Measuring the strength of earth surface process feedback from the air: new tools for quantifying dominant landscape processes and their timescales
Abstract
Earth surface process feedbacks are responsible for the genesis of patterned landscapes, long-term stability of geomorphic features, or rapid change in response to small external perturbations. Traditionally, understanding of the feedback processes driving a particular behavior has been gained deductively, by testing the plausibility of hypothesized mechanisms through extensive numerical modeling or field observation and experimentation. However, with the improvement of the resolution and types of data observable through remote sensing, together with the infusion of new techniques for inductive learning in the earth sciences, it is now possible to delineate the significance, strength, and temporal characteristics of the interactions between pairs of variables for which large spatial or temporal datasets are available.
We applied an information-theory approach to evaluate the extent to which vegetation canopy characteristics reduce uncertainty in sedimentation and erosion and vice-versa in a deltaic floodplain (Wax Lake Delta, Louisiana) and inland floodplain (Big Spring Run, Pennsylvania), based on airborne and terrestrial lidar imagery. In the deltaic floodplain, we found that only native emergent and submersed vegetation species served as dominant controls on elevation, whereas in all vegetation communities, canopy volume served as a significant control on rates of sedimentation. In the inland floodplain, none of the vegetation classes present induced sufficient deposition or erosion to exert an effect on elevation distinguishable from other driving processes (e.g., sediment supply), but, as in the deltaic floodplain, canopy volume across all vegetation types controlled rates of sedimentation. These same tools, when applied to time series, also reveal critical timescales over which one variable controls another. When applied in a data pre-processing context, this information can pinpoint timescales of input data aggregation and time lags needed for generating forecasts using machine learning. Using the Dry Creek experimental watershed (Idaho) as a case-study, we demonstrate how this approach leads to rapid convergence on well-performing stream discharge forecast models compared to traditional approaches for training models.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFMEP43B..01M
- Keywords:
-
- 1824 Geomorphology: general;
- HYDROLOGYDE: 1942 Machine learning;
- INFORMATICSDE: 4217 Coastal processes;
- OCEANOGRAPHY: GENERALDE: 4558 Sediment transport;
- OCEANOGRAPHY: PHYSICAL