GRIME AI GCC Validation Against PhenoCam

Overview

This dataset provides validation evidence for green chromatic coordinate (GCC) calculations produced by GRIME AI (GaugeCam Remote Image Manager Educational Artificial Intelligence). Two comparison analyses are included: (1) cross-platform agreement between GRIME AI and the PhenoCam Network pipeline (Northern Arizona University, benchmark reference), and (2) within-platform consistency between GCC computed on in-memory images versus those same images after saving to disk and reloading as JPEG. GCC outputs from each platform were manually compiled into a spreadsheet for comparison.

Contents

``` Software - Generate Test Images/ create_greenness_test_images.py Image generation script create.bat Windows batch launcher requirements.txt Python dependencies test_images/ masks/ 4 mask images (white/black, RGB/grayscale) in JPG, PNG, and TIF solids/ flat/ 65 solid flat-fill green JPEGs + 20250718_1200_Solid_Flat_Images_Info.xlsx textured/ 65 solid stippled green JPEGs + 20250718_1500_Solid_Textured_Images_Info.xlsx splotches/ flat_flat/ 32 JPEGs: flat clay background, flat green splotches + Splotch_flat_flat_Images.xlsx flat_textured/ 32 JPEGs: flat clay background, stippled green splotches + Splotch_flat_textured_Images.xlsx textured_flat/ 32 JPEGs: stippled clay background, flat green splotches + Splotch_textured_flat_Images.xlsx textured_textured/ 32 JPEGs: stippled clay background, stippled green splotches + Splotch_textured_textured_Images.xlsx Test Files Sent or Uploaded to NAU on 2025_07_18/ Flat and stippled solid JPEGs + metadata XLSXs submitted to NAU PhenoCam for GCC computation

Test Results & Analysis/ GCC - Test Results - Memory Image vs. File Image (JPG) Differences.xlsx GCC outputs from GRIME AI (memory and file) and PhenoCam (file) with computed differences ```

Synthetic Image Design

All images are 1024 x 768 pixels, saved as JPEG at quality=100. Images were generated on 2025-07-18 using create_greenness_test_images.py.

Solid flat images — 65 images. Colors span an 8 x 8 grid of HSV saturation (0.2–1.0) and value (0.2–1.0) at a fixed green hue (100°), plus one pure green [0, 255, 0] image. Each image is a uniform fill of a single color. These images were used in the GCC comparison analyses.

Solid stippled images — 65 images. Same color grid as the flat set. Each image applies Gaussian-blurred random noise (amplitude ±20 DN) to the base color, producing a stipple texture. These images were used in the GCC comparison analyses.

Splotch images — 32 images per permutation (4 permutations), not used in the GCC comparison analyses. Each image composites 50 circular green splotches of randomized radius (20–100 px), saturation (0.2–1.0), and value (0.2–1.0) onto a clay-colored background (RGB: 160, 82, 45). Background and splotch patches are independently flat or stippled across the four permutations.

Mask images — 4 images covering all combinations of fill color (white, black) and channel depth (1-channel grayscale, 3-channel RGB), each saved in JPG, PNG, and TIF formats.

Workflow

The solid flat and solid stippled images were submitted to the PhenoCam Network (NAU) for independent GCC computation. The files submitted are preserved in Test Files Sent or Uploaded to NAU on 2025_07_18/. GCC values returned by PhenoCam and computed independently by GRIME AI were manually entered into the comparison spreadsheet.

Validation Results Summary

Cross-platform differences are attributed to floating-point rounding variation across CPU architectures. Memory-vs-file differences are attributed to JPEG lossy compression (e.g., [0, 255, 0] → [0, 255, 1] after a save-reload cycle).

Reproducing the Test Images

bash pip install -r requirements.txt python create_greenness_test_images.py

Output is written to test_images/ with subdirectories for solids, splotches, and masks.

Software

GRIME AI source code: https://github.com/JohnStranzl/GRIME-AI

Citation

Stranzl Jr., J. E. (2026). GRIME AI GCC Validation Against PhenoCam, HydroShare. [DOI pending]

License

Creative Commons Attribution CC BY 4.0 — http://creativecommons.org/licenses/by/4.0/

Code is licensed under the Apache License 2.0.

Funding

This work was supported by the U.S. National Science Foundation under Award No. 2411065 — Innovative Resources: Cyberinfrastructure and community to leverage ground-based imagery in ecohydrological studies.