Digital images, color spaces, color appearance models and performance metrics data - Chairul Ichsan et al

This dataset accompanies the manuscript "Comparative Performance of Tristimulus and Color Appearance Models in Smartphone Digital Colorimetry for Soil CO2 Detection" by Chairul Ichsan, Khoirun Nisa, Siti Rodiah, and M. Mahfudz Fauzi Syamsuri (submitted to Talanta). It provides raw and processed data from experiments using a phenol red-based optical sensor exposed to controlled CO2 concentrations (0–7.8729 wt%) in soil, captured via smartphone for digital colorimetry analysis. Key components include: Color parameters extracted from tristimulus spaces (sRGB, CIELAB, CIE XYZ) and color appearance models (CIECAM02, CIECAM16, CAM16UCS, ZCAM), such as lightness, chroma, hue angle, etc. Machine learning regression performance metrics (RMSE, MAPE, R², LOD) for models like Partial Least Squares (PLS) and Bayesian Ridge, evaluated with Leave-One-Out Cross-Validation. Embedded sample digital images of the sensor at various CO2 levels. The data is structured in an XLSX file with three sheets: "color spaces, CAMs & metrics" (raw parameters and error analysis), "ML reg performance" (model benchmarks and LOD notes), and "digital images samples" (concentrations with embedded images). This dataset enables reproduction of the study's findings on CAM superiority for low-cost, field-deployable CO2 monitoring in CCS verification and precision agriculture. For full details, refer to the README.txt file included in the upload. Data collection methods: Controlled chamber experiments with high-purity CO2 (99.9%), phenol red sensor, and Samsung Galaxy A12 smartphone imaging. Processing used Python libraries (Pillow, Scikit-Image, Colour-Science). Limitations: Small sample size (n=6 concentrations); metrics are estimates. Keywords: Carbon Dioxide Monitoring, Color Appearance Model (CAM), Digital Colorimetry, Smartphone Sensor, Soil Gas Detection, Machine Learning Regression. Funding: None.