The blending of observational data sets with dynamic environmental models depends on control parameters such as initial conditions, boundary conditions, model errors, and empirical model parameters. The uncertainty of these parameters lead to forecast errors. Data assimilation is a method to minimize the growth of these forecast errors while staying true to the observational data sets.
CIRA has a wide range of activities in variational and ensemble data assimilation research. Our variational techniques range from 1-dimensional variational (1DVAR) satellite profile retrieval and surface parameter retrieval systems, to 2-dimensional (2DVAR) deep soil moisture systems that exploit the temporal and vertical dynamics of land surface models to obtain what otherwise is an indirect remote sensing observation of the surface. Extensive 3D and 4D NWP-focused systems include work with the operational WRF-3DVAR, and a mesoscale cloudy 4DVAR data assimilation using the CSU Regional Atmospheric Modeling Data Assimilation System (RAMDAS). Advanced ensemble data assimilation systems using WRF and several other models is performed using the Maximum Likelihood Ensemble Filter (MLEF) also developed at CIRA. CIRA work has also expanded data assimilation theory to non-Gaussian probability distributions with work underway using hybrid Normal/log-Normal data assimilation using 3D and 4D systems.
The target of these activities range from direct satellite radiance assimilation, spatial representation filters using Backus-Gilbert techniques, preconditioner development, improvement of quality control procedures and minimization techniques, and advanced control theory and systems using ensemble techniques. A wide range of observational studies using cloudy data assimilation, land surface data assimilation, and climate systems are underway in several projects. Our science team members are highly active with invited talks at several conferences, workshops, and other scientific community involvement with many Govt. agencies that require these skill sets. We welcome collaborations as a way for us to expand the educational reach of data assimilation as a needed scientific tool for better understanding of our environment, and as a means toward improved environmental forecast performance and identification of important feedbacks within these complex environmental systems.