Probability is only fixed in a controlled situation, such as tossing a fair coin, or rolling a roulette wheel. In the real world - the weather, earthquakes, asteroid strikes - probability is not fixed. When we talk about the probability of a certain area being flooded next year, we rely on historical data, stretching back possibly over thousands of years. Unfortunately in an age of global warming and changing trends, this data is not reliable. Many incidents of flooding have been described as a one in a thousand year event. You might think this means that area is flooded on average once in a thousand years. More accurately maybe, records show that that area has been flooded once in a thousand years. This same area could have been flooded three times in three years. So extreme an occurrence probably means that the probability of that area being flooded has changed, and we must change the value of an area being flooded in the calculations insurance companies make in setting insurance premiums, and the calculations governments make when allocating money for flood defences. These changes to risk are most relevant in mountainous area or on flood pains on either side of a river. The probability of an area being flooded must be constantly update and is often underestimated if it is based on historical data.