YAPC Overview

YAPC is a prototype photon classifer for the NASA ICESat-2 ATL03 Global Geolocated Photon product. It was developed by Jeff Lee (GSFC) with the goal of supporting and simplifying science applications for the NASA Ice Cloud and land Elevation Satellite-2 (ICESat-2).

YAPC is a customized inverse-distance kNN algorithm developed to determine the significance (or weight) of individual photon events. The weight of each photon is indicative of localized density based on it surrounding K neighbors and the inverse distances. For this prototype product:

  • Signal confidence is no longer surface-type specific

  • There is no penalty for sloped surfaces

YAPC Goals

  • Separate photon events that are likely signal from likely noise

  • Improve processing efficiency compared to the current histogram-based algorithm

  • Be significantly easier to maintain and extend for products and applications

  • Reduce the amount of sensitivity studies required for analysis

  • Possibly reduce the size of ATL03 product files

YAPC Algorithm

For each segment:

  • Calculate h_win_width as the spread of heights in a segment

  • Initialize photon weights to 0.0 and perform initial checks

    • Number of photons >= min_ph

    • Along-track spread >= min_xspread

    • Height spread >= min_hspread

  • Calculate the size of a dynamic selection window

    • Calculate density using the number of photons in the segment h_win_width and the span of the along-track distance

    • From the density, calculate the area necessary to contain at least min_ph photons

    • Calculate the horizontal (win_x) and vertical (win_h) window sizes for the aspect ratio

  • OR dynamically calculate the number of neighbors * Calculate k as half of the square root of the number of photon events

  • For each source photon in the segment, calculate inverse distances from its neighbors

  • Select and count target photons whose along-track distance and height are within win_x/2 and win_h/2 of the source photon

  • Calculate and record the inverse distances for the k nearest photons

inv_dist =  win_x/2 - abs(delta(x)) + win_h/2 - abs(delta(h))

  • Calculate the weight of each source photon as the sum of the k largest inv_dist values

  • Normalize weights by dividing by half of the perimeter of the window (win_x/2.0 + win_h/2.0) and the k value

For the python version, the distances between photons can be calculated using:

Credits

These notes are based upon presentations by Jeff Lee to the ICESat-2 Land Ice Science Team.