2.1. Model and Implementation
 The model Simple Biosphere, Version 3 (SiB3) is a land surface model originally designed [Sellers et al., 1986] for use with General Circulation Models but used here in an “offline” mode to represent ecosystem physiology as driven by weather and vegetation. SiB [Sellers et al., 1996a, 1996b, 1996c] estimates gross photosynthesis (P) with a modified version of Farquhar et al.'s  model [Collatz et al., 1991] scaled by the canopy integration scheme of Sellers et al. [1996c] and coupled to the Ball-Berry stomatal conductance model [Ball et al., 1988; Collatz et al., 1991, 1992]. Net photosynthesis (Pnet) accounts also for the canopy-integrated leaf respiration, which is on the order of 10% of gross photosynthesis. Gross photosynthesis, net photosynthesis, and leaf respiration are scaled by stresses from the lack of soil water, high or low temperatures, and low air humidity as described further in Appendices A and B. SiB scales these “top leaf” photosynthetic rates to the canopy with the integration scheme of Sellers et al.  that defines a canopy-scale, photosynthetically active radiation use parameter (Π) estimated from the fractional absorption of photosynthetically active radiation (fPAR) derived from remotely sensed vegetation reflectances, here Normalized Difference Vegetation Index (NDVI), and vegetation type.
 The model's respiration rate depends on temperature and moisture and was scaled to produce annual balance with photosynthesis as by Denning et al. . This scaling is related to the model's lack of carbon pools then substituted by a first-order approximation that assumes respiration is driven by recent photosynthesis. Implemented as a calendar year annual balance the approach does not account for long-term sources and sinks. A recent addition to SiB partitions ecosystem respiration into autotrophic (Ra) and heterotrophic (Rh) components following Schaefer et al. . Among other things, this enables estimation of net primary production (NPP) from gross, canopy-scale photosynthesis minus autotrophic respiration (NPP = PΠ − Ra). However, owing to uncertainty in the parameters controlling respiration's partition as well as weakness in the assumption of a calendar year balance between gross photosynthesis and ecosystem respiration, our analysis centers on photosynthesis terms and only examines NPP to place these results in the general context of those published. Still, favorable comparison between SiB's net ecosystem CO2 exchange and that measured with the eddy covariance technique lends confidence to the model's process-level representation of ecosystem metabolism [Colello et al., 1998; Verma et al., 1992; Baker et al., 2003; Denning et al., 2003].
 The model's surface energy balance includes separate vegetation and ground temperatures that change according to net radiative input minus sensible and latent heat fluxes. The model's surface water balance is composed of canopy interception, ground ponding, soil, and snow balance equations that include inputs from precipitation or direct condensation, losses to evaporation, transpiration or sublimation, as well as interlayer exchanges, where appropriate. Runoff occurs as deep gravity drainage and lateral flows during periods of infiltration excess.
 The newest version (SiB3) includes a prognostic canopy air space for temperature, moisture, and CO2 [Vidale and Stöckli, 2005], and a 10-layer soil with explicit treatment of temperature and moisture based on the common land model [Dai et al., 2003]. Also added is a mixed plant canopy physiology used here to simulate canopy-atmosphere exchanges for C3 and C4 plants separately [Hanan et al., 2005], though they have the same soil water, radiation, and canopy air space environment.
 SiB3 was run for the period 1982–2003 with a 10-min time step on a 1° by 1° latitude/longitude grid encompassing the African continent. Model inputs for this implementation included soil texture [Tempel et al., 1996], vegetation type [DeFries et al., 1998; Hansen et al., 2000], and the fraction of C4 vegetation [Still et al., 2003]. To obtain an observationally consistent vegetation parameterization, as well as phenomenologically correct soil water limitation of photosynthesis in the Sahelian zone (open savannas to the south of the Sahara desert) it was necessary to extend the coverage of short, wooded C4 grassland further north to replace the C3 bare soil parameterization that was prescribed in the original vegetation data set. This change is consistent with field observations of vegetation cover in the region. Surface weather was prescribed on the basis of the National Centers for Environmental Prediction Reanalysis 2 (data made available online by National Oceanic and Atmospheric Administration Climate Diagnostics Center, Boulder, Colorado, 2003). However, the 6-hourly precipitation was adjusted to obtain total monthly rainfall that matches the remote sensing plus gauge data merged Tropical Rainfall Measuring Mission (Tropical Rainfall Measuring Mission Science Data and Information System (TSDIS) Interface Control Specification, 2006, available at ftp://disc2.nascom.nasa.gov/data/TRMM/Gridded/3B43_V6/) 3B43 rainfall product for 1998 through 2003, or Climate Research Unit (CRU) [Mitchell and Jones, 2005; New et al., 2000] monthly totals for 1982–1997 but adjusted to be consistent with the TRMM 3B43 monthly average spatial pattern from their period of overlap.
 The Simple Biosphere model contains algorithms, described by Sellers et al. , Los , and Los et al.  for estimating a suite of light interception, surface roughness, resistance, and physiological parameters based on remotely sensed vegetation index. In this work we used the twice monthly, ∼8 km NDVI made available by the Global Inventory Modeling and Mapping Studies (GIMMS) team [Pinzon, 2002; Pinzon et al., 2005; Tucker et al., 2006] derived from Advanced Very High Resolution Radiometer (AVHRR). While this NDVI data set contains corrections for satellite orbital drift, differing instrument calibrations, sensor degradation, and volcanic aerosols, we found large negative spikes of NDVI in many areas prone to cloud cover, and therefore replaced the lower 20% of NDVI of each biweek across years and at each ∼8-km grid cell with the mean of the upper eighty percent. Furthermore, we found unusual seasonal dynamics in NDVI even after this lower fifth replacement, and discovered that this seasonal pattern is strongly anticorrelated with pyrogenic or dust aerosol contamination as seen from MODerate Resolution Imaging Spectroradiometer (MODIS) Terra Level-3 global, monthly atmospheric aerosol optical thickness (MOD08_M3). Therefore we performed an ad hoc adjustment to the AVHRR NDVI data so that their biweekly average seasonality matches the average seasonality seen with a filled, 5 km MODIS NDVI product (A. Huete, personal communication, 2006) covering the 5-year period of 2000–2004. This approach retains seasonal and interannual variability in NDVI and hence vegetation structure and function, but removes much of the erroneous seasonality associated with aerosol and water vapor contamination.
 Since each of these data sets (weather, soil type, vegetation type, NDVI, etc.) contains a unique resolution or grid, we selected a land point mask based on the 0.0727 degree by 0.0727 degree NDVI data set and regridded the other data sets to match this land point mask using a bilinear interpolation as needed, with the exception of vegetation type which was regridded on the basis of assignment of a nearest neighbor without interpolation. These 0.0727 degree data sets were then upscaled to the 1° grid again with bilinear interpolation except for vegetation type which was assigned the most frequent occurrence.
 The core analysis in this paper quantifies limitation of canopy-scale photosynthesis by specific physiological and biophysical factors according to a procedure described briefly in words here and more formally in Appendix C. To put it simply, SiB calculates photosynthesis from the product of a physiologically defined rate with a biophysical effective area scaling that extends a single leaf's flux to the entire canopy. The model's diagnostics make it possible to calculate precisely how each biophysical and physiological factor limited photosynthesis in every time step and can be annually summed as performed here for a climatological analysis of what limits annual photosynthesis and drives its variability.
 On the physiological side, the leaf-scale gross photosynthetic rate (P) is reduced from a potential maximum (Ppot) according to multiplicative stress modifiers for temperature and soil water [Sellers et al., 1996c] as well as regulation of intercellular CO2 by stomatal closure [Collatz et al., 1991, 1992; Sellers et al., 1992, 1996c]. On the biophysical side, the canopy-scaling parameter (Π) is reduced from its potential maximum (Πmax). Appendix A describes biophysical and physiological terms, and Appendix B defines and illustrates stress controls. Combining the physiological and biophysical conditions, canopy-scale gross photosynthesis is obtained from PΠ (analogous to what many would define as gross primary productivity) and it has a potential maximum value of PpotΠmax.
 It is important to notice the model's joint limitation of canopy-scale photosynthesis by physiology and biophysics according to the product PΠ. A key consequence of this feature for the current analysis is that the magnitude of limitation by physiology depends on the biophysical state and vice versa. In other words, the marginal photosynthesis return from additional water depends on leaf area extent, and similarly the marginal photosynthesis return from additional leaves depends on the plant physiological state such as hydration. This highlights a point of departure between models that grow vegetation from photosynthesis and are inherently consistent with climate compared to models like SiB that prescribe vegetation dynamics from observations. The latter class of models risks inconsistency between prescribed vegetation and climate driven parameterizations of stress (Either the prescribed vegetation is wrong or the parameterization of the effects of climate is wrong, e.g., plant water stress parameterization is wrong and/or precipitation is wrong). In contrast, dynamic vegetation models force vegetation to match their particular parameterization of climate driven stresses even if either or both are wrong.
 One advantage to prescribing vegetation from observations is its ability to capture the real suppression of canopy-scale photosynthesis where vegetation is sparse in spite of conditions that may favor more extensive vegetation cover. This could arise where human or natural disturbances lower vegetation density, a process that is not captured in ecosystem process models that grow vegetation that is inherently consistent with a recent history of weather/climate. While prescribing vegetation can complicate attribution it also presents an opportunity for exposing apparent inconsistencies. For example, a case of very high stress coincident with a highly vegetated state would seem unsustainable from a biological point of view as chronic stress would cause vegetation dieback. Alternatively, the case of anomalously low stress with low vegetation is also inconsistent as vegetation is expected to more completely use available resources for photosynthetic gain. Such apparent inconsistencies are revealed by our analysis and help us to learn about potential model or model input errors, or missing processes (e.g., land use).