Interactive comment on “ Combining surface reanalysis and remote sensing data for monitoring evapotranspiration ”

After discussing Josh’s comments with Kevin, I agree that focusing the paper on improving operational LSM’s with the PT-JPL LEC component, which was the original intent of the paper, would make it shorter (particularly in the discussion section) and a cleaner read. To this end, when the paper passes to the editing phase, Kevin and I will remove portions that discuss the LES and LEI components and add material that strengthens the LEC portions of the paper. This should be a simple and straightforward process, as Kevin and I have already started the editing.


Introduction
Actual evapotranspiration (AET) is the only water flux that connects the atmospheric and land water cycles (Hartmann, 1994).In semi-arid regions of Africa, AET is the dominant component of the water budget and is highly variable across vegetation types (Ramier et al., 2009).These changes can have dramatic effects on the interannual and interdecadal variability of rainfall (Zeng and Neelin, 2000).A positive land-atmosphere feedback between vegetation and rainfall, for example, has been observed in the western Sahel.A decrease in rainfall and natural vegetation, leads to an increase in surface albedo and decrease in AET, which decreases moisture input to the west African Figures monsoon circulation, thus contributing to persistent droughts there (Zeng, 2003).Understanding this important moisture flux and its relationship to rainfall is particularly important in sub-Saharan Africa, where more than 70 % of the region's livelihood is rain-fed agriculture.Near real time AET estimates modulated by regional vegetation can be a powerful tool to facilitate this endeavor.
Literature reviews of continental to global scale AET modeling techniques can be found in Diak et al. (2004), Glenn et al. (2007), and Kalma et al. (2008).For the purposes of this paper, modeling approaches can be divided into two general categories.In the first category, remote sensing derived energy flux is combined with ancillary meteorological data using simple relationships to estimate near real-time estimates of AET as a residual of the energy balance.The remaining terms in the energy balance equation (ground heat flux -G, sensible heat flux -H, and net radiation-R N ) can be determined from empirical relationships with remotely sensed surface temperature and vegetation.Remote sensing models which derive flux directly from empirical relationships of meteorological and remote sensing data are not considered in this category.
Algorithms which produce global estimates of AET as a residual of the energy balance are described in Nishida et al. (2003), Leuning et al. (2008), Mu et al. (2007a), and Fisher et al. (2008).These models have been recently used in sub-Saharan Africa to estimate water use efficiency for arid rangelands (Palmer and Yanusa, 2011) and extrapolate biological nitrogen deposition from wildfires (Chen et al., 2010).One of the greatest limitations of AET estimates inferred from satellite data is that they often use once-a-day satellite overpasses and the evaporative fraction to extrapolate AET estimates over longer time scales.The evaporative fraction is the ratio of AET to R N -G, which can be used to scale AET over the day (Mintz and Serafini, 1992).Early reports showed errors in AET using this method to be 20-40 % from surface measurements (Nichols and Cuenca, 1993).Improved satellite data and parameterization have reduced these errors, but persistent cloud cover and overestimation of net radiation due to poor cloud filtering continue to make the application of this technique difficult in the tropics.Figures The second category of models, termed Land Surface Models (LSMs), can provide near real-time, continuous and global estimates of AET using sophisticated processbased techniques driven by assimilated ground, satellite, and reanalysis data (Rodell et al., 2004).In the case of this paper, LSM's are run offline (i.e.uncoupled to the atmosphere).AET from these models is driven primarily by precipitation with latent heat being solved as a residual of the water balance and corrected using assimilated fields of the energy terms.LSM's overcome the temporal limitations associated with remote sensing based estimates of AET, because LSMs are driven by reanalysis fields of surface climate.Land surface flux has been compared from several LSM's at the global level: Project for Intercomparison of Land-Surface Parameterization Schemes (Henderson-Sellers et al., 1995), the Global Soil Wetness project (Dirmeyer et al., 2006), and a remote sensing based and LSM inter-comparison (Jim énez et al., 2011).The AMMA Land-Surface Model Inter-comparison Project (Boone et al., 2009) includes LSM comparisons in Africa in order to better understand the impact of land surface flux on the West African Monsoon.Two of the major drawbacks of LSMs is that their strong theoretical framework often makes them less robust, due to conspiring factors attributed to a multitude of data types (most importantly R N ) and empirically-based parameters (Rosero et al., 2009).
The objective of this paper is to identify and integrate components from a remote sensing based and LSM approach to improve estimates of AET.The integration of near real-time remote sensing AET models with LSM AET can help maximize the benefits of both approaches.For example, land surface temperature derived from remote sensing data are used to recalibrate AET estimates from a soil vegetation atmosphere transfer model in Olioso et al. (1999) and from a residual energy balance model in Boni et al. (2001).AET estimates in this study are improved further downstream in the modeling process by direct insertion, whereby the AET component of one model, in this case one that drives canopy AET with regular remotely sensed vegetation, is substituted with canopy AET from an LSM.The hybrid model is developed through a series of exploratory exercises, given the fundamentally different approaches each Introduction

Conclusions References
Tables Figures

Back Close
Full The Berkeley model (Fisher et al., 2008) is selected for integration with an LSM.This model is driven primarily by vegetation controls on Rn.The Berkeley model is chosen, because it has been shown to outperform other AET models in the tropics when compared with eddy covariance flux tower data and it can be readily applied over large areas using remote sensing and surface reanalysis data.In addition, it requires no calibration (optimization) or spin-up.A major drawback to this model is that the soil and wet evaporation components are highly empirical power functions which are driven by specific humidity, a notoriously uncertain parameter in surface reanalysis datasets.Therefore, improvements in the Berkeley model can be made by further re-parameterizing wet canopy and soil evaporation.The LSM used is the NCEP, Oregon State University (OSU), Air Force, and Hydrology Research Laboratory at NWS (NOAH) model (Chen et al., 1996).NOAH is a community model which has undergone several intermodel comparisons and ground-based validations, leading to vast improvements in its parameterization since its inception at OSU.It is comparable to other LSMs when modeling latent heat flux (LE) driven by AMMA Land Surface Model Intercomparison Project forcing and parameters over large areas in western Africa (Boone et al., 2009) and generally performs better than other LSMs under semi-arid (sub-Saharan Africa) conditions (Hogue et al., 2005).The major limitation of the model is that the canopy (transpiration) component of AET is driven by long-term monthly averages of the Normalized Difference Vegetation Index (NDVI).Therefore, further improvements in NOAH AET in semi-arid regions can be made by including a dynamic vegetation component.
In the first phase of the analysis, the Berkeley model is compared with flux tower data to identify possible sources of error in model parameterization and to judge the performance of the model across various land cover types in sub-Saharan Africa.A sensitivity analysis is then performed on the model inputs to guide model re-parameterization.For example, identifying the parameters that represent a large source of model error or insensitivity would justify re-parameterization or substitution of the parameter with Introduction

Conclusions References
Tables Figures

Back Close
Full long-term mean data.This type of exploratory analysis facilitates the development of a hybrid model, which combines the canopy component of the Berkeley model and wet canopy and soil evaporation components of the NOAH LSM.In the final phase of the analysis, the combined model is evaluated using the flux tower data.

The Berkeley model for actual evapotranspiration
The Berkeley model is a modified version of the original Priestley and Taylor (1972) model for potential evapotranspiration (PET).A complete description of the model and bibliography can be found in Fisher et al. (2008).The Priestley-Taylor formulation for PET reduces the advection term in the original Penman (1948) formulation to a simple coefficient (α).The Priestley-Taylor model is therefore driven primarily by net radiation and tends to perform best in humid areas and worse in moisture limited areas (DehghaniSanij et al., 2004).The Berkeley model retains the original α (1.26) and instead uses six parameters to modify PET: where f SM , f WET , f C , f G , f T , and f M are the soil moisture constraint, relative surface wetness, fractional total vegetation cover, green canopy fraction, plant temperature constraint, and the plant moisture constraint respectively.Table 1 lists each parameter and the relevant equation used in the computation of latent heat (LE).
The equations have been arranged to express LE in terms of its three components: bare soil evaporation (LE S ), transpiration (LE C ), and wet surface evaporation (LE I ).The Introduction

Conclusions References
Tables Figures

Back Close
Full psychometric constant (γ: 0.066 kPa • C −1 ), slope of the saturation-to-vapor pressure curve (∆), R N , and G make up the Priestley-Taylor formulation for PET.Net radiation and G are readily available from surface climate reanalysis or remotely sensed spectral indices.Equation ( 1) is a modified version of the Fisher et al. (2008) model in which the soil and canopy contribution to R N are assumed to be a function of leaf area index and f C according to Beer's Law (Kelliher et al., 1995).Here, R N -G is discretized by NDVI, which has been shown to be highly correlated with f C (Sellers, 1987).Fractional total vegetation cover is expressed as a linear function of NDVI, which is determined from the red and near infrared bands of standard multi-spectral remote sensing platforms.
The transpiration component of LE has the greatest number of constraints.The green canopy fraction is a biophysical constraint expressed as the ratio of the fraction of PAR absorbed by the green vegetation (f APAR ) to the total amount of PAR absorbed by the canopy (f C ). Photosynthetically Active Radiation (PAR) absorbed by green vegetation is a linear function of the Enhanced Vegetation Index EVI.The Enhanced Vegetation Index is another remotely sensed spectral index and is more sensitive to the chlorophyll content of vegetation than NDVI (Gao et al., 2000).The plant temperature constraint is a physiological parameter and assumes that vegetation photosynthesizes at an accelerated rate until an optimal temperature (T OPT ) is achieved, after which efficiency decreases (June et al., 2004).The optimal temperature is determined over the entire available time series and is analogous to "relative greenness" in the remote sensing literature.It occurs during the primary growing season, when the daytime temperature (T MAX ) at which plant investment in light energy (f APAR ) and the availability of light (PAR) are high, and the vapor pressure deficit (VPD) is low.Surface temperature data is readily available from reanalysis data.The plant moisture constraint is the ratio of f APAR to maximum f APAR over the available time series.It assumes that the amount of light a plant absorbs varies with moisture availability: f APAR decreases only when the plant is stressed, thus lowering transpiration when PAR is high.The effect of this constraint on LE C is minimal, unless the plant is suffering from extreme moisture stress.Introduction

Conclusions References
Tables Figures

Back Close
Full The interception and soil components are governed by relative humidity (RH), which can be determined from surface pressure and specific humidity reanalysis.It is assumed that bare soil evaporates water at the potential rate (PET), provided that the soil is saturated (Bouchet, 1963).When the soil is not saturated, relative humidity above the surface decreases and latent heat decreases due to lower soil water storage.The soil moisture constraint is a power function of RH and the vapor pressure deficit (VPD), which acts to reduce the supply of water to the atmosphere as the soil dries.Midday relative humidity and VPD are used, because these values show better results in the original paper than daily averages.The theoretical justification is that the coupling between soil moisture and atmospheric humidity is strongest in the daytime when vertical mixing is high.Similarly, f WET is a power function of relative humidity and indicates the probability that the surface is wet; when relative humidity is 100 %, the soil and canopy are completely wetted and evaporate moisture at the potential rate.

NOAH Model for actual evapotranspiration
The earliest version of the NOAH-LSM can be traced back to Chen et al. (1996), who integrated an explicit canopy component with a simple soil water balance model developed at OSU (Pan and Mahrt, 1987).The model has undergone several revisions since that time, including improvements to bare soil evaporation estimates with the introduction of skin temperature and a dynamic soil moisture component (Betts et al., 1997) and to transpiration estimates with the introduction of a monthly fractional total vegetation cover climatology (Chen and Dudhia, 2001).The NOAH model also includes three components of latent heat.Unlike the Berkeley model, where constraints are driven primarily by changes in vegetation and atmospheric humidity, the NOAH model takes a water balance approach driven primarily by precipitation.Each energy term (LE S , LE C , and LE I ) is summed after constraints on PET have been computed.Potential evapotranspiration in the NOAH model (E P ) is an energy version of Penman (1948) with a detailed description in Mahrt and Ek (1984).Introduction

Conclusions References
Tables Figures

Back Close
Full where LE (Wm −2 ) is the latent heat flux (actual evapotranspiration), β is the fraction of total soil moisture not used by the canopy (Mahfouf and Noilhan, 1991), B C is a function of atmospheric and stomatal resistance (Jacquemin and Noilhan, 1990), W C is the water holding capacity of the canopy defined as the residual of water balance terms, and S is the maximum water holding capacity of the canopy (calibrated constant).As in the original Berkeley formulation, f C is a function of LAI.It is computed from a 0.15 • resolution climatology of AVHRR NDVI (Gutman and Ignatov, 1998).Soil moisture availability (β) includes a dynamic soil moisture component constrained by the wilting point and the field capacity of the soil, both of which are functions of soil texture.The constraints are elaborated upon in Table 2, while a more detailed description can be found in Chen and Dudhia (2001).The NOAH-LSM in this paper (version 2.7.1) is driven uncoupled to the atmosphere by the 0.25 • resolution Global Land Data Assimilation System (GLDAS) forcing and paremeters (Rodell et al., 2004), enabled by the Land Information System (Kumar et al., 2006).Details on GLDAS forcing data can be found on NASA's Hydrology Data and Information Services Center webpage (http://disc.sci.gsfc.nasa.gov/hydrology),while details on the parameterization datasets used can be found at http://ldas.gsfc.nasa.gov/gldas/GLDASvegetation.php.Forcing data will be referred to as GLDAS for the remainder of the paper, while GNOAH will refer to NOAH-LSM output driven by GLDAS forcing and parameterization.
The hybrid model developed in this paper sums the LE I and LE S components from the GNOAH with LE C from the Berkeley model.Introduction

Conclusions References
Tables Figures

Back Close
Full

Data processing and handling
The Berkeley model is run globally at a quasi-0.05The Global Land Data Assimilation System uses 0.25 • resolution climatological forcing data for this exercise.GLDAS uses NOAA/GDAS atmospheric fields, Climate Prediction Center Merged Analysis of Precipitation fields, and observation-driven shortwave and longwave radiation using the Air Force Weather Agency's AGRicultural ME-Teorological modeling system.These data are produced at 3-hourly intervals and aggregated to a monthly time step.The latent heat simulation from the NOAH-LSM is at 0.25 • resolution and resampled using the "nearest neighbor" approach to 0.05 • resolution for comparison purposes with the Berkeley model.The choice of resampling assumes that spatial heterogeneity is driven primarily by vegetation.
The NDVI and EVI 0.05 • resolution vegetation index inputs come from the Moderate Resolution Imaging Spectroradiometer (MODIS) on board the Earth Observing System-Terra platform (Huete et al., 2002).The vegetation indices are composited over 16 day periods.The NDVI and EVI data are distributed by NASA with complementary information describing the reliability and quality of the data.Pixels are masked for low data reliability, aerosols, clouds, and water bodies.A piecewise weighted least squares regression filter (Swets et al., 1999) is applied to the datasets to further reduce atmospheric interference on the vegetation signal.Introduction

Conclusions References
Tables Figures

Back Close
Full

Statistical analysis
The analysis is done in three phases: (1) evaluation of the Berkeley model using station data and comparison of ground inputs with surface reanalysis data, (2) sensitivity and residual analysis of the Berkeley model, and (3) development of a combined model and inter-model comparison.The observed data are from three sources: the African Monsoon Multidisciplinary Analysis (AMMA) programme (www.amma-international.org), CarboAfrica (www.carboafrica.net)and Fluxnet (www.fluxdata.org).Table 3 lists the field sites used to estimate micro-climate and energy fluxes and includes the station I.D. and name, country, geographic coordinates, the period data is obtained, ecosystem type using the International Geosphere-Biosphere Programme (IGBP) naming convention, and climatology.Five of the sites (BW-Ma1, CG-Euc, NE-Waf, NE-Wam, and ZA-Kru) include halfhourly measurements of surface air temperature ( • C), incoming longwave and shortwave radiation (W m −2 ), outgoing longwave and shortwave radiation (W m −2 ), relative humidity (%), precipitation (mm), and latent heat (W m −2 ).NE-Wam and NE-Waf were in close proximity and shared a rainfall gauge between them (Ramier et al., 2009).LE was measured using the eddy covariance method (Baldocchi et al., 1988).With this technique, latent heat flux is determined by correlating changes in water vapor concentration at the surface and at height measured using hygrometers and a sonic anemometer.The eddy covariance method is the most widely used method for measur-Introduction

Conclusions References
Tables Figures

Back Close
Full neural networks (Papale and Valentini, 2003), however these data are not available, so persistent data gaps at some of the sites do exist.The purpose of the first phase of the analysis is to extend comparison of the model with micrometeorological data to semi-arid zones in sub-Saharan Africa and to qualitatively assess the performance of LE parameterizations (transpiration, wet canopy evaporation, and bare soil evaporation).The comparison also helped to identify sources of error in the surface reanalysis to aid in the interpretation of the sensitivity and residual analysis.Near infrared and red reflectance are not available from the stations, so NDVI and EVI from MODIS are used to drive the local model run instead.Point to sensor scale comparison are performed by taking the average of remote sensing and surface reanalysis inputs over adjacent grid cells to the cell corresponding to the latitude and longitude of the stations.The averaging covers a footprint typically larger than that of a flux tower (<1000 m), however the averaging is deemed necessary, because many of the stations fall on the border of two or more adjacent pixels.Each component of LE is computed and compared visually with observed LE.The coefficient of determination (R 2 ) and root mean squared error (RMSE) are the primary metrics used for comparison of monthly Fisher LE to observed LE.It is assumed that LE C is the largest and the primary seasonal component of LE.Wet surface evaporation is typically the smallest component and is assumed negligible.Soil evaporation tends to track well with precipitation (Nagler et al., 2007), so precipitation is used as a proxy for LE S evaluation.
Each GLDAS input (R N , T , specific humidity -q, and surface pressure -p) is plotted against the micrometeorological data.The bias, RMSE, and correlation coefficient is calculated for each.
The purpose of the second phase of the analysis is to identify those GLDAS inputs that could potentially lead to large errors in the Berkeley model.A sensitivity analysis is used to identify which GLDAS and MODIS inputs the Berkeley model is most sensitive to.The outcome of interest in the analysis is: the model is sensitive to a particular input that is also a large source of error, necessitating calibration.Minor improvements during model calibration can also be made in the case that the input was erroneous Introduction

Conclusions References
Tables Figures

Back Close
Full but not an important contributor to AET variability by substituting the input with the long-term mean.The residuals (observed LE minus Berkeley modeled LE) are plotted against each input variable to identify the inputs that lead to the greatest errors in modeled LE.Two cases of the sensitivity analysis are performed.In the unconditional case, all input variables (R N , T , q, p, NDVI, and EVI) are constrained to mean values except for the test variable (Haan, 2002).The test variable is perturbed 10 000 times between ±3 σ.The model output LE is regressed against the test input and the slope of the relationship is computed to determine the relative weight of each input to the output.
In the conditional case, the test input or a combination of test inputs is constrained to the mean and the remaining variables are allowed to run freely.This case tests The monthly GLDAS data used to drive the Berkeley model (R N , T MAX , q, and p) are plotted against observed data for ZA-Kru in Fig. 2. It is expected that GLDAS pressure and specific humidity data correlates well with observed relative humidity.Actual specific humidity and surface pressure are not used for direct comparison, because these data are not available for most of the sites.GLDAS R N , T , and q showed good agreement with observed data.The inverse relationship between p and relatively humidity is present, but obscured by outliers above 98.0 kPa (Fig. 2d).The outliers are from the first year of the reanalysis when the GDAS reanalysis fields are used exclusively.BW-Ma1 is the only other site that showed such a relationship.Table 4 summarizes the results for all of the stations with sufficient field data for the comparison.Some of the relationships are clearly non-linear, however statistics from a linear fit are chosen for comparison purposes between inputs and across sites.Temperature from the GLDAS dataset consistently shows high correlations across all the sites.Specific humidity is the second best predicted input.The correlation between specific humidity from the reanalysis and observed relative humidity are highest at the two driest sites (NE-Waf and NE-Wam) and lowest (negative) at the most humid site (CG-Euc), which could be due to the low variability in relative humidity and relatively larger homogenous area at the dry sites.GLDAS R N was the next best predicted model input.The longest time series (CG-Euc and ZA-Kru) had the highest correlations to observed R N , but with large biases.The longer time series and presence of outliers could therefore be inflating the correlations.In either case, the general pattern (seasonality) in R N is being predicted well for these sites, but assuming all inputs are weighted equally, overestimates in modeled LE are likely to occur.Conversely, the weak and negative relationships between GLDAS R N and observed R N at the two driest sites lead to underestimates of R N in modeled LE.Reanalysis pressure by far has the poorest correlations with observed data, as high biases are seen across all the sites.Introduction

Conclusions References
Tables Figures

Back Close
Full

Sensitivity analysis
The comparison in Sect.3.1 revealed that Fisher LE C performs the best of the three LE components and that GLDAS temperature, which is used exclusively in the transpiration component of the model, is the best performing input with available observed data.This exploratory exercise indicates that Fisher LE I and LE S may be good candidates for re-parameterization.The second phase of the analysis shows the relative contribution of inputs to model error.The results of the unconditional case of the sensitivity analysis are shown in The ratio of the standard deviation of the model when the test variable(s) are kept at their mean to the standard deviation of the model if all inputs are allowed to run freely represent the relative sensitivity of the model to that input or combination of inputs.The evaluation of all possible combinations reveals important synergistic effects that may enhance or suppress the predictability of LE.The various combinations are plotted against modeled LE with all inputs allowed to run freely and R is computed as well, but this chart was not included, because it did not contribute new information.Ratios less than one indicate a reduction in predictive power, while ratios greater than one reveals Introduction

Conclusions References
Tables Figures

Back Close
Full dampening effects.As shown in Fig. 3, keeping q (input 2) and EVI (input 6) at their mean has the largest impact on modeled LE, while T (input 3) and NDVI (input 5) act to suppress modeled LE.EVI, which is used to compute the green fraction, explains nearly 60 % of the model variability alone (combination 1, 2, 3, 4, 5).Keeping NDVI and R N at their mean (combination 4, 5) reveals no noticeable drop in modeled LE.The model inputs for EVI and q are plotted against a time series of modeled LE (not shown).EVI tends to capture the troughs in ET (dry season) when transpiration is significantly larger than soil evaporation and specific humidity is better at capturing the peaks in ET (wet season) when soil evaporation is a much larger component of the budget.

Hybrid model
The was burned in July 2008, so that no grass remained until the site recovered in mid-October (data not available for the analysis).Grass was absent and the soil humidity is less than 2 % by volume, so H is the dominant energy flux.The combined model outperforms the GNOAH model for the majority of sites and equally as well for the remaining sites.The greatest improvements are made between the Fisher and combined model, particularly in the reduction in RMSE.These improvements are greatest at the sites with dense vegetation (CG-Euc and BW-Ma1).At these sites, the combined model tends to underestimate the peaks and overestimate the troughs.At the driest sites (NE-Wam, NE-Waf, and SD-Dem), all three models grossly underestimate peak LE, however the combined model performs the best.The major limitation of GNOAH, namely lack of a dynamic vegetation component, is reflected on the time series at the ZA-Kru site.The smooth LE signal produced by the GNOAH model does not capture the natural variability of this semi-arid site.The Berkeley model represents the other extreme, as it overestimates the variability in observed LE.The combined model is a compromise between the two.All three models tend to perform poorest for CG-Tch, missing the peak in 2007 and the secondary peak in 2008.

Discussion
The study represents an initial attempt to use a suite of new flux tower data in sub-Saharan Africa to improve AET estimation using remote sensing and surface reanalysis data for regular near real-time continental scale monitoring.The integration of the evaporation from GNOAH with the Fisher transpiration further improves the correlations and reduces the RMSE for both humid and semi-arid sites.The fractional total vegetation cover of the GNOAH model is the single most important variable controlling transpiration.This is consistent with other findings that showed LAI was a strong control on the ratio of AET/PET at savanna sites (Williams et al., 2009).The use of climatology to derive this component often suffers in semi-arid climates where variability in vegetation can be high.This is most apparent at the savanna sites, where the GNOAH Introduction

Conclusions References
Tables Figures

Back Close
Full The greatest improvements in modeled LE using the combined model are at the sites with dense vegetation classes.The predictability of LE for the various vegetation using the combined model can be attributed in part to light and water limitation (Mu et al., 2007b).Dense vegetation in the wet tropics is light limited because soil is sufficiently wetted throughout the year, so one would expect that LE is driven primarily by R N (Priestley-Taylor formulation for PET).Sparse vegetation in the dry tropics on the other hand is water limited (humidity driven), because light is sufficient to maximize stomatal conductance throughout the year and plants close their stomata in response to dry conditions to conserve water and avoid stress.The poor performance of GLDAS R N and q at the dry and wet sites respectively further augments this relationship.The use of GNOAH wet canopy evaporation (results not shown) improves correlation with and reduces the RMSE in observed data, though not as significantly as soil evaporation, reflecting its lower importance to total AET.The greatest improvement is seen at the humid site (CG-Euc), which is a challenging region to model, albeit less representative of land-cover in sub-Saharan Africa.The modest improvements can be attributed to the substitution of poor GLDAS specific humidity data and poor parameterization of LE S with an improved LE S parameterization driven by a more reliable input variable (precipitation).The small improvements made at the drier sites using the combined model are more difficult to justify, because GLDAS showed higher correlations with observed relative humidity and the performance of the Fisher LE S component is relatively better at these sites.Erroneous GLDAS R N appear to be a culprit at the drier sites, but the insensitivity of LE to changes in PET formulization and lower sensitivity of the Berkeley model to R N , suggest that the use of GNOAH LE S may make some limited contribution.Although the model is less sensitive to pressure, erroneous values can lead to a 16 % difference between the Fisher and the combined model.This is most Introduction

Conclusions References
Tables Figures

Back Close
Full apparent in 2000 at ZA-Kru, where pressure values from the GDAS fields were several kPA higher than for other years.Improvements in GLDAS specific humidity and pressure fields will undoubtedly reduce the variability in future studies that use the Berkeley model driven by surface reanalysis.Vapor pressure could be used, for example, from a study that recently determined it from MODIS estimates of dew-point temperature in Korea (Ryu et al., 2008).In the meantime, formulations for f WET and f SM involving more certain surface reanalysis inputs than specific humidity/pressure or precipitation, such as maximum and minimum daily surface temperatures, could be used as a substitute.Idso et al. (1975) and more recently Wang et al. (2007) have suggested that the diurnal temperature range is a good indicator of soil moisture status.Under moderately dry to wet conditions, minimum (nighttime) temperature is highly correlated with observed vapor pressure, while midday temperature is indicative of saturation vapor pressure.A formulation for LE that combines temperature driven soil and canopy components can be a powerful tool at both point and grid cell resolution.
The large discrepancy of the models from peak AET at the driest sites could be due to several factors.Although energy balance closure problems are typically associated with errors in measuring turbulent fluxes, measurement error in R N remains a possibility.Ramier et al. (2009) and Ardo et al. (2008) report only modest daily average energy balance biases of 14.2 W m −2 and 16.8 W m −2 for NE-Wam and SD-Dem respectively, which does not account for the model underestimation of peak LE by more than 30 %. Another possibility in LE underestimates at NE-Wam and NE-Waf can be the result of unexpectedly low average daytime net radiation during the wet season that appear to result from large increases in longwave outgoing minus longwave incoming (∆LW) near sunset.The increase in absolute ∆LW can be attributed to increases in soil moisture (lower sensible heat and longwave outgoing) and cloud cover (increased longwave incoming) that is typical during the wet season (Ramier et al., 2009).These factors could lead to a drop in LE of nearly 17 W m −2 , assuming a one standard deviation drop in R N from the sensitivity analysis, which still does not account for the underestimation of Introduction

Conclusions References
Tables Figures

Back Close
Full In dry areas, it was expected that the advection term, which is included in the Penman formulation for PET, would improve the hybrid model performance over Priestley-Taylor PET.The Penman PET and its input parameters are not available from GLDAS, so a simplified version of Penman PET described in Allen et al. (1998) was evaluated with no improvement at the drier sites.Priestley-Taylor PET may be better at large (satellite) scales, however, because PET at that scale is driven primarily by the expansion of the convective boundary layer, which in turn is driven by R N (Raupach, 2000).Future calibration can therefore include integration of Priestly-Taylor PET into the GNOAH model.
The residual analysis is less revealing than the sensitivity analysis, as the relationship between GLDAS data and model error are inconsistent across stations.This can be due to a number of factors, including strong multi-collinearity which obscures the effects of independent variables, and the inadequate length of many of the time series.Some results of the residual analysis are worth discussion.For ZA-Kru, the longest time series, heteroskedasticity is observed for all the variables except pressure.This is not unexpected given the non-linearity of the model (larger values in the input can lead to greater variability in the errors).The residual analysis for pressure revealed an inconsistency between the first year and the remaining years of the model run, Introduction

Conclusions References
Tables Figures

Back Close
Full suggesting that improvements in pressure can be made by refining GLDAS synthesis across datasets.Given the dependence of the Berkeley model on R N from the literature it was initially suspected that PET could be an important source of model error.The Penman version of PET described in Allen et al. (1998) and the Priestley-Taylor formulation are compared against the model residuals and no significant relationship are observed for either.This further corroborates the relatively lower sensitivity of the Berkeley model to R N than vegetation in tropics.
It is difficult to make a more detailed assessment on the performance of the combined model, however, because the comparison is done using point data scaled up to moderate and coarse data.Ideally, the satellite and surface reanalysis scale data should capture the average of a grid cell corresponding to the station used for validation.This assumption is problematic and scaling flux tower data to coarser scales therefore remains an active area of research.The greatest challenge is finding flux towers that lie within homogenous and flat terrain corresponding to the spatial resolution of satellite or surface reanalysis data.If the fetch of the flux tower includes heterogeneous and/or rough terrain, eddy formation can be highly variable and may not be consistent with the areal average (Baldocchi et al., 1988).A possible solution can be to evaluate these models using several flux towers within a grid cell, representing the various land cover types and taking a weighted average to compare with coarser scale data.Another difficulty arises in the nature of the input variables.Many input variables, such as NDVI, scale non-linearly (McCabe and Wood, 2006).Coarser scale inputs capture regional estimates of ET well, so the model can be further validated using larger scale surrogates of ET.For example, the combined model is recently used to develop a crop stress index that successfully tracked multi-year district-level crop yield and identified historical food insecure hotspots in Kenya (Marshall et al., 2011a).Runoff data, which can be used with precipitation to validate the model using a water balance approach, is extremely limited after the year 2000 for basins in sub-Saharan Africa.These data are being used to evaluate the model in a future study which extends the ET time series back to 1981 (Marshall et al., 2011b).Introduction

Conclusions References
Tables Figures

Back Close
Full The quantity and quality of field data remains an important obstacle in developing an AET model for sub-Saharan Africa.Several sites that are part of CarboAfrica and AMMA have not undergone any rigorous pre-processing.CG-Tch, for example, had frequent failures with the sonic anemometer between December 2006 and July 2007, which contributed to a lower monthly average, likely reflecting the models' inability to capture the peak LE at this time.Other stations which can help to calibrate and validate AET models for other landcover classes and over larger areas are just coming online.AMMA, for example, has deployed a network of stations in West Africa.At least one of the sites has been scaled to grid resolution using readings from multiple locations within the grid and land cover fractions.At the time this paper is written, the data from those stations included all of the necessary inputs to run the Berkeley model, but did not include LE necessary for model evaluation.In the next few years, when stations have been online long enough, the process of model development should be revisited.

Conclusions
The need to understand energy and water fluxes in data sparse regions of the world is an important field of research that demands the use of cost-effective and efficient modeling approaches.In this paper, a new approach to modeling AET, an important energy and moisture flux, has been introduced.For the first time, an AET model has been evaluated using the eddy covariance technique over areas representing major land-cover classes in sub-Saharan Africa.The paper highlights many of the obstacles and limitations of such an analysis.Even with these shortcomings, a model, which combines LE components driven by remote sensing vegetation and precipitation reanalysis has been developed that provides better AET estimates than the individual models.Perhaps the most important contribution of this work is the assessment of the two models across diverse land-cover types and climatic zones.and the use of specific humidity reanalysis to model soil evaporation in humid areas.R N reanalysis is of limited use in semi-arid regions, though its contribution to overall model error is less than vegetation and specific humidity.Future work will evaluate the model using historical forcing and test data.This will allow researchers for the first time to make reliable estimates of the surface water budget and to conduct water balance and water resource assessment type studies across sub-Saharan Africa.In the meantime, the current model is being used in conjunction with precipitation indices to enhance crop monitoring in sub-Saharan Africa.Introduction

Conclusions References
Tables Figures

Back Close
Full  Full  Full  Full  Full Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | model takes to estimating AET and the equally diverse strengths and weaknesses in parameterization.
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Fig.1.There is no surrogate for wet canopy evaporation and its low variability revealed that it is merely a background signal in total LE.The magnitude and timing of modeled LE for the five stations with sufficient data to drive the Berkeley model are captured well at three of the five sites.At the driest sites (NE-Waf and NE-Wam), the timing is captured well, but the magnitude of peak LE from the Berkeley model is less than 50 % of actual peak LE.NE-Wam was not included in the figurebecausethe results were similar to NE-Waf.The average R 2 and RMSE across the five sites (BW-Ma1, CG-Euc, NE-Waf, NE-Wam, amd ZA-Kru) are 0.78 and 36.81W m −2 , respectively.The fallow site (Ne-Waf) has the highest R 2 (0.93), reflecting the ability of the model to capture timing, but also the highest RMSE (72.47 W m −2 ).The wooded site had the lowest R 2 (0.65).ZA-Kru had the lowest RMSE (13.75 W m −2 ), due in part to data gaps during peak ET (wet season).During this season, sensors are frequently turned off to prevent damage from lightning and power surges.At the driest sites, the model tends to underestimate wet season LE (peaks) and overestimate dry season LE (troughs).Latent heat from the eucalyptus plantation (CG-Euc: included inFisher et al., 2009)  is overestimated fairly consistently throughout the year.The time series at the two driest sites (NE-Wam and NE-Waf) are insufficient to discern interannual patterns, but given the sparse vegetation and lack of an advection term in the PET approach used, it is expected that the model underestimates peak latent heat for both wet and dry years.The canopy component follows the timing of total LE for all the sites, as expected, given the dominance of transpiration to total LE.Without actual transpiration data, however, it is impossible to determine the representativeness of Fisher LE C magnitudes.Assuming a lagged relationship between soil evaporation and rainfall, LE S performs well at the sites with low vegetation density and poorly at the sites with high vegetation density.Therefore, based on the limited field data that is available, it appears that Fisher LE C is at least capturing the timing of observed LE well, while the Fisher LE S parameterization tends to suffer in vegetated (humid) areas.
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | low dependence of the Berkeley model on surface pressure and spurious GNOAH pressure data suggests that minor improvements in modeled ET could be made by substituting long-term means in the GLDAS pressure data.The relatively poor parameterization of Fisher LE S and LE I and conflicting results of the GLDAS specific humidity data in humid areas and strong dependence of the Berkeley model on specific humidity, suggests that modest improvements in LE can be made by substituting LE S and LE I driven by specific humidity in the Berkeley model with a parameterization that uses a surrogate, such as precipitation or nighttime temperature.The GNOAH LE S and LE I components are driven primarily by precipitation, are well conceived in the literature, and therefore served as parameter substitutes in the Berkeley model.Time series of observed LE and modeled LE using Fisher, GNOAH, and the combined model (Fisher LE C + GNOAH LE I,S ) for six of the eight stations are shown in Fig. 4. Table 6 includes goodness-of-fit statistics for six of the eight stations: ZM-Mon and NE-Wam are omitted, because the former had less than one year of data and the latter showed no noticeable difference from NE-Waf.Two of the stations (BW-Ma1 and CG-Tch) showed obvious bad (flat) LE data at the beginning and end of the time series when the sensors were just coming online or being neglected, so statistics are computed after these points were omitted.The last month of CG-Tch was omitted as well, because the site Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | model shows a rather smooth annual signal.The introduction of a dynamic component dramatically improves the correlation between observed and modeled LE.The Fisher and GNOAH model use two different satellite sensors (AVHRR and MODIS), and this undoubtedly plays a role in the results as well.
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |LE.The largest contributor to the underestimation of LE is uncertainty in vegetation dynamics: MODIS resolution data is too coarse to adequately capture the vegetation in the presence of heterogeneous terrain, given the strong dependence of the model on EVI and the sparse vegetation at these sites.Even with gross underestimation of R N and EVI, the comparison of observed data with the Berkeley model reveals that LE remains difficult to model in semi-arid areas.The primary purpose of this paper is to integrate the existing components of the Berkeley model with an LSM.Therefore specific improvements to Berkeley model LE C , such as the use of the EVI ratio to formulate f C as inMu et al. (2007b), calibration of the S in the GNOAH model for semi-arid regions, or the use of different forcing and parameterization data, should be addressed in future studies to reduce uncertainties in semi-arid regions.
• resolution daily time step from 2000-2008 for sub-Saharan Africa and is aggregated to a monthly time step for visualization and comparison purposes with field data.Daily values were aggregated to monthly values for several reasons: (1) ground heat flux was not calculated in the Berkeley model (ground heat flux is near zero in warm regions at a monthly time step), (2) daily station flux data was often spurious or missing, (3) the vegetation data used (see below) is extremely noisy at a daily time step and (4) dekadal to monthly timesteps are adequate to address research questions in food security monitoring and other continental scale studies.

Table 5 .
The results of this test are shown for ZA-Kru and are representative of the sites.The slope and intercept are generated from linear fits of model output data to input data in normalized space.The slope of the fit therefore represents the average expected increase in LE due to a standard deviation increase in the test variable.The model is most sensitive to EVI and q, because these variables have the largest slopes, while the model is least sensitive to p. Increases in NDVI and T act to suppress LE, as indicated by the negative slopes.NDVI in the Berkeley model is in the denominator of f G , since it is used to indicate overall absorption of incoming radiation by the canopy.As NDVI increases, f G decreases, given NDVI is more variable than EVI.Temperature is explicitly handled in the plant temperature constraint function-increases in temperature above the optimum temperature leads to increases in this constraint and lower LE.The results of the unconditional case are shown in Fig.3.Six input variables yield sixty-two possible combinations or model realizations.

Table 2 .
Model parameters and equations for the Noah Model, where k is Beer's Law extinction coefficient, LAI is the leaf area index, r is the atmospheric resistance, R C is the stomatal resistance, C h is the surface exchange coefficient for heat and moisture, Θ 1 is the soil moisture in the top soil layer at a given timestep, Θ W is the wilting point, Θ REF is the field capacity, and f is a scaling constant typically equal to 1 or 2. The change in water holding capacity is defined by P (precipitation), D (drainage) and LE C (in mass units).

Table 3 .
Eight micrometeorological stations throughout sub-Saharan Africa.Ecosystems are identified with the IGBP convention: croplands/natural vegetation mosaic (CRO), evergreen broadleaf forest (EBF), open shrublands (OSH), savanna (SAV), and woody savanna (WSA).P and T A are the annual total precipitation and average air temperature respectively.

Table 6 .
Summary results from the comparison of observed monthly LE to the Fisher, Noah, and hybrid (Fisher LE C + Noah LE (I,S) ).