Integrating Phenological Priors with Deep Spatio-Temporal Features for tree species mapping

Street trees make cities healthier and more livable, but global research has been limited by scarce shared data. We created StreetTree, a public collection of over 12 million street-level images covering more than 8,300 tree species in 133 countries. Tests show that current image classification methods still struggle with real street scenes, where trees can look alike, change with seasons, and appear in varied conditions. StreetTree can support tree mapping, planning and greener city management.

Surface crevasses play an important role in glacier flow, iceberg calving, and meltwater routing, yet their depths remain poorly quantified at catchment or larger scales. We use ICESat-2 laser altimetry to estimate crevasse depths on Sermeq Kujalleq (Jakobshavn Isbræ) in 2019. Over ten thousand crevasses are mapped from along-track detection. Results show deepest crevasses occur ~20 km inland, demonstrating the potential of ICESat-2 for high-resolution crevasse depth mapping.