Toward integrated seasonal predictions of land and ocean carbon flux: lessons learned from NASA’s subseasonal-to-seasonal predictions

Abstract:

Seasonal forecasts made by coupled atmosphere-ocean general circulation models (GCMs) are increasingly able to provide skillful forecasts of climate anomalies. At some centers, the capabilities of these models are being expanded to represent carbon-climate feedbacks including ocean biogeochemistry (OB), terrestrial biosphere (TB) interactions, and fires.

Here, we examine whether land and ocean carbon flux anomalies that occurred over the past 10 years could have been predicted months in advance. This period included both El Nino and La Nina episodes, which are predictable several months in advance, and a wealth of remote sensing data to evaluate performance. We explore this topic using NASA’s Goddard Earth Observing System (GEOS) model, which routinely produces an ensemble of seasonal climate forecasts, and a suite of offline dynamical and statistical models that estimate carbon flux processes. Using GEOS forecast fields from 2007-16 to force flux model hindcasts shows that these models are able to reproduce significant features observed by satellites. Specifically, OB hindcasts are able to predict anomalies in chlorophyll distributions with lead times of 3-4 months. Statistical fire forecasts driven by ocean climate indices are able to predict burned area in the tropics with lead times of 3-12 months. The ability of TB hindcasts to reproduce land flux anomalies is controlled by the skill of the climate forecast. Application of a bias correction to address substantial errors in the seasonal forecast meteorology greatly improves TB predictions, supporting skillful forecasts out to 3 months. Though TB predictions display skill in reproducing estimates produced by the same TB model driven by observed meteorology, predictions from different models can also disagree with each other, highlighting the need for multi-model approaches and for continued model development to improve the reliability of forecasts.

While seasonal forecasting remains an active area of research, these results demonstrate that forecasts of carbon flux processes can support a variety of applications, allowing scientists to understand carbon-climate feedbacks as they happen and to capitalize on more flexible satellite technologies that allow areas of interest to be targeted with lead times of weeks to months.