Comparative performance of tristimulus and color appearance models in low-cost smartphone digital colorimetry for...

The verification of carbon capture and storage (CCS) sites and the monitoring of soil health demand cost-effective and scalable methods for detecting carbon dioxide (CO₂) leaks. While smartphone-based digital colorimetry offers a promising alternative to traditional analytical techniques, its accuracy is fundamentally constrained by the choice of color representation model, especially under variable field conditions. This study presents a systematic comparison between traditional tristimulus color spaces (sRGB, CIELAB, CIE XYZ) and advanced Color Appearance Models (CAMs including CIECAM02, CIECAM16, and CAM16UCS) for quantifying soil CO₂ concentrations using a phenol red-based optical sensor and machine learning regression. A phenol red sensor was exposed to known CO₂ concentrations (0-7.8729 wt%) in a controlled soil environment. Sensor images captured by a smartphone were processed to extract color data from various color spaces and CAMs. The performance of multiple machine learning models, including Partial Least Squares (PLS) and Bayesian ridge regression, was evaluated using leave-one-out cross-validation. The method was evaluated under controlled laboratory conditions using simulated soil CO₂ leakage scenarios. Results demonstrate that CAMs, particularly CAM16UCS, significantly outperform tristimulus models. The CAM16UCS-PLS combination achieved a superior limit of detection (LOD) of 0.0427 wt% CO₂, surpassing both the spectrophotometric benchmark (0.0780 wt%) and the best tristimulus model, CIELAB-PLS (0.0463 wt%). This enhanced performance is attributed to the superior perceptual uniformity, hue linearity, and adaptive viewing condition compensation inherent to CAMs, which provide a more robust colorimetric signal for machine learning interpretation. This work establishes CAM-integrated digital colorimetry as a highly sensitive, low-cost, and laboratory-validated protocol that offers rapid analysis capabilities for environmental monitoring, CCS verification, and precision agriculture.