Introduction to radiative transfer modelling

J Gómez-Dans (UCL & NCEO)

j.gomez-dans@ucl.ac.uk

Introduction to modeling lectures

  1. Motivation
  2. Modelling leaf optical properties
  3. Modelling canopies in the optical domain

Motivation

  • Model: formulation of physical principles that produce signals you measure.
  • Models are simplifications of reality, and thus are not reality
    • We need to know strengths AND weaknesses of models
  • In EO, we are interested in processes that describe how photon-atmosphere-canopy interactions $\Rightarrow$ radiative transfer theory (RT)
  • Some interesting quotations (H/T Mat Disney@UCL

“All models are wrong but some are useful” – George Box

“The purpose of models is not to fit the data but to sharpen the questions” – Samuel Karlin

“No one trusts a model except the person who wrote it. Everyone trusts an observation except the person who made it.” – Anon.

Some uses of physcal models

  • Inform data collection
    • Observng System Observation Experiments (OSSEs)
    • Satellite mission simulators
  • Consistent treatment of data from different instruments
    • Can treat spectral, angular and spatial acquisition characteristics
    • Data pre-processing and conditioning
  • Consistent use of data from e.g. microwave, optical and thermal domain.
  • Parameter retrieval
    • The model can be used to interpret observations in terms of model input parameters
  • Interpolation/Extrapolation
    • Extend observations
  • Understand sensitivity of observation

  • modelling entails assumptions
  • You need to understand the assumptions
  • ... and assess whether they are acceptable for your application!

Physical models and EO

  • Typically want to monitor processes over vast areas, different time scales...
  • EO is an indirect measurement, so data we acquire needs to be interpreted
  • Eg: we measure radiance from a sensor
    • ... but you said you were interested in delimiting land use!
    • ... bit you said you were interested in assessing fire danger!
  • Data interpretation can be done in two ways:
    • Empirical models: measure magnitude of interest $\leadsto$ pairing with EO data $\leadsto$ finding a statistical relationship
    • Mechanistic models: by considering the processes that produce the observations, and using this understanding to quantify variables of interest.

An example of an empirical model

Ouyang et al (2012)

*From Ouyang et al (2012)

An example of a mechanistic model

Lewis' millet raytrace

Assumptions!

  • In both cases, there are assumptions:
    • An empirical model is typically derived for a particular
      • Location
      • Time
      • Sensor
    • A mechanistic model makes assumptions on
      • processes included and ignored
      • simplifications of the processes
  • Physical models explicitly enconde basic properties of system
    • Energy conservation
    • Reciprocity

Model error

  • Observations have experimental error (e.g. errorbars!)
  • Model error encodes the price that the model inadequacy in predicting reality
  • It is usually very hard to assess formally
  • ... and thus often disregarded