Data Assimilation Meetings at Reading

Date Meeting type Speakers
10 June 2015
Maths 314
Invited speaker Luke Smallman (University of Edinburgh)
Improving our understanding of ecosystem processes using data assimilation
Data assimilation (DA) has become an invaluable tool for combining observations and models to improve our understanding of the current state and process knowledge of the Earth system. Current research foci in Edinburgh include site level analyses of managed forests and continental scale analyses investigating the role of fire in controlling ecosystem processes. Managed forests present a valuable opportunity due to the existence of spatially explicit datasets of planting dates etc. which can be combined in an appropriate DA framework. Fire has a significant impact on global terrestrial carbon balance. Moreover fire has been identified as playing a key role in the maintenance of savannah ecosystems preventing succession. Therefore fire can impact ecosystem processes, the extent of which needs to be better quantified. We use a Metropolis Hastings – Markov Chain Monte Carlo DA procedure to investigate the terrestrial carbon cycle. The DA analyses combine both remotely sensed information and site level information with prior knowledge via ecological and dynamical constraints (EDCs). Site level analysis at Duke Forest, USA (Loblolly pine) show that assimilation of a single observations of woody carbon stocks can substantially reduce analysis error (rmse = 7 MgC ha-1→ 3.5 MgC ha-1) and uncertainty (> 55% reduction). However errors in remotely sensed LAI can result in significant errors in the retrieved parameter distributions while still accurately simulating ecosystem carbon stock accumulation through compensating errors. Australia's mean median net carbon sink was 0.60 PgC yr-1 (2001-2010). The 'finger print' of fire emerges from posterior parameter distributions of key process parameters (e.g. NPP:GPP). Moreover cluster analysis of the parameter posteriors highlight areas with substantial fire activity as having distinct parameter combinations. Indicating existing land cover maps do not adequately discretize the land surface into functional groups."

Page navigation