductance versus albedo. J. Climate, 17 , 1459–1473.
Hall, F G., and Coauthors, 2006: ISLSCP Initiative II global data sets: Surface boundary conditions and atmospheric forcings for land-atmosphere studies. J. Geophys. Res., 111 .D22S01, doi:10.1029/2006JD007366.
Harshvardhan, and Davies, R., Randall D A. , and Corsetti T G. , 1987: A fast radiation parameterization for general circulation models. J. Geophys. Res., 92 , 1009–1016.
Henderson-Sellers, A., Dickinson R E. , Durbridge T B. , Kennedy P J. , McGuffie K. , and Pitman A J. , 1993: Tropical deforestation: Modeling local to regional-scale climate change. J. Geophys. Res., 98 , 7289–7315.
Hong, S., and Pan H-L. , 1996: Nonlocal boundary layer vertical diffusion in a medium-range forecast model. Mon. Wea. Rev., 124 , 2322–2339.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77 , 437–471.
Kanamitsu, M., Ebisuzaki W. , Woollen J. , Yang S-K. , Hnilo J J. , Fiorino M. , and Potter G L. , 2002: NCEP–DOE AMIP-II reanalysis (R-2). Bull. Amer. Meteor. Soc., 83 , 1631–1648.
Kiehl, J T., Hack J J. , Bonan G. , Boville B A. , Williamson D L. , and Rasch P J. , 1998: The National Center for Atmospheric Research Community Climate Model: CCM3. J. Climate, 11 , 1131–1149.
Kim, Y., and Wang G. , 2007: Impact of vegetation feedback on the response of precipitation over North America in the coupled land–atmosphere model CAM3–CLM3. J. Hydrometeor., 8 , 513–533.
Kinter, J L., and Coauthors, 1997: The COLA atmosphere-biosphere general circulation model. Vol. 1: Formulation. COLA Tech. Rep. 51, 46 pp. [Available from the Center for Ocean–Land–Atmosphere Studies, 4041 Powder Mill Rd., Suite 302, Calverton, MD 20705.].
Koster, R D., and Coauthors, 2004: Regions of strong coupling between soil moisture and precipitation. Science, 305 , 1138–1140.
Lean, J., and Rowntree P R. , 1997: Understanding the sensitivity of a GCM simulation of Amazonian deforestation to the specification of vegetation and soil characteristics. J. Climate, 10 , 1216–1235.
Marx, L., 2002: New calculation of saturation specific humidity and saturation vapor pressure in the COLA atmospheric general circulation model. COLA Tech. Rep. 130, 25 pp. [Available from the Center for Ocean–Land–Atmosphere Studies, 4041 Powder Mill Rd., Suite 302, Calverton, MD 20705.].
Mellor, G L., and Yamada T. , 1982: Development of a turbulence closure model for geophysical fluid processes. Rev. Geophys. Space Phys., 20 , 851–875.
Misra, V., and Coauthors, 2006: Validating ENSO simulation in coupled climate models. COLA Tech. Rep. 210, 50 pp. [Available from the Center for Ocean–Land–Atmosphere Studies, 4041 Powder Mill Rd., Suite 302, Calverton, MD 20705.].
Moorthi, S., and Suarez M J. , 1992: Relaxed Arakawa–Schubert: A parameterization of moist convection for general circulation models. Mon. Wea. Rev., 120 , 978–1002.
Nicholson, S E., Tucker C J. , and Ba M B. , 1998: Desertification, drought, and surface vegetation: An example from the West Africa Sahel. Bull. Amer. Meteor. Soc., 79 , 815–829.
Pitman, A J., Zhao M. , and Desborough C E. , 1999: Investigating the sensitivity of a land surface scheme’s simulation of soil wetness and evaporation to spatial and temporal leaf area index variability within the Global Soil Wetness Project. J. Meteor. Soc. Japan, 77 , 281–290.
Polcher, J., and Laval K. , 1994: A statistical study of the regional impact of deforestation on climate in the LMD GCM. Climate Dyn., 10 , 205–219.
Reynolds, R W., and Smith T M. , 1994: Improved global sea surface temperature analyses using optimal interpolation. J. Climate, 7 , 929–948.
Ropelewski, C F., Janowiak J E. , and Halpert M F. , 1985: The analysis and display of real-time surface climate data. Mon. Wea. Rev., 113 , 1101–1107.
Rowell, D P., and Blondin C. , 1990: The influence of soil wetness distribution on short-range rainfall forecasting in the West Africa Sahel. Quart. J. Roy. Meteor. Soc., 116 , 1471–1485.
Sela, J G., 1980: Spectral modeling at the National Meteorological Center. Mon. Wea. Rev., 108 , 1279–1292.
Shukla, J., and Mintz Y. , 1982: Influence of land-surface evapotranspiration on the earth’s climate. Science, 215 , 1498–1501.
Shukla, J., Nobre C. , and Sellers P. , 1990: Amazon deforestation and climate change. Science, 247 , 1322–1325.
Sud, Y C., and Molod A. , 1988: A GCM simulation study of the influence of Saharan evapotranspiration and surface-albedo anomalies on July circulation and rainfall. Mon. Wea. Rev., 116 , 2388–2400.
Tiedtke, M., 1984: The effect of penetrative cumulus convection on the large-scale flow in a general circulation model. Beitr. Phys. Atmos., 57 , 216–239.
Xie, P., and Arkin P A. , 1997: Global precipitation: A 17-year monthly analysis based on gauge observations, satellite estimates, and numerical model outputs. Bull. Amer. Meteor. Soc., 78 , 2539–2558.
Xue, Y., 1997: Biosphere feedback on regional climate in tropical North Africa. Quart. J. Roy. Meteor. Soc., 123 , 1483–1515.
Xue, Y., and Shukla J. , 1993: The influence of land surface properties on Sahel climate. Part I: Desertification. J. Climate, 6 , 2232–2245.
Xue, Y., Sellers P J. , Kinter J L. , and Shukla J. , 1991: A simplified biosphere model for global climate studies. J. Climate, 4 , 345–364.
Zeng, X., 2001: Global vegetation root distribution for land modeling. J. Hydrometeor., 2 , 525–530.
Zhao, M., and Dirmeyer P. , 2003: Production and analysis of GSWP-2 near-surface meteorology data sets. COLA Tech. Rep. 159, 22 pp. [Available from the Center for Ocean–Land–Atmosphere Studies, 4041 Powder Mill Rd., Suite 302, Calverton, MD 20705.].
The 10-yr mean differences between IAC and MAC cases of energy flux terms: (a) net longwave radiation, (b) net shortwave radiation, (c) sensible heat flux, (d) latent heat flux (W m−2), and (e) cloud cover (dimensionless percentage) averaged for 10 ensemble members in GCM simulations.
Citation: Journal of Hydrometeorology 9, 3; 10.1175/2007JHM931.1
Scatterplot of the fPAR anomaly vs differences in vegetation transpiration (black ○, mm day−1) and soil evaporation (gray +, mm day−1) between IAC and MAC for different vegetation types over southern Africa (60°S–0°, 0°–60°E) across the 10-yr period: (a) tropical rain forest, (b) savanna, (c) grassland, (d) steppe, and (e) cropland. Each symbol represents one grid point in 1 yr.
Citation: Journal of Hydrometeorology 9, 3; 10.1175/2007JHM931.1
Temporal ACCs between (a) MAC and (b) IAC and observed precipitation as well as (c) the difference in their significance levels (IAC − MAC); RMSE (units of mm day−1) between (d) MAC and (e) IAC and observed precipitation as well as (f) the difference in their RMSE (IAC − MAC). Also shown in the bottom-right corner of (a)–(f) is the fraction of the global area with positive values for temporal ACCs and the fraction of the global area with negative values for RMSE. In (c), the bottom labels and the vertical labels in the middle represent the significance levels of MAC and IAC, respectively. The gray color indicates no change between the significance levels of MAC and IAC.
Citation: Journal of Hydrometeorology 9, 3; 10.1175/2007JHM931.1
Same as in Fig. 3, but for near-surface air temperature. The regions with the local significance level less than 90% (white color) are masked out in (a) and (b). Units are K for RMSE.
Citation: Journal of Hydrometeorology 9, 3; 10.1175/2007JHM931.1
The 10-yr mean of global (land only, 60°S–90°N) (a) spatial ACCs and (b) RMSE between MAC, IAC, and observations for precipitation (circle) and near-surface air temperature (triangle) in GCM simulations.
Citation: Journal of Hydrometeorology 9, 3; 10.1175/2007JHM931.1
Precipitation anomalies (mm day−1) from the observation, MAC, and IAC simulations over North America for the months of May, June, and July during the 1993 summer flood.
Citation: Journal of Hydrometeorology 9, 3; 10.1175/2007JHM931.1
Seasonal precipitation anomalies (mm day−1) from the (top) observation, (middle) MAC, and (bottom) IAC simulations over North America (left to right) during the 1988 summer drought (JJA), over India and Africa during the dry year of 1986 (JJAS) and the wet year of 1994 (JJAS), and over South America during the dry year of 1990 (FMAM), and the wet year of 1995 (FMAM).
Citation: Journal of Hydrometeorology 9, 3; 10.1175/2007JHM931.1
Time series of root zone soil wetness averaged for different regions from the GSWP-2 multimodel analysis (line) and for soil layers 2–4 for the GCM (symbols).
Citation: Journal of Hydrometeorology 9, 3; 10.1175/2007JHM931.1
Spatial correlations among pairs of the 10-yr mean differences between the IAC and MAC cases of energy fluxes and cloud cover, averaged across 10 ensemble members in the coupled simulations.
The fraction of the global area with positive values for the differences in total temporal standard deviation (σtotal) between the IAC and MAC cases ( f ), and the global mean of the differences of σtotal between the IAC and MAC cases (d). Also shown are the confidence levels. Units of d for energy flux terms are W m−2 and for water flux terms are mm day−1.
Average of all the extracted regions composited by positive (i.e., 53) and negative (i.e., 107) fPAR anomalies for differences between the IAC and MAC cases of various land surface variables in the coupled simulations.