Скачать или смотреть Exploring Crop Variability - Analyzing NDVI Distributions with Cumulative Distribution Functions

In the video, Jesse discusses a project analyzing an area of interest (area with a history of disturbed soils via excavaiton) using NDVI (Normalized Difference Vegetation Index) data.

I've been struggling with 2 large concepts:
1. How to use statistical analysis with high resolution multispectral maps, because the incredibly dense point data means that normal significance tests tend to always be significant even with minor changes due to the extremely high sample size.
2. How to quantify the amount of variability in a field - since typical central tendency statistics like mean and median, do not do a good job of describing relatively small bare areas that get lost in the huge number of data points being considered, but are very important when assessing crop growth in a reclamation or remediation scenario.

In this demonstration I compare an area of interest to a buffer zone and explore the natural variability in the field through Zonal Statistics and cumulative distribution functions (CDFs). Using ‪@replit‬ and Claude Sonnet V2 and ChatGPT 4o and o1-preview - I built a ‪@streamlitofficial‬ webapp on Replit which generates Cumulative Distribution Functions to visualize NDVI distributions and employs Earthmover's Distance to quantify differences between these distributions. The analysis reveals that the mean NDVI for the area of interest is lower than background area. And that although the mean is only slightly lower than background, it falls outside the "normal" amount of variability seen within the field. Jesse invites feedback and collaboration on this analytical approach.

I invite any comments or disucssion into the concepts discussed. I know that I have a lot of work to do to see if this concept holds water, but i thought it was an interesting idea to flesh out, and it was an excellent way to use my AI enababled coding skills to calculate, visualize and play with the data.

I was thinking about adding Coefficent of Variation and Cohen's D as helping to quantify these things further.

https://github.com/lawrencejesse/CDF-...