Skillful seasonal prediction of key carbon cycle components: NPP and fire risk

We investigate the skill of the GloSea5 seasonal forecasting system for two carbon cycle processes, which are strong contributors to global CO2 variability: the impact of meteorological conditions on CO2 uptake by vegetation (characterised by net primary productivity, NPP), and on fire occurrences (characterised by fire risk indices). Current seasonal forecasts of global CO2 concentrations rely on the relationship with the El Nino–Southern Oscillation (ENSO), combined with estimated anthropogenic emissions. NPP and fire are key processes underlying that global CO2–ENSO relationship: In the tropics, during El Nino events, CO2 uptake by vegetation is reduced and fires occur more frequently, leading to higher global CO2 levels. Our study assesses the skill of these processes in the forecast model for the first time. We use the McArthur forest fire index, calculated from daily data from several meteorological variables. We also assess a simpler fire index, based solely on seasonal mean temperature and relative humidity, to test the need for additional complexity. For NPP, the skill is high in regions that respond strongly to ENSO, such as equatorial South America in boreal winter, and northeast Brazil in boreal summer. There is also skill in some regions without a strong ENSO response. The fire risk indices show significant skill across much of the tropics, including Indonesia, southern and eastern Africa, and parts of the Amazon. We relate this skill to the underlying meteorological variables, finding that fire risk in particular follows similar patterns to relative humidity. On the seasonal-mean timescale, the McArthur index offers no benefits over the simpler fire index: they show the same relationship to burnt area and response to ENSO, and the same levels of skill, in almost all cases. Our results highlight potentially useful prediction skill, as well as important limitations, for seasonal forecasts of land-surface impacts of climate variability.