Fixed uncertainty for GEDI AGBD

Use this Forum to find information on, or ask a question about, NASA Earth Science data.
Post Reply
gcaria
Posts: 4
Joined: Mon Jun 29, 2020 11:02 am America/New_York
Answers: 0

Fixed uncertainty for GEDI AGBD

by gcaria » Fri Mar 04, 2022 4:04 am America/New_York

Hi,

I am using the L4A V2 GEDI data for an area in Colombia, and when I look at the uncertainty variable "agbd_se" I only see values which are either around 13 or around 3. It seems that while the "agbd" variable itself can vary across a wide range of values, the uncertainty "agbd_se" can pretty much be either 13 or 3.

Could you please help in understanding what I see?

Tags:

ORNL - mthornton
User Services
User Services
Posts: 20
Joined: Tue Jun 23, 2020 2:46 pm America/New_York
Answers: 0
Location: ORNL DAAC
Contact:

Re: Fixed uncertainty for GEDI AGBD

by ORNL - mthornton » Fri Mar 04, 2022 9:33 am America/New_York

@gcaria We have our GEDI subject matter expert working on your question. They will respond soon.

ORNL - shresthar
Subject Matter Expert
Subject Matter Expert
Posts: 3
Joined: Thu May 14, 2020 4:35 pm America/New_York
Answers: 0

Re: Fixed uncertainty for GEDI AGBD

by ORNL - shresthar » Fri Mar 04, 2022 1:50 pm America/New_York

Hi @gcaria, The `agbd_se` value should vary based on which prediction stratum the particular shot belongs to and its RH metrics. As an example, I have attached a plot of shots for one beam ('BEAM0110') of a GEDI L4A file `GEDI04_A_2020207182449_O09168_03_T03028_02_002_01_V002.h5`. In the figure, x-axis is the first predictor variable (xvar_0) and y-axis is agbd_se in Mg/ha.
If you can share the name of the files you are having the issue with, we can investigate them further. Thank you !
agbd_se.png
agbd_se.png (129.37 KiB) Not viewed yet

gcaria
Posts: 4
Joined: Mon Jun 29, 2020 11:02 am America/New_York
Answers: 0

Re: Fixed uncertainty for GEDI AGBD

by gcaria » Sat Mar 05, 2022 2:18 am America/New_York

Thanks for your answer.

In the plots you have posted I can see the same issue I'm reporting, that is agbd_se with an almost constant value (I'm assuming agbd itself actually does vary significantly in magnitude in the data used for your plots).

To be more specific about my case, I've made the attached histograms taking file "GEDI04_A_2019108110901_O01966_02_T02493_02_002_01_V002.h5", selecting only the full power beams, and clipping to a polygon defined by a WRS2 scene (path=9, scene=55, available here: https://www.usgs.gov/media/files/landsat-wrs-2-scene-boundaries-kml-file).

In order to find, download and clip the data I've followed the procedures outlined here: https://github.com/ornldaac/gedi_tutorials

Although the sample used to make these histograms is small, I see this pattern (agbd with wide range but agbd_se with an almost constant value of 13 Mg/ha, and less often 3 Mg/ha) consistently for all GEDI orbits data that fall within this polygon, and actually also for a much larger (5x) area in the same region of South America.
Attachments
Screen Shot 2022-03-05 at 8.16.21 am.png
Screen Shot 2022-03-05 at 8.16.21 am.png (12.14 KiB) Not viewed yet
Last edited by gcaria on Sun Mar 06, 2022 2:47 pm America/New_York, edited 1 time in total.

ORNL - mthornton
User Services
User Services
Posts: 20
Joined: Tue Jun 23, 2020 2:46 pm America/New_York
Answers: 0
Location: ORNL DAAC
Contact:

Re: Fixed uncertainty for GEDI AGBD

by ORNL - mthornton » Mon Mar 07, 2022 2:14 pm America/New_York

@gcaria, @ORNL - shresthar is working with the GEDI Science Team to investigate your question. A response is forthcoming.

ORNL - shresthar
Subject Matter Expert
Subject Matter Expert
Posts: 3
Joined: Thu May 14, 2020 4:35 pm America/New_York
Answers: 0

Re: Fixed uncertainty for GEDI AGBD

by ORNL - shresthar » Thu Mar 10, 2022 10:06 am America/New_York

Hi @gcaria , The GEDI Science Team has advised to use "agbd_t_se" to calculate prediction intervals. "agbd_se", which represents a back-transformed "agbd_t_se", might not be meaningful. The section around "In [33]" and "In [34]" in the tutorial shows how to compute the prediction intervals at different alpha levels. I hope this helps!

Post Reply