Calibrating environmental models and accurately quantifying their uncertainties are essential for enhancing the reliability of simulations of environmental processes. These advancements improve our predictive understanding of ecosystems and support climate-resilient decision-making. Traditional calibration methods often face challenges with high computational demands and difficulties in precise uncertainty quantification (UQ). To address these challenges, we developed three AI/ML-driven methods. The first, surrogate modeling, constructs a fast approximation of the costly environmental model, which is then used to streamline UQ and reduce computational expenses. The second method employs invertible neural networks to learn a bijective mapping between model inputs and outputs, enhancing the efficiency of solving both forward and inverse UQ tasks. The third method, diffusion-based uncertainty quantification, leverages generative AI to enable amortized Bayesian inference, quickly generating parameter posterior samples from given observations to quantify uncertainties. We applied these methods to the Energy Exascale Earth System Model land model, demonstrating their effectiveness in improving computational efficiency and accuracy in UQ.