The big data-big model (BDBM) challenges in ecological research

Abstract:

The field of ecology has become a big-data science in the past decades due to development of new sensors used in numerous studies in the ecological community. Many sensor networks have been established to collect data. For example, satellites, such as Terra and OCO-2 among others, have collected data relevant on global carbon cycle. Thousands of field manipulative experiments have been conducted to examine feedback of terrestrial carbon cycle to global changes. Networks of observations, such as FLUXNET, have measured land processes. In particular, the implementation of the National Ecological Observatory Network (NEON), which is designed to network different kinds of sensors at many locations over the nation, will generate large volumes of ecological data every day. The raw data from sensors from those networks offer an unprecedented opportunity for accelerating advances in our knowledge of ecological processes, educating teachers and students, supporting decision-making, testing ecological theory, and forecasting changes in ecosystem services.

Currently, ecologists do not have the infrastructure in place to synthesize massive yet heterogeneous data into resources for decision support. It is urgent to develop an ecological forecasting system that can make the best use of multiple sources of data to assess long-term biosphere change and anticipate future states of ecosystem services at regional and continental scales.

Forecasting relies on big models that describe major processes that underlie complex system dynamics. Ecological system models, despite great simplification of the real systems, are still complex in order to address real-world problems. For example, Community Land Model (CLM) incorporates thousands of processes related to energy balance, hydrology, and biogeochemistry. Integration of massive data from multiple big data sources with complex models has to tackle Big Data-Big Model (BDBM) challenges. Those challenges include interoperability of multiple, heterogeneous data sets; intractability of structural complexity of big models; equifinality of model structure selection and parameter estimation; and computational demand of global optimization with Big Models.