A multimodel analysis, validation, and transferability study of global soil wetness products

Multimodel ensemble forecasting has been shown to offer a systematic improvement in the skill of climate prediction with atmosphere and ocean circulation models. However, little such work has been done for the land surface component, an important lower boundary for weather and climate forecast models. In this study, the authors examine and evaluate several methods of combining individual global soil wetness products from uncoupled land surface model calculations and coupled land-atmosphere model reanalyses to produce an ensemble analysis. Analyses are verified against observations from the Global Soil Moisture Data Bank (GSMDB) with skill measured by correlation coefficient and root-mean-square error (RMSE). A preliminary transferability study is conducted as well for investigating the feasibility of transferring ensemble regression parameters within two specific regions (Illinois and east-central China) and between these two regions of similar climate and land use. The results show that when sufficient validation data are available, one can use a seasonally dependent linear regression to improve the skill of any individual model simulation of soil wetness. Further improvements in skill can be achieved with more sophisticated ensembling methods, such as the regression-adjusted multimodel ensemble mean analysis and regression-adjusted multimodel analysis. However, all the ensembling schemes involving regression usually do not help improve the skill scores as far as the simulation of anomalies of soil wetness is concerned. In the absence of calibration data, the simple arithmetic ensemble mean across multiple soil wetness products generally does as well or better than the best individual model at any location in the representation of both soil wetness and its anomaly. Transferability from one subset of stations from the Illinois or east-central China dataset to another gives satisfactory results. However, results are poor when transferring regression weights between different regions, even with similar climate regimes and land cover. Such an exercise helps us to understand better the virtues and limitations of various ensembling techniques and enables progress toward creating an optimum, model-independent analysis from a practical point of view.