Integrating Automated Data into Ecosystem Models: How Can We Drink from a Firehose?

Sensors and imaging are changing the way we are measuring ecosystem behavior. Within short time frames, we are able to capture how organisms behave in response to rapid change, and detect events that alter composition and shift states. To transform these observations into process-level understanding, we need to efficiently interpret signals. One way to do this is to automatically integrate the data into ecosystem models. In our soil carbon cycling studies, we collect continuous time series for meteorological conditions, soil processes, and automated imagery.

To characterize the timing and clarity of change behavior in our data, we adopted signal-processing approaches like coupled wavelet/coherency analyses. In situ CO2 measurements allow us to visualize when root/microbial activity results in CO2 being respired from the soil surface, versus when other chemical/physical phenomena may alter gas pathways. While these approaches are interesting in understanding individual phenomena, they fail to get us beyond the study of individual processes.

Sensor data are compared with the outputs from ecosystem models to detect the patterns in specific phenomena or to revise model parameters or traits. For instance, we measured unexpected levels of soil CO2 in a tropical ecosystem. By examining small-scale ecosystem model parameters, we were able to pinpoint those parameters that needed to be altered to resemble the data outputs. However, we do not capture the essence of large-scale ecosystem shifts.

The time is right to utilize real-time data assimilation as an additional forcing of ecosystem models. Continuous, diurnal soil temperature and moisture, along with hourly hyphal or root growth could feed into well-established ecosystem models such as HYDRUS or DayCENT. This approach would provide instantaneous “measurements” of shifting ecosystem processes as they occur, allowing us to identify critical process connections more efficiently.