The metabolic activity of water-limited ecosystems is strongly linked to the timing and magnitude of precipitation pulses that can trigger disproportionately high (i.e., hot-moments) ecosystem CO2 fluxes. We analyzed over 2-years of continuous measurements of soil CO2 efflux (Fs) under vegetation (Fsveg) and at bare soil (Fsbare) in a water-limited grassland. The continuous wavelet transform was used to: (a) describe the temporal variability of Fs; (b) test the performance of empirical models ranging in complexity; and (c) identify hot-moments of Fs. We used partial wavelet coherence (PWC) analysis to test the temporal correlation between Fs with temperature and soil moisture. The PWC analysis provided evidence that soil moisture overshadows the influence of soil temperature for Fs in this water limited ecosystem. Precipitation pulses triggered hot-moments that increased Fsveg (up to 9000%) and Fsbare (up to 17,000%) with respect to pre-pulse rates. Highly parameterized empirical models (using support vector machine (SVM) or an 8-day moving window) are good approaches for representing the daily temporal variability of Fs, but SVM is a promising approach to represent high temporal variability of Fs (i.e., hourly estimates). Our results have implications for the representation of hot-moments of ecosystem CO2 fluxes in these globally distributed ecosystems.
Keywords: arid grasslands, precipitation variability, machine learning, soil respiration, wavelet analysis, rain pulses