Extended Data Fig. 3: Comparing global predictions to national maps of IRES in the USA and Australia.

Comparison of a, the US National Hydrography Dataset (NHDPlus, medium resolution) and d, the Australian hydrological geospatial fabric, with our model predictions based on two thresholds of flow intermittence, either ≥1 zero-flow day per year (b, e), or ≥1 zero-flow month (30 days) per year (c, f), on average. Only rivers and streams with MAF ≥ 0.1 m³ s⁻¹ are shown for the USA (a–c) and with drainage area ≥10 km² for Australia (d–f). The US reference dataset portrays 19–22% of the length of rivers and streams as non-perennial, depending on whether reaches without flow intermittence status are assumed to be perennial or removed; our estimates range from 51% (≥1 zero-flow day per year) to 36% (≥1 zero-flow month per year). We hypothesize that the remaining gap in IRES prevalence is attributable to a tendency of our model to overpredict intermittence across the eastern USA and an under-accounting of intermittence in medium to large rivers by the national dataset. The Australian reference dataset portrays 91% of the length of rivers and streams as non-perennial; our estimates range from 95% (≥1 zero-flow day per year) to 92% (≥1 zero-flow month per year). See Extended Data Fig. 7b for data sources. Mapping software: ArcMap (ESRI).