Ecosystem Informatics (EI) is addressing research problem in many areas of ecosystem science. Research aims to produce outcomes that are publishable not only in major ecology journals but also in engineering, computer science, and mathematics journals. The four research areas in EI are:  

Carbon cycling, climate, climate change, fire: Decomposition, Growth and Mortality, Carbon Dynamics of forests, and Airshed dynamics.

Community stability and resilience, disease, invasion: Vector-borne disease, Epidemiology, and Invasive species.

Hydrology, streams, network dynamics: Hydrology, River restoration and engineering, Benthic macroinvertebrates, and Sediment transport.

Wildlife and changing landscapes, streamscapes, and seascapes: Ecosystem consequences of land-use change, Marine Geography, Modeling demographic processes, Seafloor Mapping, and Species distribution modeling. 

Examples of research approaches and contributions (many of which are catalyzed by the EI program based at Oregon State University) include:Combining alternative modeling approaches to estimate uncertainty, e.g., in ecosystem flux estimates. Generalized linear uncertainty evaluation (Beven and Binley 1992) reveals that parameter equifinality (many combinations of apparently feasible parameter values) limits predictive accuracy of complex ecosystem models (e.g., BIOME-BGC), resulting in underestimates of net ecosystem CO2 exchange, while model structure uncertainty (i.e., in soil hydrology) plays a substantial part in model-data mismatch (Mitchell, Beven Freer Ecol Mod 2008). Thus, simpler models (e.g., harmon standcarb ref) may be more effective in revealing effects of forest management, such as tradeoffs between fuels reduction and carbon sequestration (Mitchell/Harmon Ecol Apps) in part because they can be validated by analytical solutions (Ngo 2006).                                                                                                                                                                             

Quantifying disturbance processes and predicting effects of changing disturbance regimes. Hazard rate functions (probability distributions)(Johnson/Gutshell, Johnson/Reed) permit predictions of forest age class distributions for age-specific and changing disturbance regimes of e.g. forest fire (Tepley/Thomann). “Heavy-tailed” probability distributions (in which rare events are more likely than assumed by the usual exponential distribution) help characterize the likelihood and scaling behavior of extreme events, such as extreme air and sea surface temperatures (Fuentes, Henry, Reich 2009), and unexpectedly long residence times of solutes (McGuire and McDonnell 2005, Haggerty et al grl 2002). Applied probability and statistics also can help understand the spatial scaling behavior of extreme precipitation andflood events (Gupta and Waymire, Gupta and Dawdy).    

Automated detection and identification of organisms and events using novel instrumentation technologies and multiple monitoring datasets. Computer science and engineering approaches can help ecologists manage and extract information from large datasets. For example, machine learning algorithms based on Bayesian inference can automate detection of errors in e.g., climate datasets (Derezynski and Dietterich 2007, Derezynski and Dietterich 2009 ACM) and automate identification of aquatic insects (Dietterich ref). Sensor networks (Porter et al 2005) for environmental variables, including sound, provide huge spatially and temporally diverse datasets suitable for analysis using machine learningapproaches to detect responses of organisms to environmental change (e.g., Laiolo 2008). Modeling complex communities and disease systems and their susceptibility to perturbations. Ongoing changes in ecosystems are altering the behavior of pests and diseases, with changes in persistence and spread of vector-borne disease epidemics in the environment.

Key EI problems relate to modeling disease behavior at intermediate (i.e., transient, non-equilibrium) time scales, or integrating multiple species interactions. For example, using mathematical models of disease, Hosack et al (2008) developed a new theory for disease epidemics that predicts whether or not a population or disease may increase in the 5 short term even if it cannot persist in the long term, and Moore et al (2009) demonstrated how predation may be sufficient to control pathogen prevalence indirectly via the vector.  Quantifying species distributions, correlations with environmental factors, and changes in distributions. Species are responding rapidly to climate and land-use change, and a recent explosion of papers aim to detect changing animal and plant distributions building on existing, large datasets (e.g., Elith et al 2006). Machine learning approaches are helping to improve the performance of species mapping (Phillips et al 2009 ecol apps) and calibrate models (Phillips and Elith, ecology, in press).