H33D-01
An Auto-calibration System Used to Assimilate AMSR-E Data Into a Land Surface Model for Monitoring Soil Moisture and Surface Energy Budget
Low-frequency microwave brightness temperature is strongly affected by near-surface soil moisture; therefore, it can be assimilated into a land surface model to improve modeling of soil moisture and the surface energy budget. This study presents a new variational land system used to assimilate AMSR-E brightness temperature of vertical polarization of 6.9 GHz and 18.7 GHz. The system consists of a land surface model (LSM) used to calculate surface fluxes and soil moisture, a radiative transfer model (RTM) to estimate the microwave brightness temperature, and an optimization scheme to search for optimal values of soil moisture by minimizing the difference between modeled and observed brightness temperature. The LSM is an improved simple biosphere model for sparse vegetation modeling and the RTM is a Q-h model that can account for the effects of surface roughness and vegetation. Several parameters in the LSM and RTM can significantly affect the outputs of the land data assimilation system but their values are either highly variable or unavailable. To solve this problem, we developed a dual-pass assimilation technique. Pass 1 inversely estimates the optimal values of the model parameters with long-term (~months) forcing data and brightness temperature data, while Pass 2 estimates the near-surface soil moisture in a daily assimilation cycle. This system is driven by well-established reanalysis data and global data sets of leaf area index, precipitation, and surface radiation, and was tested at a CEOP (Coordinate Enhanced Observing Period) reference site on the Tibetan Plateau. The system not only detected the effect of precipitation events that were missing in the forcing data, but also led to a significant improvement in modeling of the surface energy budget.
H33D-02
Evaluation of Radar-Rainfall Uncertainties by a Highly Dense Rain Gauge Network
It is well acknowledged that there are large uncertainties associated with radar-rainfall estimates. Numerous sources of these errors are due to parameter estimation, the observational system and measurement principle, and not fully understood physical processes. To describe these uncertainties, rain gauge data are usually considered as a good approximation of the true rainfall values. However, in the vast majority of the cases, the available networks are too sparse to accurately describe the true rainfall process, adding uncertainties to the radar-rain gauge comparison. The authors will use a large (seven years) dataset of rainfall measurements by a highly dense rain gauge network deployed during the HYdrological Radar EXperiment (HYREX) in the Brue catchment, south-west part of England. This network presents unprecedented opportunity for the investigation of radar-rainfall uncertainties. In addition to the length of the dataset, one unique characteristic resides in its configuration: on a regular grid with 2×2 km2 resolution, there are 20 pixels with one gauge, seven pixels with two gauges and two super-dense pixels with eight gauges. The radar-rainfall estimates for the same time period are from C-band weather radar located at approximately 40 km from the catchment. Focusing on the two very dense pixels, the authors describe the uncertainties in radar-rainfall estimates for different accumulation times (5-minute to daily), modeling the errors using an additive and a multiplicative model.
H33D-03
Toward Improved Calibration of Distributed Hydrologic Models via Uncertainty Analysis
Hydrologic models are generally dependent on physical and/or empirical parameters that may be difficult or impossible to measure directly. Sophisticated parameter calibration routines have been developed as a means of increasing model performance while both characterizing and reducing model uncertainty. Lumped models, which often contain a relatively low-dimensional parameter vector, in particular, have profited from such routines. As distributed models are employed more frequently to make use of spatially distributed measurements, there is a need to implement similar procedures to improve distributed model efficiency. However, application in distributed models often requires analysis of a larger parameter vector with a greater degree of dimensionality. Issues regarding parameter identifiability, cross-correlation, and search algorithms are further confounded by this increase in dimensionality. This research aims to demonstrate that fully distributed models (e.g. TIN-based Real-time Integrated Basin Simulator or tRIBS) can also benefit from implementation of similar calibration procedures (i.e., Generalized Likelihood Uncertainty Estimation or GLUE). Results from application of GLUE to12-month simulations (May 1996 to May 1997) of the Peacheater Creek basin near Eldon, Oklahoma, demonstrate the concept of "equifinality" between many different parameter sets. Furthermore, many of these parameter sets achieve Nash-Sutcliffe efficiencies greater than 0.7. GLUE not only increases model performance, but it provides estimates of model output uncertainty as well as discrete approximations of the posterior parameter probability distributions. These probability distributions can then be used as part of an ensemble-modeling scheme. A notable limitation in this procedure, however, is the computational burden of a Monte Carlo simulation using a complex hydrologic model.