Dissolved organic matter (DOM) dynamics influence aquatic ecosystem metabolism with ecological and biogeochemical effects. During microbial degradation, certain DOM molecules accumulate in the environments constituting the residual refractory carbon (C) pool that has a key role in the global carbon cycle in lakes and oceans. The present study aims to model the factors driving bacterial C-consumption, thus predicting the potential residual carbon accumulation. We developed mechanistic models to represent bacterial C-consumption, considering the contribution of DOM quality and phosphorus (P) and nitrogen (N) concentrations in the total carbon pool. Based on 59 different environments, we established DOM components and nutrient concentration for deep lakes, shallow lakes, high-altitude lakes, and wetlands from North-Andean Patagonian glacial lake district (around 41°S). We applied Bayesian methods to estimate model parameters from laboratory C-lability experiments performed in 26 environments. We tested the statistically predictive accuracy of our models with an external dataset consisting of C-lability experiments with natural lake water enriched with organic matter from different sources. We found a model that performed excellently in both fit to training data and prediction to external experiments. The selected model showed that an increase in P concentration stimulates C-consumption, and an increase in the proportion of DOM protein-like compounds reduces the amount of residual C. Based on the statistically predictive accuracy, we showed that our model is very useful to anticipate C-accumulation due to changes in the inputs to water bodies.

溶解有机质 (DOM) 动力学通过生态和生物地球化学效应影响水生生态系统代谢。在微生物降解过程中，某些 DOM 分子在构成残留难熔碳 (C) 池的环境中积累，该池在湖泊和海洋的全球碳循环中起关键作用。本研究旨在模拟驱动细菌碳消耗的因素，从而预测潜在的残留碳积累。考虑到 DOM 质量以及总碳库中磷 (P) 和氮 (N) 浓度的贡献，我们开发了机械模型来表示细菌 C 消耗。基于 59 种不同环境，我们建立了深湖、浅湖、高海拔湖泊、和来自北安第斯巴塔哥尼亚冰川湖区（南纬 41°左右）的湿地。我们应用贝叶斯方法从在 26 种环境中进行的实验室 C 不稳定性实验中估计模型参数。我们使用外部数据集测试了我们模型的统计预测准确性，该数据集由富含来自不同来源的有机物质的天然湖水的 C-lability 实验组成。我们发现一个模型在拟合训练数据和预测外部实验方面都表现出色。所选模型表明，P 浓度的增加会刺激 C 消耗，而 DOM 类蛋白质化合物比例的增加会减少残留 C 的量。基于统计预测准确性，