Weather-driven disturbances, anthropogenic pressures, and biogenic stressors increasingly impact managed and natural ecosystems, with consequences for food security, carbon cycling, and ecosystem integrity. These disturbances of regular ecosystem functioning occur in many forms and alter ecosystem services such as net CO<sub>2</sub> uptake and physiological processes including photosynthesis and respiration. Local effects of climate change and global long-term warming motivate the need for enhanced monitoring of ecosystem stress and integrity in near-real-time. However, detecting early signs of ecosystem disturbance by tracking the response of flux magnitudes remains a major challenge due to the diversity of both disturbance and ecosystem types. To overcome this challenge and develop a more general approach for diagnosing ecosystem integrity, we focus on a feature common to all ecosystems: their response to light availability and temperature, reflected in the dominance of circadian (24 h) rhythms. To this end, we examine the response of the 24 h rhythm in net ecosystem CO<sub>2</sub> exchange (NEE). Using real-time continuous wavelet transform (CWT) analysis, we quantify the dominance of circadian rhythms in NEE across both cultivated and natural ecosystems. Specifically, we evaluate (i) whether NEE becomes ‘more circadian’ during the growing season; (ii) the directionality of the relationship between the 24 h rhythm and magnitude of NEE; and, finally, (iii) whether recent history of circadian ryhtms of NEE can be leveraged to improve estimates of NEE in a short-term forecasting setting. We evaluate our method across 45 site-years of two distinct plant functional types whose functioning is well understood: European croplands and pulse-driven ecosystems from the dry tropics. Results consistently show that circadian rhythms are more sensitive than flux magnitudes, reacting earlier to disturbances. Therefore, we show how tracking the response of circadian rhythms provides an actionable lever to enhance short-term forecasts of NEE. This framework may provide a novel avenue for the early diagnosis of ecosystem stress and integrity. Because it relies solely on routinely collected 30 min flux measurements, it can be readily implemented within existing real-time carbon monitoring networks such as ICOS, NEON, AmeriFlux, MexFlux, ChinaFlux, JapanFlux, KOF, SAEON and TERN.