Assessing the Utility of Multiple Land Surface Models for Drought Simulation

Ensemble simulation has been widely used in weather forecast and climate prediction as a way of assessing uncertainty of simulated variables. In general, combining multi-model outputs lead to increased simulation skill as the individual model errors tend to cancel each other out in an ensemble mean estimate, as demonstrated through multi-model projects such as the North American Land Data Assimilation System (NLDAS). Recent studies also suggest that the NLDAS development has pushed the constituent models to be too similar to one another, which raises questions about their true utility to the ensemble. In this study, we assess the utility of multimodel configurations of NLDAS for drought estimation. Four existing land surface models (LSMs) in the North American Data Assimilation System (NLDAS) Phase 2, namely Noah-2.8, Mosaic, VIC-4.0.3, and SAC, and four candidate models likely to be added to the next generation NLDAS, namely Noah-3.6, Noah-MP, VIC-4.1.2 and Catchment land surface models, are employed for drought simulations. All the models are running at the same spatial and temporal resolutions using the NLDAS Phase 2 forcing. Archived U.S. Drought Monitor (USDM) data will be used for assessing the effectiveness of drought simulation by the 8 models.

This study focuses on two key research questions: 1) How similar are the drought estimates from the constituent models and the common factor across the ensemble? The similarity of drought simulations is assessed using a confirmatory factor analysis method, which will find out how much a common drought factor plays in each model simulation. The 8 NLDAS models are evaluated for the similarity of drought simulation over different locations of the CONUS domain. 2) What is the contribution of atmospheric boundary conditions (primarily precipitation) in the climatological distribution of drought estimates? This is assessed using a k-means clustering method that partitions the modeling domain into a number of zones, each with similar statistical features. An analysis of the clustering patterns between the input forcing fields and the drought estimates will be used to quantify the impact of forcing inputs to drought estimation.