A hybrid machine learning approach for calibration and regionalization of LSM soil and vegetation parameters
Description
This work explores the combined use of machine learning and traditional model calibration methods to develop a high resolution (1 km) soil and vegetation parameter dataset for North America and Central America. First, we use a traditional calibration method (i.e., genetic algorithm (GA)) to calibrate soil and vegetation parameters to Soil Moisture Active Passive (SMAP) soil moisture at 25 km spatial resolution over the CONUS, Asia, and Europe. Next, we use a series of machine learning algorithms, including Neural Network, Random Forest, and XGBoost, to downscale and regionalize the optimized parameters for the National Land Data Assimilation System - Phase 3 (NLDAS-3) domain (i.e., North and Central America). The goal is to develop high resolution parameters that produce a land surface model (i.e., Noah-MP) soil moisture climatology that is more in-line with SMAP’s SM climatology.
Detailed example
The system outputs are soil and vegetation parameters at 1 km resolution.
AI / analytics pattern
Classical/Predictive Machine Learning: Models trained on data to make predictions or classifications based on identified patterns or relationships.
Automation level / stage
a) Pre-deployment – The use case is in a development or acquisition status.
Expected benefit
The ultimate benefit of this work is that it could improve land surface model simulations of the water cycle through more efficient remote sensing data assimilation (DA). Current DA methods rely on bias correction and CDF matching to translate observations (SMAP) into the same climatology as the model. However, this often filters out (as noise) useful signal about the human component of the water cycle, such as irrigation, in the observations. The ultimate goal of our work is to develop parameters that generate a model soil moisture climatology similar to SMAP, possibly bypassing the need for bias correction. As such, it may allow us to more readily incorporate information that SMAP provides about irrigation and other non-geophysical activities that are difficult to model, through DA of NASA remote sensing observations.
Controls / human review
ATO: Not reported; PIA: Not published