What does this mean...............?

It does not make any sense.

It sounds statistical to me.

To use statistics you need to have a (mathematical/statistical) model. That model might be the basis for working out tomorrow's weather, for instance. Let's make up some inputs to the model:

Today's minimum and maximum temperatures and pressures, the wind direction and average speed for the day with recent speeds being weighted more heavily than older speeds, amount of precipitation today and the day before etc. Those variables could either be parameters for our model or might form the basis for a more complex parameter based on several things. We would calculate the parameters and feed those values into our computer and set the program running. It would eventually produce a weather forecast for the following day. Strictly, that would be an estimate of tomorrow's weather and there would be error bounds saying how good the forecast was likely to be.

Suppose someone forgot to measure some wind speeds (empirical data). We would not then be able to produce an estimate of the weighted wind speed use that in our weather model. That is the situation we have here but it would have made more sense if written down in the reverse order:

"Because you have no [or some missing] empirical data, you cannot estimate [all the] parameters [so] things cannot be estimated mathematically."

Try rewording it to say:

"If your calculations rely on parameters that are derived from empirical data, it is difficult to obtain accurate results from those calculations if there is no data available for deriving the parameters."

I think that is what it means, although whether you find the rewording clearer or not might be debatable.

The meaning is that if the value of a parameter is not known, and the results of the calculation are sensitive to small changes in that parameter, then you can have little confidence in the accuracy of the results.

Wrong, you don't estimate parameters, one defines them. Even with the simplest cpu and analog PID loops, things like temperature, range, error, pressure, set point, min/max and any process variable has to be literally define in a way the processor can understand. Basically the programmer controls the outcome...but the math has to verified.

What the person probably meant is... if you don't have the initial numbers, you can't get exact results. But that doesn't mean you can't get approximate results.

Just for example... water evaporates faster when it's hot out. I don't need to know the exact temperature to know "it's hotter today than it was yesterday, so the plants need more water than they did yesterday". Knowing the exact temperature would (in theory) tell me exactly how *much* more water the plants need, but lacking that exact information doesn't negate the idea that the plants need more water on a hot day than on a cool one.

In particular, if something is hard to objectively measure--for example, the human capacity to survive in adverse conditions--then it can be essentially impossible to give *exact* numbers for the result (in this case, likelihood that humans will survive a particular set of conditions). We can crudely estimate based on comparisons--for example, "humans can demonstrably survive A, B, and C, so they can probably also survive D"--but we can't exactly calculate the probability.

It means that you need the data to do the calculations.

If you start off with the wrong assumptions, then logic will lead you astray.