Experiment-Model Integration: A Case Study with the SPRUCE Project
SPRUCE is a climate change manipulative experiment supported by Terrestrial Ecosystem Science Scientific Focus Area of ORNL's Climate Change Program. The experiment aimed toward integration of experiments with ecosystem modeling, data assimilation, and model structure evaluation to yield reliable model projections. Although the SPRUCE experiment only initiated the Deep Peatland Heating treatments in June of 2014 and the whole-ecosystem warming and elevated CO2 treatments will not start until June 2016, the integrated model-experiment approach based on pretreatment data sets and modeling activities have promoted an interactive and mutually beneficial engagement between modelers and experimentalists to advance predictions from experiments and models.
Infrastructure Challenges in the SPRUCE Experiment
The SPRUCE experiment is being operated as the first whole-ecosystem, forest-scale experiment to increase temperature and CO2 concentrations from deep soil to the tops of tree canopies. This decade-long experiment is conducted in a weakly ombrotrophic peatland with a perched water table that has little regional groundwater influence and is located in northern Minnesota in the USDA Forest Service Marcell Experimental Forest (MEF). The 8.1 ha experimental site (S1-Bog) is dominated by Picea mariana (black spruce), regenerated from strip cuts in 1969 and 1974. Located at the southern margin of the boreal peatland, this ecosystem is anticipated to be approaching its tipping point with high vulnerability in response to climate change. Shifts of plant communities at the southern margin of boreal ecosystems under climate change is not fully understood, so large-scale long-term experiments like SPRUCE are needed to improve mechanistic representation of unresolved processes in understudied ecosystems in Earth system models.
To achieve the overall goal of assessing ecosystem-level biological responses of vulnerable, high carbon terrestrial ecosystems to climate change, the SPRUCE experiment faces many infrastructure challenges including, but not limiting to, plot facilities, sensors and instruments, data acquisition and control system, automated data monitoring system, data management, model-data integration, and synthesis of model outputs. In this section, we do not attempt to describe all the research infrastructure challenges facing large-scale experiments like SPRUCE, but we highlight those key elements necessary for model-data integration.
Long-term monitoring of ecosystem dynamics is an important data source for data-model integration. But more powerful manipulative ecosystem experiments are needed to distinguish future climate change impacts from those inherent responses to natural variability. Climate change-caused warming scenarios predicted by the Intergovernmental Panel on Climate Change (IPCC 2013) are much higher than observed variation in mean annual temperatures (±2°C) under current climate. The SPRUCE experiment provides a platform to understand physiological and biogeochemical processes under future climate through a combination of multiple levels of warming up to +9°C at ambient or elevated CO2 levels. Air warming is achieved with heating infrastructures enclosed in 10 plots of 12 m diameter by 8 m high open-top enclosures (Figure 6.2). The open-top enclosures can keep warming air around the enclosed plots by limiting air turnover, while still allowing natural precipitation to fall on experimental plots. The design also enables maintenance of high concentration of CO2 (800-900 ppm) in the elevated CO2 treatment. One unprecedented data set that the SPRUCE experiment can provide for modeling activities is the Deep Peat Heating (DPH, -2 m) with belowground temperatures being consistent with future aboveground warming scenarios (Hanson et al. 2011). Belowground deep warming evaluates responses of deep peat C stocks, microbial communities, and biogeochemical cycling processes, which can be used for evaluating model projections.
The SPRUCE program also provides data from real-time automated monitoring systems, which advance data-model integration approach in an interactive fashion. All data collected from sensors or instruments are recorded by dataloggers and then transferred to a data storage server every 30 min. Data are then published via a web server and used for remote monitoring and control (Krassovski et al. 2015). The standardized data streams are then immediately available to feed and inform models.
(See color insert.) View of SPRUCE experiment infrastructure with (a) exterior view of experimental chamber, (b) interior view of experimental chamber, and (c) aerial view of the S1 Bog site. (Pictures a and c are Oak Ridge National Laboratory Images from: http://mnspruce.ornl. gov. Image b is a PHENOCAM network SPRUCE image from http://phenocam.sr.unh.edu/ webcam/gallery/.)