As climate modelers prepare to gather in College Park, Maryland, during the first week of April 2018 for the annual US Climate Modeling Summit, one topic that is likely to dominate discussions is whether we need to rethink our approach to improve model performance and accuracy.
This year’s Summit includes a workshop on Land-Atmosphere Interactions and Extremes. Recognition has grown in the scientific community that droughts, heat-waves and other catastrophic weather and climate events are not random in occurrence, nor are they caused only by variations in remote ocean temperatures altering large-scale atmospheric circulation.
Over the last 15-20 years, climate scientists have come to understand the local and regional interplay between land and atmosphere, and how their coupled interactions are a strong driver of climate variations over the continents [Santanello et al., 2018]. However, our models often do not simulate well the processes that connect anomalies in soil moisture, snow, and vegetation status back to the atmosphere, triggering or amplifying climate extremes on time scales of days, weeks, months or even longer.
When ocean models were first coupled to atmospheric models well over a quarter century ago, systematic errors in each component near their interface led to sizeable drift and unrealistic climate simulations. These errors were immediately obvious because the global distribution of sea surface temperature is well known from satellite monitoring, and the symptoms of erroneous coupled model behavior were readily expressed in that variable. This ignited theoretical and observational research that contributed to coupled model development and improved forecasts, nucleating around the El Niño phenomenon.
By comparison, coupled land-atmosphere modeling has evolved backwards to the classical scientific progression, which is the observation of natural phenomena, formulation of hypotheses to explain the phenomenon, development of experiments to elucidate the relevant processes, and finally construction of models to test emerging theories. Instead, land surface models were originally developed to provide lower boundary fluxes of momentum, heat and water for existing atmospheric models.
Whereas meteorological variables like temperature, humidity, winds and precipitation were routinely measured at weather stations around the world, there were little or no observations of land surface states like soil temperature, moisture, and snow mass, or important terms of surface energy and water budgets such as shortwave and longwave radiation, evaporation or sensible heat fluxes.
As a result of this data and knowledge void at the land-atmosphere interface, the inevitable near-surface variable errors in coupled land-atmosphere models became the problem of land surface modelers. Beginning in the 1980s, schemes were developed to adjust land surface states to ameliorate systematic atmospheric model errors in temperature and humidity. Such approaches amount to addressing one error by introducing additional ones.
Today, thanks to advances in satellite and in situ monitoring, the land surface is becoming fully observed. This enables process-driven model development and validation following the scientific method, and is enabling proper initialization of the land surface in operational forecast models.
A key result of recent research is that land-atmosphere interactions are a fully coupled problem needing a fully coupled solution [Santanello et al., 2018]. However, old habits are hard to break, and modeling centers continue to develop component models separately along historical discipline boundaries.
Coupling is typically the final step before official release of a new Earth system model, after the development phase is completed when the only adjustment available is tuning of a few parameters. This is analogous to dance partners training and practicing separately, coming together for the first time on the day of the dance.
Earth system models, particularly the land and atmosphere components that intersect precisely where we all live, grow our food and operate our economies, need to be developed together, with their coupled behavior considered from the start. Nature does not have rigid boundaries where we draw the borders between our scientific disciplines. While the modular approach to modeling is attractive from a design standpoint, it easily leads to compartmentalization and a lack of communication between disciplines. The result has been, and continues to be, underperforming weather and climate models.
—Paul A. Dirmeyer, Editor, Journal of Advances in Modeling Earth Systems (JAMES); and Department of Atmospheric, Oceanic & Earth Sciences, George Mason University; email: [email protected]