Toward sustainable compost use: Prediction of organic matter via smartphone image analysis

Abstract

Increased global emphasis on environmental sustainability and soil health requires efficient, accessible tools to evaluate compost organic matter (OM), a key contributor to soil quality and carbon/nutrient cycling. This study used smartphone image analysis to predict compost OM as an innovative, cost-effective alternative to laboratory methods. Utilizing 157 compost samples across North America, this research integrated smartphone-acquired images and machine learning (specifically, random forest models applied to features such as color, texture, spatial descriptors, and geographic location extracted from the images) to predict OM content. Results showed that dry samples yielded robust predictive performance (validation R² = 0.75, root mean square error [RMSE] = 5.63%, ratio of performance to inter-quartile distance [RPIQ] = 2.97); moist samples faced challenges due to moisture-induced variability (validation R² = 0.35, RMSE = 9.14%, RPIQ = 1.83). The better performance of dry samples was attributed to reduced surface reflectance and more stable visual features, which allowed for more accurate prediction—highlighting the importance of pre-processing in practical applications. Integrating color, texture, spatial features, and geographic location enhanced model accuracy, underscoring the importance of regional variability in compost characteristics. This smartphone-based method empowers compost producers—especially those without access to laboratory facilities—to conduct rapid, nondestructive, and on-site compost quality assessment.