A Historical Record of Actual Evapotranspiration in Sub-Saharan Africa using Climate Reanalysis and Remote Sensing Data

Climate change is expected to have the greatest impact on the world's poor. In the Sahel, a climatically sensitive region where subsistence farming is the primary livelihood, evaporation is expected to increase, leading to a decrease in water supply and increase in food insecurity. Studies on climate change and the intensification of the water cycle and how they influence environmental decision-making are still in their infancy in sub-Saharan Africa. This is due in part to poor parameterization and calibration of evapotranspiration (ET), a key input in hydrologic models. ET is an integral component of the monsoon rainfall feedback: soil water vapor enhances evaporation and subsequently rainfall during the monsoon season. Regional forcings, such as ET and global circulation predictors, such as the El Nino Southern Oscillation (ENSO) have the potential to explain much of the rainfall variability in the Sahel. Proper ET representation requires instrumentation that is costly and difficult to extrapolate over large areas and model parameters that scale linearly in space.

In this paper, we used a modified Priestly-Taylor formulation to model ET over sub-Saharan Africa from 1981-1999. The model was selected, because it yields ET totals that are more similar to flux towers representing major biomes in the tropics and sub-tropics than other common ET modeling approaches. It was recently compared with flux towers in sub-Saharan Africa against an energy balance and Penman-Monteith approach with comparable results. The primary objective of this paper is to demonstrate how the most recent advances in climate reanalysis and remote sensing inputs can be used to monitor ET in sub-Saharan Africa. This paper will also perform the following: evaluation of the 1981-1999 ET time series using runoff from major catchments, extension of the time series with a modern ET dataset (2000-2008), and characterization of spatio-temporal patterns in historical ET.

The model requires five inputs (Normalized Difference Vegetation Index- NDVI, Soil Adjusted Vegetation Index- SAVI, net radiation, maximum air temperature, and vapor pressure deficit/relative humidity) and yields average daily latent heat flux at the surface, which can be converted to ET using the latent heat of vaporization. In order to analyze intraseasonal and interannual variability in ET, the daily output was aggregated to a monthly timestep. Monthly maximum NDVI and SAVI were derived for 1981-1999 from the daily 0.05 degree resolution Advanced Very High Resolution Radiometer (AVHRR) Land Long Term Data Record (LTDR). LTDR is an improvement over the commonly used Global Inventory Modeling and Mapping Studies (GIMMS) dataset. Unlike the GIMMS dataset, LTDR corrects daily reflectance data with a rigorous atmospheric correction algorithm before computing NDVI or SAVI. A smoothing algorithm was applied to the LTDR dataset to further reduce noise from atmospheric contamination. The smoothing algorithm preferentially gives more weight to peaks than troughs, thus preserving plant phenology. The climate inputs were derived from the 3 hourly 1.0 degree resolution Sheffield et al. (2006) reanalysis. Shortwave and longwave radiation fluxes were combined to compute the daytime average net radiation. Surface specific humidity, pressure, and ambient temperature were used to compute the minimum relative humidity (RHmin) and maximum vapor pressure deficit (VPDmax). The model assumes that the relationship between soil evaporation and atmospheric humidity is strongest at this daily extreme. The time series will be extended to 2008 using a dataset previously evaluated using flux tower data and modeled ET. This dataset was parameterized with the 16-day 0.05 degree resolution Moderate Resolution Imaging Spectroradiometer vegetation indices and 3 hourly 0.25 degree resolution Global Land Data Assimilation System reanalysis.

The 1981-1999 ET will be evaluated using downscaled gauge-enhanced precipitation fields and corresponding runoff from eight large catchments in sub-Saharan Africa. Annual totals will be compared directly, while seasonal totals will be compared after a lag is introduced to represent soil storage. The historical and modern datasets will be compared to see if the assumptions of stationary and thus extension of the dataset is possible. Standard summary statistics will be used to characterize spatial patterns in ET over the continent. Non-parametric and parametric techniques will be used to characterize significant trends in ET both seasonally and annually. It is expected that: 1) the model will perform best in monsoonal climate regimes, as vapor pressure deficit (advection) is a poor indicator of ET when water stress occurs primarily outside the growing season, 2) ET will be smaller in these areas, but will have greater variability, due to substantial variation in vegetation cover, and 3) there will be an overall increase/decrease in ET in sub-tropical/tropical Africa, reflecting current trends in runoff, and the most significant trends will be during the growing season when the precipitation and ET feedback is strongest.

In the future, the 28 year time series will be combined with global climate predictors, such as ENSO, Madden-Julian oscillation, and North Atlantic Oscillation to improve rainfall forecasts in sub-Saharan Africa. Ultimately, the new ET product will be integrated into the National Aeronautic and Space Administration Land Information System (LIS) evaluated globally. The LIS software framework uses time-varying precipitation and surface meteorology estimates to drive advanced land surface models that characterize land surface states and fluxes. LIS includes the latest in computing and data management technologies, providing a test bed for high resolution data analysis. The LIS components are flexible, making re-parameterization, model inter-comparison, and validation possible. Modeled evapotranspiration is critical to hydrologic forecasting and drought prediction. Consistent and reliable estimates of ET are necessary for continued monitoring and effective mitigation to social problems linked to it.