Voice of editors is a blog from the AGU Publications Department.
The data collected by observations are essential to science. Models also make a fundamental contribution to science by helping us understand processes and predict future changes. However, observations and modeling have certain limitations. “Reanalysis” is a way of combining the two, and then building a more complete and accurate picture of the phenomenon being studied. In Earth sciences, this can be applied to the study of the atmosphere, oceans, and land surface. A recent article published in Geophysics Opinion describes the achievements of reanalyses in the earth sciences and explores the reanalysis of terrestrial ecosystems, which is a much more recent application. Here, the authors explain reanalysis, its uses, and new developments in the reanalysis of terrestrial ecosystems.
How would you explain “reanalysis” in simple language?
A reanalysis in Earth system science takes three main ingredients: a numerical model, observational data, and an optimization scheme to merge the model prediction with the current observations. The result of optimally merged model predictions and observations is called “reanalysis”.
For example, the most recent ERA5 global atmospheric reanalysis produced by the European Center for Medium-Range Weather Forecasts (ECMWF) provides a large number of atmospheric variables (such as temperature and precipitation) at a horizontal resolution of 30 kilometres, a vertical resolution of 137 levels, and an hourly time step from 1979 to today. This reanalysis took into account a very large number of in situ and remotely sensed observation data. It provides a near-optimal reconstruction of atmospheric states for the period since 1979 given our current models and model-data fusion methodology.
Recently, an increasing amount of reanalysis data has been made available for terrestrial processes such as the physical land surface, terrestrial carbon, and the terrestrial hydrological cycle.
What is the main difference between reanalysis and other data products?
Reanalysis provides “continuously optimized states” (eg, soil moisture in a land surface model) and fluxes (eg, evapotranspiration in a land surface model) over a long period. The optimization process is formally called data assimilation. Continuous optimization is generally not performed for simple data products or model output. “Continuously optimized” means that the data assimilation algorithm optimizes at each individual time step the states and flows given the available observations. In data assimilation, the numerical model is propagated up to the time step of the first observations, the model is stopped, the optimization procedure is executed, and the model is propagated up to the next time step with observations. Data assimilation takes into account uncertainties in observations, initial conditions, and may take into account uncertainty in model structure, model parameters, and forcing conditions. Continuous optimization requires large amounts of computing resources because often an ensemble of models is used to produce predictions.
What are reanalysis used for?
Large-scale global atmospheric reanalyses inform high-resolution regional reanalysis by providing boundary conditions. Atmospheric reanalyses are also used to inform ocean simulations and land simulations by providing atmospheric forcing conditions. Autonomous simulation of ocean/terrestrial processes forced by atmospheric reanalysis is called offline coupling. These coupled offline simulations provide more detailed spatio-temporal and process resolution than atmospheric reanalysis.
For water management, the coupling of groundwater and surface water is of particular interest. Carbon cycle predictions focus on carbon, energy and nutrients in ocean and terrestrial ecosystem models. In this way, the reanalysis of the ecosystem finds its place in management and policies such as the development of a blue growth strategy or in the last sixth assessment report AR6 of the IPCC.
Why is the reanalysis of terrestrial ecosystems emerging today?
There are various reasons why ecosystem reanalysis is only available now. In the past, reanalysis focused on physical (abiotic) and biogeochemical (biotic) systems and less on ecosystems (which encompass abiotic and biotic variables and their interactions). Biotic variables are often related to properties of biodiversity such as genetic composition, species population, plant traits, and faunal density.
Main observations from the Fluxnet data network (measuring, for example, the exchange of water, energy and carbon between the earth and the atmosphere), the Integrated Carbon Observation System (ICOS), vegetation indices MODIS, and for hydrology the Global Runoff Data Center (GRDC), and the GRACE mission enabled advanced understanding of processes and improved models. However, data are only available since 2000. These combined data sources allow models to couple carbon and water cycles constrained by energy and nutrient availability. The SMAP mission launched in 2015 provided data for the world’s first coupled terrestrial carbon-water reanalysis. We will see in the future more data provided to constrain ecosystem models and ecosystem reanalysis.
What recent developments will shape and benefit reanalyses of terrestrial ecosystems?
In our review article, we identified four handy steps to achieve the goal of producing an ecosystem reanalysis.
First, researchers will develop new empirical frameworks in the form of numerical models to link physical (abiotic) variables to biological (biotic) variables in integrated ecosystem models. Currently, the link between biotic and abiotic variables is rather poorly represented while biotic variables such as genetic composition, species populations contribute significantly to ecosystem functions, biodiversity as a natural resource and to the resilience of the ecosystem as a whole.
Second, agreement on essential ecosystem variables is needed. The essential variables provide the orientation for the modeling and for the observing communities.
Third, remote sensing trait observations are a research area with great potential to better relate biotic-abiotic processes to sufficient spatio-temporal coverage. This must go hand in hand with the provision of long-term multivariate in situ observations, as the European infrastructure for long-term ecosystem research does. Finally, existing databases (GBIF, TRAITS) and forthcoming high-throughput biodiversity data provide additional data for integration into ecosystem models and ecosystem reanalysis. Integrating these new models and data into model-data fusion frameworks will advance our understanding through digital ecosystem twins and provide new ways to manage terrestrial ecosystems.
—Roland Baatz (r.baatz@fz-juelich.de; 
