Mark Richardson

In this seminar, we will introduce the rationale, the methodology, and initial applications of data assimilation for Mars. We view data assimiation primarily as an optimal tool for testing of numerical models against observations; and secondarily as a tool for generating a "best guess" global dataset that can be applied "off line" to climatological and geological problems. While steeped in an aura of mystery and complexity, we propose that data assimilation should be a basic tool in the inventory of any numerical model analysis system, and that it in fact represents the simplest and most direct means of properly comparing models and data. What does that really mean? At its purest, data assimilation is simply the implementation of a measurement filter: when we have a noisy measurement of a modelable system, the optimal estimate of the system state is obtained from a combination of the model prediction (the prior estimate) and the observation, with a weight (or gain) that reflects the degree of error (variance) in both the measurement and the model (prior) estimate. The new estimate may not even be the useful information we obtain from such a filter: persistant deviations (biases) between the prior estimates and the measurements can yield enormously useful information for the testing of models, the isolation of model errors, and constraining potential improvements in the model processes. By building instrument observation patterns and "forward models" (e.g. the mapping between model state variables like temperature profiles and measurement values like radiances in specific wavelength bands) into the system, data assimilation minimizes ambiguity in model-data comparison. Finally, by simultaneously allowing arbitrary numbers of diverse measurement types into the mix, data assimilation provides the most complete and straightforward means of constraining models. Why do we need this? Mars climate is both well observed and quantitatively poorly understood (by terrestrial standards; it is spectacularly well understood by most other planetary standards). Progress in the field of Martian climate dynamics is now largely limited by the ability to properly constrain and test models with the available data. While this reflects the very significant progress that has been made in both conceptual and numerical modeling of observations in the last 20 years, it means that the current methods of model-data comparison have probably run their course. The power of the observational archive for Mars is in its length (going back to 1997 with continuous observations), its regularity (mainly from polar mapping orbit), and its diversity (with multichannel quantitative data coming from multiple instruments on four different orbiters). This represents the posibility of significant simultaneous constraints on the climate system for a continous variety of different seasonal and meteorological states. Significant further progress at this point is not really limited by extant observations (though this is still the case for low-period variable phenomena like major dust storms) or by the state of development of current numerical models (or computing power), but by the comparatively primative, cumbersome, and wasteful means we have of testing and improving models with data. The specific data assimilation system we have implemented for Mars is based on the open source NCAR Data Assimilation Research Testbed (DART) system. While we have implemented it for our planetWRF (Mars General Circulation Model - GCM) system, the goal of NCAR's DART is to make it as easy as possible to integrate any numerical model. We have continued in this spirit by maintaining a clean separation of instrument models (currently just MGS TES) and the required planetary timing system based on the Mars24 package, from the numerical model. DART uses an ensemble of numerical model simulations to create the covariances for a given data assimilation step or cycle. DART then uses the observations and observation covariances, the forward model of the observing system, and the model state and model covariances to yield a new state estimate (the assimilation or "reanalysis"). That state estimate is then used to spawn a new set of model ensemble simulations for the next data assimilation cycle. Useful information for testing models can be obtained with such a system for Mars with only a few tens of days of integration with DART. The version of DART that includes the Martian calendar, the TES nadir forward model, and the MarsWRF GCM model is fully open source can be obtained from www.marsclimatecenter.com/models.html