Harnessing the power of AI technologies for ecology: The knowledge learning analysis system (KLAS) for spatially-distributed, continuous ecological data

TitleHarnessing the power of AI technologies for ecology: The knowledge learning analysis system (KLAS) for spatially-distributed, continuous ecological data
Publication TypeConference Paper
Year of Publication2020
AuthorsPeters D, Ramirez G, Anderson J, Savoy HM, Huang H
Conference NameEcological Society of America
Date Published08/2020
Conference LocationVirtual Conference
ARIS Log Number378827
Abstract

Background/Question/Methods
Sensors are increasingly being used to capture information about biotic processes and environmental conditions at finer spatial and temporal resolutions across broader spatial extents than possible from data collected manually. This sensor-based information is leading to new insights about ecological systems through the integration of data across scales, disciplinary interests, and levels of organization. However, sensors can also produce large volumes of different types of data at higher frequencies for more spatially-distributed locations than ecologists are used to handling or managing. Artificial intelligence (AI) technologies provide one approach to effectively distilling, managing, and analyzing these large and diverse amounts of data to be used for ecological problems. Our objectives were to develop an AI system to evaluate and learn from streaming data that: (1) would automate the QA/QC of sensor data, and (2) would be sufficiently flexible to account for changes in climate through time. We developed an AI system in R and MatLab combined with the GCE Toolbox. We analyzed sensor data from 15 meteorological stations spatially-distributed at the Jornada USDA-LTER site in southern New Mexico. We then tested the scalability of our AI system by applying it to an additional 60 spatially-distributed meteorological stations locations at the Jornada, and to primary production data.
Results/Conclusions; Our AI system effectively assimilated and analyzed large volumes of streaming data from meteorological stations, and provided level one QA/QC data for all 15 locations. The data were uploaded to a national data repository (EDI) with ecological metadata within one day following data collection. These metrics contrast with our previous manual approach to checking and correcting the data that resulted in monthly time lags before uploading the data. The AI system was successfully applied to an additional 60 locations with meteorological stations, and to primary production data. Given the automated data collection and AI computational process, adding more stations did not increase the time required for checking and uploading the data to EDI, and the errors associated with the data did not increase. Recent additions to our approach to identifying outliers from extreme values based on comparisons with nearby locations result in an approach that is sufficiently flexible to account for changes in climate through time. Our AI system is a dynamic approach that can be applied to other data types, locations, and ecosystems where non-stationarity through time is important. Poster #85379.