Robert Pincus (PI)
NOAA/CIRES
Robert.Pincus@colorado.edu

Can Assimilating Satellite Observations of Clouds Improve Forecast Skill?

Accurate weather forecasts rely heavily on obtaining accurate initial conditions, and data assimilation has emerged as the optimum way to use observations to arrive at an estimate of the instantaneous state of the atmosphere. Current data assimilation schemes use a wide range of observations but exclude all those that are related to clouds, despite the fact the cloud properties are part of every forecast model’s description of the atmosphere. Clouds have been neglected because they are very hard to assimilate, in large part because cloud schemes in forecasting models couple in complicated and sometimes non-unique ways to the rest of the model state. Nonetheless, introducing cloud observations into the assimilation process might constrain the state of the atmosphere more tightly, leading to better initial conditions and better forecasts. Satellites provide the most likely source of observations, since these are available globally and contain more quantitative information than do surface observations.   We propose to determine how much forecast skill can be gained by adding satellite observations of clouds to the assimilation process. We will focus on observations of clouds made by MODIS aboard the Terra and Aqua platforms, which we will assimilate in four global models using an ensemble Kalman filter system. We will assess the degree to which assimilating these observations can be expected to improve the characterization of the state of the atmosphere, and to what extent these improved initial conditions increase short-term forecast skill for each model.   The assimilation of satellite observations of clouds is made possible by the convergence of two relatively new ideas. Ensemble assimilation systems make it possible to assimilate observations without the construction of an adjoint. This is a particular advantage with respect to clouds, since cloud schemes in forecast models contain thresholds and strong non-linearities that make the derivation of adjoints especially difficult. There has also been substantial progress in the mapping between large-scale model state and fine-scale observations through the generation of subcolumns, such as in the “ISCCP simulator.” These success provide guidance for how we may cast model states and high-resolution observations into similar forms (e.g. joint probability distributions) to make assimilation tractable.   We will experiment with the assimilation of observations in circumstances that move from highly idealized to highly realistic for both single observations and sets of observations. We will assess the maximum possible increment to forecast skill that could arise by assimilating cloud observations through “perfect model” experiments in which the “observations” are taken directly from the model state vector. The impact of the forward operator will be tested by assimilating “pseudo-observations” created by using the exact model state vector through the subcolumn generator. Finally, we will assimilate MODIS observations of cloud top pressure and cloud optical depth, first on a single-granule basis, and then for the entire globe. We will repeat these experiments for a range of models with different cloud schemes to test how sensitive the assimilation is to the cloud scheme being used.

+ Back to Participant Listing