We critically examine the ability of future neutrino telescopes, including Baikal-GVD, KM3NeT, P-ONE, TAMBO, and IceCube-Gen2, to determine the flavor composition of high-energy astrophysical neutrinos, i.e., the relative number of νe, νµ, and ντ , in light of improving measurements of the neutrino mixing parameters. Starting in 2020, we show how measurements by JUNO, DUNE, and Hyper-Kamiokande will affect our ability to determine the regions of flavor composition at Earth that are allowed by neutrino oscillations under different assumptions of the flavor composition that is emitted by the astrophysical sources. From 2020 to 2040, the error on inferring the flavor composition at the source will improve from > 40% to less than 6%. By 2040, under the assumption that pion decay is the principal production mechanism of high-energy astrophysical neutrinos, a sub-dominant mechanism could be constrained to contribute less than 20% of the flux at 99.7% credibility. These conclusions are robust in the nonstandard scenario where neutrino mixing is non-unitary, a scenario that is the target of next-generation experiments, in particular the IceCube-Upgrade. Finally, to illustrate the improvement in using flavor composition to test beyond-the-Standard-Model physics, we examine the possibility of neutrino decay and find that, by 2040, combined neutrino telescope measurements will be able to limit the decay rate of the heavier neutrinos to below 1.8 × 10 −5 (m/eV) s −1 , at 95% credibility.