Quantifying temperature dependence of Fe reduction in humid tropical soils: a Bayesian model-data integration

This resource includes the Python scripts of the modeling framework and figures for the manuscript entitled "Quantifying temperature dependence of Fe reduction in humid tropical soils: a Bayesian model-data integration." The Jupyter notebook files contain the code for Bayesian calibration and evaluation of the one-pool, two-pool and three-pool soil Fe-redox models (based on whether Fe(II), Fe(III), and DOC variables are included) and generating the figures found in the manuscript. To find the updated version of the codes, visit the Github link: https://github.com/ShibliSadik/T-dependence-of-Fe-redox The abstract of the manuscript: Humid tropical forests are critical regulators of the global carbon (C) cycle, yet their soil C stocks are increasingly vulnerable to warming. Predicting potential losses requires a mechanistic understanding of the processes that govern soil C stabilization and mineralization, particularly in Fe-rich soils, where iron (Fe) redox cycling plays a dual role in both protecting and decomposing organic matter. However, the temperature dependency of these Fe-mediated processes remains poorly understood. In this study, we quantified the temperature dependence of Fe(III) reduction by conducting anoxic incubations at 23, 27, and 33 C and calibrating four kinetic models of increasing complexity to estimate the Q10 and Arrhenius (Ea) using a Markov Chain Monte Carlo (MCMC) framework. Model performance was evaluated using Bayesian information criteria (WAIC, LOO, and LPML) to assess fit, complexity and uncertainty. Short-term warming significantly accelerated Fe-reduction rates, potentially destabilizing mineral-associated organic carbon and enhancing microbial respiration. Estimated Q10 and Ea values ranged from 1.5 to 2.1 and 30.8 to 56.5 kJ/mol respectively, comparable to the temperature sensitivity values measured in temperate and tropical biomes. With the available data, Bayesian information criteria preferred the simplest one pool Fe(II) model due to its parsimony. In contrast, the most complex (three pool) model, which includes dissolved organic carbon (DOC) dynamics alongside Fe reduction and oxidation, was generally the least preferred by Bayesian information criteria due to increased uncertainty from unconstrained additional processes. These results underscore the importance of temperature-dependent Fe redox processes in regulating soil C cycle in humid tropical soils and emphasize the need to balance model complexity with data availability when modeling coupled C-Fe interactions.