Integrating machine learning and model-based causal attribution to decipher stage-specific trade-offs in nitrogen...

It is difficult to reconcile nitrogen mitigation with maturation due to nonlinear interactions and temporal autocorrelation. This study integrated machine learning, background-shuffling causal attribution, and phased meta-analysis to decode stage-specific mechanisms in 513 samples. Rank normalization within each study ensured robust generalization, resulting in R² values exceeding 0.85 and mean absolute errors below 0.18. To quantitatively evaluate the trade-offs between gaseous emissions and maturation, a composite performance index (CPI) was constructed. Results showed a benefit offset in the active phase, where the benefits in maturation were outweighed by penalties in emissions, especially at aeration rates of 0.10-0.40 L min^-1 kg^-1. pH showed a minimum benefit at the median rank (pH 7.80 ± 0.71), after which increased maturation and nitrous oxide suppression outweighed ammonia emission penalties. Ammonium nitrogen showed a benefit plateau at rank 0.25 (2429.58 ± 1473.78 mg kg^-1), where nitrogen retention benefits were offset by phytotoxicity. The estimated driver effect of organic matter on CPI showed a reversal between phases, with a critical inflection at rank -0.25 (59.07 ± 11.07%) during maturation due to reduced emission risks and increased humification. Composite additives were the most effective intervention, increasing CPI by 40.1-40.9 percentile points, as revealed by phased meta-analysis. This framework offers reference ranges and strategies for stage-specific parameter control to balance environmental sustainability and resource recovery during composting.