Fig. 2: Trends, changes and attribution of global streamflow based on observation-constrained models in 1981-2020.

a Spatial pattern of 40-year streamflow annual trends obtained from four observation-constrained models. SD, D, I, SI denote significant decrease, insignificant decrease, insignificant increase, and significant increase at the significant level of α = 0.05 (Mann–Kendall test), respectively. Purple dots in global map indicate that the trend is significant. b 40-year global area-weighted anomaly streamflow time series and trends for the four observation-constrained models (mean ± 1std), and p_min represents the minimum p value from the four models. c Attribution results for global area-weighted streamflow changes for the four observation-constrained models, with 90% upper and lower bounds and median values for the scale factor of the regularized optimal fingerprinting method being the median of the corresponding values for the four models. If the range of the scale factor at the 90% significance level is greater than 0 and contains 1, the driving factor (climate change (CLI), elevated CO₂ (eCO₂), and land-use change (LUC)) can be attributed; otherwise, it cannot.