Using Landsat and ECOSTRESS land surface temperature to parameterize sub-grid tiling schemes and enable tile-level calibration of land surface models

Abstract:

The ever-increasing availability of hyperresolution (<100 m) global environmental datasets from satellite remote sensing is leading to significant improvements in the parameterization of sub-grid heterogeneity in Earth system models (ESMs) (i.e., tiling schemes). However, these improvements have generally not been accompanied by appreciable improvements in model performance. This discrepancy is related to the remaining tile-level model structural, parameter, and forcing uncertainties. As such, there is a growing need to develop tile-level calibration techniques that can fully leverage the contemporary advances in tiling schemes. In this presentation we show how field-scale (<100 m) satellite retrievals of land surface temperature (LST) can be used to address this persistent challenge. In specific, we used the time-varying observed field-scale estimates of LST from Landsat and ECOSTRESS to improve the definition of sub-grid tiles and to tune the tile-level parameters in the HydroBlocks land surface model.

This study is performed over the ARM Southern Great Plains site in central Oklahoma in the United States. The HydroBlocks LSM is setup to run from 2010 to 2019 at an effective 30-meter spatial resolution and hourly temporal resolution. The datasets used to parameterize and force the model include the Princeton Climate Forcing meteorological dataset (PCF), POLARIS soil dataset, and the National Land Cover dataset (NLCD). To improve the tiles representation of the landscape, we modified the cluster algorithm used to assemble HydroBlocks’ sub-grid tiles to also include the observed 100-meter temporal mean and standard deviation of LST from Landsat and ECOSTRESS. To identify the best performing tile-specific LSM parameters, a 250-member Latin Hypercube sample is then used to run the LSM with varying model parameters per sub-grid tile. This approach illustrates the value of leveraging existing high-resolution satellite-based datasets of states and fluxes to improve tile configuration and to calibrate LSM parameters at the tile-level. We will finalize by discussing how this approach can then be used to help field-scale resolving land surface model to meet its intended goal to provide reliable and robust field-scale estimates of the water, energy, and carbon cycles.