To grapple with flux processes in the earth system, the carbon cycle, nitrogen cycle, oxygen cycle, ocean heat distribution, global heat distribution, etc. You define your reservoirs, reservoir times, reservoir conditions, the products that go into the reservoir, the products that come out of them and so on. The reservoirs represent indvidual boxes; you slice the cycle up into different stations, and where needed you subdivide to smaller detail. If you turn this into a mathematical/computational model the transformation from chemical reactions/processes to variables and operators do not alter the basic structure much (though the necessary translation becomes essential).

Understanding the layout of the fluxes aids in predictive qualities - similar to the Leontief matrix as you describe. One difference I see seems that Leontief uses economic components as building blocks, and not reservoirs. Somewhat logical, as geology constantly mucks with the influence of time and kinetics.