Core Concepts
An approach integrating spatiotemporal fusion and machine learning to upscale point-scale soil moisture measurements to a 100 m resolution, enabling improved spatial evaluation of model simulations and satellite retrievals.
Abstract
The study presents an upscaling approach that combines spatiotemporal fusion and machine learning to extrapolate point-scale soil moisture (SM) measurements from 28 in-situ sites to a 100 m resolution for a 100 km × 100 km agricultural area.
Key highlights:
- Spatiotemporal fusion was used to downscale MODIS albedo, NDVI, and land surface temperature (LST) to 100 m resolution, improving the representation of agricultural landscapes.
- The eXtreme Gradient Boosting (XGBoost) model was employed for the ML-based upscaling of in-situ SM measurements.
- Four-fold cross-validation and cross-cluster validation were conducted, demonstrating consistent correlation performance ranging from 0.6 to 0.9.
- The upscaling approach was able to capture the spatial variability of SM within areas not covered by in-situ sites, with correlation performance between 0.6 and 0.8.
- The importance of regional nuances in training strategies was highlighted, with the cross-cluster training revealing more pronounced SM variations compared to the global training.
The proposed upscaling approach offers a pathway to extrapolate point-scale SM measurements to a spatial scale more comparable to climatic model grids and satellite retrievals, enabling improved evaluation of these datasets.
Stats
The study area receives an average annual precipitation of approximately 400 mm.
The in-situ SM data was collected from the OzNet Hydrological Monitoring Network, with 28 sites used for the upscaling and validation.
The geospatial predictors used for upscaling include indices and surface temperature data from MODIS and Landsat, climatic variables from ANUClimate 2.0, Smoothed Digital Elevation Model, and soil data from the Soil and Landscape Grid Australia.
Quotes
"The upscaling of the in-situ measurements to a commensurate resolution to that of the modelled or retrieved SM will lead to a fairer comparison and statistically more defensible evaluation."
"The cross-cluster validation underscored the capability of the upscaling approach to map the spatial variability of SM within areas that were not covered by in-situ sites, with correlation performance ranging between 0.6 and 0.8."
"The observed variations in SM across the landscape underscore the importance of considerations of regional nuances in training strategies for accurate and context-specific predictions."