The Effect of Rainfall Error Characterization on the Efficiency of a Land Data Assimilation System for Soil Moisture Prediction

Date of Completion

January 2012


Engineering, Environmental|Remote Sensing




This study assesses the impact of satellite-rainfall error structure on soil moisture simulations with the NASA Catchment Land Surface Model (CLSM) and the efficiency of assimilating near-surface soil moisture observations in the NASA Land Data Assimilation System (NASA-LDAS). A multi-dimensional satellite rainfall error model (SREM2D) is compared to the standard error model (CTRL) used to generate rainfall ensembles as part of the NASA-LDAS. The error analysis is assessed in terms of rainfall ensembles and corresponding soil moisture predictions. Comparisons of rainfall ensembles generated by SREM2D and CTRL against reference radar rainfall show that both rainfall error models preserve the rainfall error characteristics across a range of spatial scales. However, SREM2D generates rainfall replicates with higher variability that better envelope the reference rainfall than those generated by CTRL. On the other hand, uncertainty in soil moisture is shown to be less sensitive to the complexity of the precipitation error model. This is attributed to the fact that soil moisture processes dampen the variability in the precipitation forcing. A sensitivity analysis is then conducted to investigate the contribution of rainfall-forcing uncertainty relative to model uncertainty in predicting soil moisture. Specifically, rainfall-forcing uncertainty is introduced by SREM2D, whereas errors in CLSM are modeled with two approaches: either by perturbing model parameters or by adding randomly generated noise to model prognostic variables. A reasonable spread in soil moisture is achieved with relatively few parameter perturbations, while the same ensemble width requires stronger perturbations with the prognostics perturbation method. The probability of encapsulating the reference soil moisture simulation increases when rainfall-forcing uncertainty and model uncertainty approaches are combined (compared to rainfall uncertainty alone). This improvement is more significant when perturbing parameters as opposed to perturbing prognostics. When assimilating near-surface soil moisture data through LDAS, soil moisture estimates exhibit improved performance metrics (higher anomaly correlation coefficients and lower root mean square errors). However, no significant dependence on the rainfall error model complexity is shown when assimilating actual satellite soil moisture observations, suggesting that the simple rainfall error model may be adequate in many applications. ^