Theil-Sen estimated median change in rain normalised soil moisture 2001-2016, Indonesia

Map: Theil-Sen estimated median change in rain normalised soil moisture 2001-2016, Indonesia

Model OSSL data: 4 manual settings

Thomas Gumbricht bio photo By Thomas Gumbricht

This is the last post in a series on organising and analysing data from the Open Soil Spectral Library (OSSL). It is also the last post in a sub-series on how to apply Machine Learning (ML) for predicting soil properties from spectral data. To run the scripts used in this post you need to setup a Python environment, and clone or download the python scripts from a GitHub repository (repo), as explained in the post Clone the OSSL python package.

Introduction

In the previous posts you developed models for predicting different soil properties from spectral data. If you applied the different options for selecting a subset of features and/or hyper-parameter tuning, you have probably discovered that a subset of the original data is sufficient for predicting a particular soil property. And that each predicted property relies on a different set of covariates (spectra and spectra derivatives). You will most likely also have discovered that a particular regressor gives better result for each target feature.

This last post describes how you can use the knowledge you gained on the combination of covariates, regressor and model parameterisation for a-priori definition of models for the target properties.

You can manually define a set of a-priori covariates and then apply hyper-parameter tuning using only this limited set. Doing so will speed up the hyper-parameter tuning.

You can also use manual setting to specifically select covariates that theoretically or empirically relate to the target feature you aim at predicting, and then test the relative performance of the different regressors, with or without hyper-parameter tuning.

Prerequisites

To follow the hands-on instructions, this post requires that you completed the processing as outlined in the posts on downloading and importing the OSSL spectral data. You must also have access to a Python interpreter with the packages matplotlib and scikit learn (sklearn) installed.

Manually defining the covariates

In the json command file used for parameterising the process-flow of the OSSL_mlmodel module you can define a fixed set of spectral wavelengths and wavelength derivates as the covariates to use. If you do that, all the feature selection steps are removed from the process-flow:

"manualFeatureSelection": {
    "comment": "Manual feature selection overrides other selection alternatives",
    "apply": true,
    "spectra": [
      "675",
      "685",
      "705",
      "715",
      "735"
    ],
    "derivatives": {
      "firstWaveLength": [
        "675"
      ],
      "lastWaveLength": [
        "735"
      ]
    }
  },

While all feature selection options are automatically removed, even if you requested them, when you apply a “manualFeatureSelection”, you can still apply:

  • Feature importance evaluation,
  • Hyper-parameter tuning, and
  • Model fitting and evaluation.

You can also choose to change the default definition of the hyper-parameters for the regressors you want to apply with the manually selected features. You do this by adding the “hyperParams” settings in the json command file (see the section on Regression models in post on the process-flow.