Emulators are regression models that are able to approximate the processing of an RTM, although at a fraction of the computational cost. To enable emulation of an RTM, the first step involves building a statistically-based representation (i.e. an emulator) of the RTM from a set of training data points derived from runs of the actual RTM. The second stage uses the emulator built in the first step to compute the output that otherwise would be generated by the RTM. The Emulator toolbox is foreseen to open new opportunities in the use of advanced RTMs, in which both consistent physical assumptions and data-driven machine learning algorithms live together.
ARTMO’s Emulator toolbox v.1.10.
Currently, the following multi-output MLRAs are included:
- Partial Least Squares Regression (PLSR)
- Neural networks (NN)
- Kernel Ridge Regression (KRR)
- Multi-output Gapussian processes regression (MO-GPR)
- Multi-output Support Vector Regression (MO-SVR)
And the following MLRAs (to be used together with PCA):
- Random Forests (FR)
- Support Vector Machine (SVR)
- (VH) Gaussian processes regression (GPR)
In short, the Emulator toolbox enables:
- To apply and evaluate multiple MLRAs according to customized training strategies, e.g. with different noise and train/validation partitioning.
- Data can either come from ARTMO-RTMs or coming from external LUTs imported as text file.
- The MLRAs are evaluated on their emulator capacities using RMSE goodness-of-fit statistics.
- The best performing MLRA can then function as emulator. Its performance can be tested against the original RTM.
- Emulators can approximate RTMs and generate LUTs at a tremendous gain in processing speed.