Daniel L Warner
University of Delaware;Delaware Geological Survey
| Subject Areas: | Hydrology, greenhouse gas dynamics, soils |
Recent Activity
ABSTRACT:
The contained grids were derived by applying a novel downscaling methodology to the coarse, remotely-sensed ESA Climate Change Initiative Soil Moisture Product version 4.5. We employed an ensemble of kernel K-nearest neighbors models to refine the grid cell resolution from 27 km to 100 m using ancillary terrain data and interpolated meteorological observations. The downscaled grids were validated against independent surface soil moisture (SSM) network observations, which revealed an improved performance over the low resolution grids. Performance was further improved when grid cell values were normalized and then rescaled based on the daily SSM minima and maxima observed by the monitoring network. The downscaled grids are independent of vegetation and land cover data, allowing for investigations into spatial relationships between SSM, vegetation and land cover.
For a code example of the KKNN approach, see: https://github.com/warnerdl/DownscalingDelawareSSM
Contact
| (Log in to send email) |
| All | 0 |
| Collection | 0 |
| Resource | 0 |
| App Connector | 0 |
Created: Nov. 20, 2020, 7:21 p.m.
Authors: Warner, Daniel · Guevara, Mario · John Callahan · Rodrigo Vargas
ABSTRACT:
The contained grids were derived by applying a novel downscaling methodology to the coarse, remotely-sensed ESA Climate Change Initiative Soil Moisture Product version 4.5. We employed an ensemble of kernel K-nearest neighbors models to refine the grid cell resolution from 27 km to 100 m using ancillary terrain data and interpolated meteorological observations. The downscaled grids were validated against independent surface soil moisture (SSM) network observations, which revealed an improved performance over the low resolution grids. Performance was further improved when grid cell values were normalized and then rescaled based on the daily SSM minima and maxima observed by the monitoring network. The downscaled grids are independent of vegetation and land cover data, allowing for investigations into spatial relationships between SSM, vegetation and land cover.
For a code example of the KKNN approach, see: https://github.com/warnerdl/DownscalingDelawareSSM