Evaluation of Multiple Land Surface Models for a Data Sparse Region

Several studies have shown that multi-model ensembles tend to produce and capture anomalies and variability better than the individual models.  In regions with sparse data networks and observations, like East Africa, multi-model simulations are particularly useful for hydrologic and drought monitoring purposes.  Also, ensembles could be used to derive better estimates of evapotranspiration (ET), streamflow and water availability over single model experiments.  In this study, we evaluate four land surface models (LSMs), against their ensemble, and independent in situ, remotely sensed and reanalysis products.  The LSMs selected are part of the current Famine Early Warning System Network (FEWS NET) Land Data Assimilation System (FLDAS), which consist of Noah version 3.3 and the Variable Infiltration Capacity (VIC), version 4.12.   Two additional LSMs, which incorporate prognostic water table schemes, are the NASA/GMAO’s Catchment LSM and the Noah Multi-Physics (MP) LSM.  In addition, the LSMs’ surface and subsurface runoff are routed through the Hydrological Modeling and Analysis Platform (HyMAP) to simulate surface water dynamics.   The four LSMs are each spun-up for 105 years (three times from 1981-2015) from which then their respective open-loop (OL) runs were generated for the period of 1981-2016.  The LSMs are driven by NASA / GMAO’s Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2), which contains global, hourly atmospheric forcing fields.  For precipitation, the USGS and UCSB Climate Hazards Group InfraRed Precipitation with Station (CHIRPS) daily rainfall dataset is used and temporally subsetted by the MERRA-2 hourly precipitation fields.  The NASA/GSFC Land Information System (LIS) software framework integrates these forcing datasets and drives the four LSMs and HyMAP. 

This presentation reports on the evaluation of the four LSMs’ long-term simulations over East Africa and the challenges of evaluating them in such a data sparse region.  We will also draw upon examples and experience of evaluating the models from other reanalysis projects evaluated in data-rich areas, like the North American LDAS (NLDAS) and its test bed analysis.  The Land Verification Toolkit (LVT) is used for the evaluation, as it provides model ensemble metrics and the ability to compare against a variety of remotely sensed measurements, like different ET and soil moisture products, and other reanalysis datasets that are available for this region. Comparison of the models’ energy and hydrological budgets are shown for this region (and sub-basin level, e.g., Blue Nile River) and time period (1981-2015), along with evaluating ET, streamflow, groundwater storage and soil moisture, using different metrics (e.g., anomaly correlation, RMSE, Nash-Sutcliffe efficiency, etc.).