I think one of the features/requirements of Markov chains is that they be stateless. Previous states cannot influence the current probability. It really depend

I think one of the features/requirements of Markov chains is that they be stateless. Previous states cannot influence the current probability. It really depends on how granular you want to cut it. You could make 10 rainy states, each with an increased probability of sunny, equal chance to stay or go to a lower probability, and no chance of going to a state with a higher probability. So R90 could be S = 10%, R90 = 45%, and R80 = 45%. R80 could be S = 20%, R80 = 40%, and R70% = 40%. And so on. That way you don't need to track how many days it's been raining, it will naturally filter down to lower rainy percentages until it inevitably goes back to sunny.