Estimating Ecosystem Resilience From Noisy Observational Data

The resilience of an ecosystem indicates its capacity to recover from disturbances, a quality essential for maintaining ecosystem persistence under global change. Temporal autocorrelation ( TAC $$ \mathrm{TAC} $$ ) of ecosystem states has been increasingly used to measure the change of ecosystem resilience, with increasing TAC $$ \mathrm{TAC} $$ representing a decline in resilience and approach toward potential tipping points. However, observations of ecosystem states are inevitably embedded with noise of different kinds, and the extent to which measurement noise may affect resilience assessments remains unclear. This study employs mathematical derivation, idealized experiments, and remote sensing datasets with varying noise levels to examine the effect of measurement noise on the TAC $$ \mathrm{TAC} $$ calculation. Our analyses indicate that TAC $$ \mathrm{TAC} $$ estimates from noisy datasets are systematically lower than those from noise-free datasets, with the degree of underestimation varying with noise levels, observational frequencies, and pulse-like disturbance intensities. Specifically, higher temporal resolution of observation and greater disturbance intensity enhances the accuracy of TAC $$ \mathrm{TAC} $$ estimates under constant noise levels. Additionally, we highlight that temporal changes of noise and disturbance characteristics may bias the trend of TAC $$ \mathrm{TAC} $$ , potentially resulting in spurious early warning signals of critical transitions. Employing observations with higher temporal resolution, together with appropriate data processing techniques, can partially mitigate the influence of noise and thereby enable more accurate assessments of global ecosystem resilience.