Earth's arid and semi-arid regions have been hypothesized to contribute significantly to interannual variability of the global terrestrial carbon sink. Arid and semi-arid regions such as the Western U.S. also show high vulnerability to climate extremes in the form of droughts, heatwaves, and large forest fires, compelling a need to quantify their climate-carbon responses. We quantify the net ecosystem exchange of (NEE) using a high-resolution regional inverse model with constraints from both in situ and space-based observations during 2015-2016. Posterior fluxes are evaluated against withheld aircraft observations across the North American observation network. Observationally constrained fluxes suggest moisture-driven seasonal compensation in NEE across the 2 years, evidenced by a decrease in NEE of 244-262 [Tg C] during the early carbon uptake period and an increase in NEE of 281-389 [Tg C] later during the carbon uptake period, relative to 2016. Atmospheric constraints on NEE, combined with remote sensing and machine learning-based upscaled gross primary productivity (GPP) products, allow for benchmarking of net and gross fluxes as estimated by a suite of terrestrial biosphere models. These benchmarks allow us to link changes in NEE and GPP constrained by atmospheric and space-based observations to changes in stomatal conductance and water use efficiency. High-precision in situ observations, such as those from the Global Greenhouse Gas Reference Network, are key constraints for bias-free carbon flux estimates.