Rapid prediction of poly(butylene adipate-co-terephthalate)/poly(glycolic acid) (PBAT/PGA) agricultural films based on UV-accelerated aging tests with applicability to the environment

Biodegradable plastic mulches (BPMs) possess great possibility as alternative materials for traditional non-degradable agricultural films. However, research on the degradation behaviors of biodegradable films remains relatively nascent, which is a crucial determinant in applications. Ultraviolet accelerated aging method offers an effective approach to simulate the outdoor or field degradation in a shortened period. In this research, poly(butylene adipate-co-terephthalate)/poly(glycolic acid) (PBAT/PGA) films were prepared and subjected to UV-accelerated degradation (UAD) and natural environmental degradation (NED). The variation of performance parameters including haze, transmittance, tensile strength, elongation at break and melting temperature were monitored at varying degradation intervals. Due to the UV-accelerated aging experimental conditions were well matched with natural environmental factors, the data derived from UAD and NED were highly correlated, indicating the feasibility of predicting film properties based on the UAD test. Random forest algorithm displayed superior stability and high accuracy in constructing degradation prediction model, achieving R 2 of 0.984 and 0.979 for training and test sets, respectively. Equations derived from this model demonstrated the mapping between NED days and UAD days, which facilitated a rapid evaluation of film out-door performance by indoor UV-accelerated aging tests. Machine learning provides a novel and efficient approach for constructing degradation prediction models, which can enhance the adoption of biodegradable films and thus contribute to addressing the plastic pollution problems in agriculture. • Biodegradation performance of PBAT/PGA films under NED and UAD evaluated. • Relationship of performance parameters between NED and UAD correlated. • Prediction models of films degradation property established by machine learning based on UAD.