Using physics-based machine learning to estimate unobserved quantities: A case study for landscape-scale soil and vegetation conductances to heat and water vapor