Statement of Research Interests
Alexey Kaplan

General. My research interests are in studies of climate and ocean variability, its modeling, hindcast, and prediction, and in developing new techniques for such studies.

Past work. In the last few years I was actively exploring the low-dimensional (reduced space) approach to the optimal estimation of climate fields. This relatively novel tool proved to be quite effective for estimating climate fields on the basis of sparse and erroneous data via scale separation. It involves analyzing the covariance of a target field and using its leading eigenvectors (often called empirical orthogonal functions, or EOFs) to approximate the large-scale portion of the field; the contribution of the small-scale parts is estimated separately. In many applications this division can be roughly described as the separation of synoptic scale and mesoscale types of variability.

I applied this approach to a wide variety of interesting and important problems arising in ocean and climate research: assimilation of a sparse observational data set into a physical model for the tropical ocean; low-dimensional representation and initialization of the coupled tropical ocean-atmosphere system; analysis (interpolating gaps and suppressing observational/sampling error) of historical data sets of marine climate observations; reconstruction of marine climate fields on the basis of paleoclimatic data sets. As a result of these efforts, new techniques of data assimilation and optimal data analysis, as well as analyzed data products were developed. Historical (150 years long) analyses of sea surface temperature (SST) and sea level pressure based on such an approach have enjoyed a wide recognition; use of a similar method for climate reconstructions from paleoclimatic proxies resulted in a few trial reconstructions and a deeper understanding of the intrinsic problems associated with climate records in biological systems. My data assimilation work and the associated hindcast of the tropical Pacific sea level height fields contributed to the improved technique for the real-time El Niño forecasting, developed by Dake Chen and others in the LDEO Climate Modeling Group.

Current work. I am working on a new analysis of the SST which is globally complete, of high (1x1 degree) spatial resolution, and accompanied by the analysis of ice concentration. This involves the use of the satellite era climate data coverage in order to provide optimal reconstructions of climate fields from in situ data. I also work on analyses (trial versions have been produced) of ship observations of other basic marine meteorological variables (winds, humidity, air temperature, cloudiness), and on the extension of analyses back to 1820s. Future work will involve producing historical analyses of ocean-atmosphere flux fields, suitable for the use as boundary conditions of the ocean models. Multivariate analyses of historical states of the tropical Pacific ocean-atmosphere system useful for the hindcast and forecast studies are also planned for the near future. The task of producing gridded objective analysis of subsurface ocean data (newest version of the World Ocean Atlas) is especially important due to the need to distinguish between steric and eustatic contributions in the current global mean sea level rise.

Future foci. From my work on these subjects, a few projects which I find particularly interesting have emerged. The first of them is the in-depth study of mesoscale variability of climate variables. This is the physical variability that is responsible for the most of the sampling error in the practical gridded representations of climate fields. It is complementary to their typical low-dimensional representations (which only captures synoptic scales, at best), and often can be successfully described statistically, for the purpose of synoptic scale analyses. Advent of remote-sensing observational systems allows unprecedented detail in describing this variability. These descriptions may then be used for the optimal analyses of a pre-satellite period. This type of variability also often lends itself to physical modeling, and that makes its study particularly exciting. My paper published in JGR in 2004 analyzed small-scale (mesoscale) variability in the tropical sea level height and used it for interpreting errors in ocean simulation and assimilation products. I also recently derived an explicit connection between small-scale sea level height variability and surface geostrophic kinetic energy (in revision for JAOT). I believe that this line of research will result in a better description (and possibly correction) of errors in ocean models and observational data sets. It is also of genuine fundamental scientific interest: spectral distribution of ocean variables were studied theoretically for more than half a century already. Finally, satellite observations and high resolution ocean models can check those theories and provide some definite answers.

These directions of my research are currently boosted by the NOAA and ONR research grants, for which I committed to produce sampling error models for satellite SST and sea surface heights data and to test submesoscale parameterizations of high resolution ocean models for Monterey Bay area, respectively. Mesoscale variability in SST and other ocean variables are important for my other research areas as well, in particular, for the high resolution historical analyses. An especially interesting case is coastal variability. Coastal temperatures are often well observed, and they may be quite different from the open ocean temperatures nearby. Yet it is often a coastal temperature which is being recorded by biological proxies, and which affects the coastal hydroclimatic cycle in general.

Regarding coupled climate modeling and data assimilation, a special interest to me presents the following problem: how does one initialize a coupled model for optimal prediction, if the imperfections of the model itself result in the best initial state for the prediction being different from the best nowcast analysis. Main directions of my work on this subject involve (1) systematic studies of the error dynamics in a coupled system; (2) finding out how relatively small errors and biases of oceanic and atmospheric model components interact to produce dominant biases of the coupled system whose climatology is quite different from that observed; (3) developing techniques for optimal initialization in presence of these errors, whose dynamics does not obey a standard textbook assumption on error randomness and independence. Some of the results obtained so far were published in collaboration with D.Chen, R.Cañizares, A.R.Karspeck, M.A.Cane, and S.E.Zebiak, but we are still quite far from understanding these problems well.

A special focus area of my interests is an adequate representation of uncertainty in climate fields, which, in my view, can only be achieved via ensembles of realizations sampled from posterior distributions obtained for climate fields on the basis of all information entering the estimation problem. Such an approach naturally requires the Bayesian description of uncertainty and efficient sampling algorithms (MCMC). In the last few years I used Bayesian approach to describing uncertainties in a few simple climatological problems [ Fairbanks et al., 2005; Kent and Kaplan, 2006; Farmer et al., 2007], but my goal is to make it possible to represent uncertainty in high dimensional climate fields estimated from observations and models in way that could be practically useful and widely distributable in the climate research community.

Another area of my interests is the climate of the nineteenth century (including not very well-known climate of the first few decades of the twentieth century as well). This is the period when instrumental, historical, and paleo data are abundant enough for independent pictures to be drawn from these different sources. These pictures often do not match. It is important to be able to reconcile these different sources, and verify the error of each in the process. In particular, I find it imperative to reconcile instrumental and historical chronologies of El Niño events from the 1820s to present. This makes necessary a better understanding of (un)certainty with which various kinds of land or coastal impacts recorded in the historical documents can be caused by different ENSO stages. It also involves revisiting some historical texts and re-evaluating the associated climatic evidence traditionally used by historical climatologists.

Further development of high-resolution paleoclimatic reconstructions is another traditional area of my interests. In recent years it became very clear to me and colleagues (M.N.Evans, M.A.Cane, J.E.Smerdon) that the limitations of currently used statistical techniques and the lack of direct physical models based on local climate variables hurts badly the reliability of paleoreconstructions. I intend to conduct further research on reconstruction methodology and direct modeling or model testing for biological proxies for climate (e.g. corals and trees), with an eventual goal of inverting such models and improving paleoreconstructions.

I will also proceed to work on a few methodological issues pertinent to the reduced space optimal analyses which are of a significant methodological importance for the climate analysis field in general. These are the issues of errors in the covariance estimates, their influence on the optimal analysis procedures, stability of the reduced spaces, objective procedures of climate proxy calibration, objective procedures for defining climate indices, and connections between prediction problems and optimal estimation procedures. I am interested in exploring the full potential of low-dimensional representations of various climate subsystems for recovering their low-order dynamics and identifying the limits of their predictability.