lalligagger / snr_modeling

early stab at thermal instrument SNR, may integrate w/ Pyxel

snr_modeling

SNR Modeling Introduction

Goals of model:

1. Develop a single pixel model of VIS-NIR & thermal observations. Calculate the at-lens radiance for each band. Inputs for this are:

   • Reference solar spectrum
   • Blackbody temperature
   • Scene reflectance/ emissivity
   • (Optional) Atmospheric transmission

2. From first-order instrument parameters, calculate optical power from scene onto pixel and signal-to-noise ratio (SNR)/ or noise-equivalent differential temperature (NEDT)​.
   This requires first order optical, filter, and detector properties:

   • Focal length
   • Aperture Diameter
   • Pixel pitch
   • Detectivity

3. Parameterize the pixel detectivity based on operational / environmental parameters (e.g. integration time, temperature).​
   Also, include a first-order contribution for lens and filter thermal emission (bb curve * residual coating emissivity), which will drive calibrations.
   Generate SNR/ NEDT over range of detector parameters​.

4. Run sensitivity studies based on scene model parameters, as well as system conditions of (3).

Phase 1: Scene Signal Modeling

Blackbody radiance calculations complete with dimensional analysis (using Pint) are being developed in the HYTRAN module.
In the meantime, an end-to-end SNR modeling prototype will be developed using a combination of validated pyspectral radiance functions and HYTRAN instrument models. (More on the complete planned HYTRAN capability later.)

The below image is a great high-level example of what we're after. This is a visual comparison of the Landsat 7 & 8 vs. the Sentinel-2 bands vs. wavelength, superposed on a standard atmospheric transmission spectrum.

Atmospheric spectra will be a part of our model, but we will go one step further and dynamically model the signal of interest over the visible, NIR/SWIR bands dominated by solar reflectance as well as the thermal signal of longer wavelengths.

Dash app specific functions:

• Pick a reference spectrum from a dropdown
• Set a ground temperature
• Select a reference atmosphere, once we have data
• Pick an instrument band selection for display, as in above plot

All of the data for this phase can be conveniently held on the same x-axis of wavelength (we'll use microns). The bands for e.g. Landsat (except TIRS) & Sentinel-3 are already available through Pyspectral. ECOSTRESS/ ASTER reference spectra are available through SPy. See spy_basics.ipynb for simple interface setup, or NEDT_calcs.ipynb for a more complete treatment with HYTRAN.
The script spy_scene.py is a much simpler reworking of the above studies, to enable quick development of a Dash application along the same lines. It outputs six total curves based on user-selected temperature, wavelength domain, and a reference reflecatance spectra. (Uses the standard TOA for solar and currently no atmosphere.)

Below is a plot of what spy_scene.py currently produces. Everything is all mashed up in one dataframe/ plot which needs some cleaning up!

Dependencies

HYTRAN

In-house python module for scene and instrument configuration to run performance modeling. Will support the inclusion of various spectral and atmospheric data once interfaces are defined.

Pyspectral

A helpful if basic package that has similar goals for scene modeling as HYTRAN, plus some spectral response curves of various Earth observing instruments. Currently used features:

• Spectral Response Functions
  See spy_basics.ipynb for a test case with Sentinel-3 bands.

* [Solar TOA Reference Spectrum and Blackbody Generation](https://pyspectral.readthedocs.io/en/master/rad_definitions.html) See the above spaghetti plot for an idea of how this all comes together.

Spectral Python (SPy)

Containes ECOSTRESS spectra interface, basic set of hyperspectral imaging algorithms.

• https://www.spectralpython.net/libraries.html#ecostress-spectral-library
  See spy_basics.ipynb for a test case of a spectrum lookup. Need a good user interface for Dash app.