Modelling Rainfall-Induced Respiration Pulses in East African Savannas

Rainfall drives carbon-water dynamics in tropical drylands, yet the mechanisms underlying ecosystem respiration responses to rainfall pulses remain poorly quantified. This study investigates the ecophysiological drivers of ”Birch”-pulses in an East African semi-arid savanna using the eddy-covariance technique between 2018–2024. High-frequency fluxes were segmented into rainfall-pulse sequences and analysed through empirical and data-driven modelling frameworks, including a moisture corrected method (Rmoisture), an event segmented Peak–Tail (RPT) model, and an explainable machine-learning model (XGBoost). The RPT formulation reproduced respiration dynamics with higher accuracy (R² = 0.70; RMSE = 1.56 µmol m-2 s-1) than temperature-only and Rmoisture, successfully capturing short-term flux peaks and decay patterns. SHapley Additive exPlanations (SHAP) analysis indicated that the rain pulse response was primarily structured by antecedent drying history and the timing and intensity of rewetting, whereas temperature acted only as a secondary modifier. We show that respiration pulses arise primarily from complex abrupt dry–wet transitions rather than rainfall magnitude or mean soil moisture. Collectively, the findings advance our understanding of how antecedent hydrology, rewetting intensity, and thermal context interact to shape respiration in drylands. By integrating empirical and machine-learning approaches, this work provides a mechanistic framework for scaling and applying pulse-focused respiration partitioning to dryland sites across the region and beyond. Moreover, this approach will complement existing flux partitioning tools to work reliably in dryland ecosystems.