That the weather of two weeks from now cannot, in any practical sense, be predicted is not a failure of meteorology but a consequence of mathematics. The equations that govern the atmosphere are sensitive to initial conditions in a way that compounds small errors into large ones at a rate set by the system itself; a fractional uncertainty in today's temperature field is, in some number of days, indistinguishable from total ignorance. The deeper insight, established in the 1960s through the work of Lorenz on a deliberately simplified model, was not that prediction is hard but that it is intrinsically bounded — there exists, for systems of this kind, a horizon beyond which prediction is impossible regardless of how good the observations or the computers become.
The result is sometimes misunderstood. Chaos in this technical sense does not mean randomness; the equations are entirely deterministic, with no chance term, and the same initial state, exactly specified, would always evolve in exactly the same way. The unpredictability arises from the fact that initial states cannot be exactly specified — every measurement has a finite precision — and that the system magnifies the unspecified part exponentially. The behaviour is, in the strict sense, deterministic chaos, a phrase whose apparent contradiction dissolves once one notices that determinism and predictability are different properties of the same system.
The misunderstanding matters because it shapes what one expects from improved instruments. A naive reader of meteorology might suppose that better satellites, denser observing networks, and faster computers will, eventually, push the prediction horizon to a month, a year, a decade. The mathematics suggests otherwise. Doubling the observational precision pushes the horizon by a fixed amount whose size depends on the system's growth rate, and for the atmosphere this amount is modest; reaching one extra week of useful prediction requires roughly an order-of-magnitude improvement in precision. The returns to investment in observation are not zero but, after a point, deeply diminishing.
What the picture leaves room for is the prediction of statistics rather than states. One cannot say what the temperature will be in Delhi on a particular day three months hence; one can, given enough understanding of the climate, say something useful about the likely distribution of temperatures over that month. The horizon for statistics is much further out than the horizon for individual states, and a great deal of useful work on weather and climate proceeds in this statistical register. To collapse the two — to treat statistical prediction as a weaker form of state prediction — is to mistake the kind of claim each is making.
The primary purpose of the passage is to: