Yes, you can do it but it depends the data quality and quantity. It is essentially a time series prediction.

Yes, you can forecast univariate time series by implementing machine learning / AI-based data-driven techniques / wavelet analysis . When only rainfall data are available to forecast rainfall, forecasting depends on the quality and length of rainfall time series along with a certain number of time lags.

Article Effective Forecasting of Hourly Typhoon Rainfall Using Suppo...

Dear Abdoulaya

Yes. You can forecast anything using any previously acquired trends. When it comes to forecasting rainfall, you are really talking about the weather. Even if the same amount of rain is reported within the same period of time, say one year, its distribution is grossly erratic. The same amount may fall in one year, three months, or in three days, etc. Weather forecasters are struggling with the problems every hour of the day because no human has the situation under control. And if you cannot control a phenomenon, you cannot predict it accurately. That is the case with the weather, in this case rainfall. As mentioned above, meteorological departments have all the sophisticated gadgets you can think of, yet they cannot forecast accurately all the time because, scientifically speaking, no one can correctly predict occurrences of events over which the individual has no control.

Hello. Thanks to all, dear Liangliang Shi , Saikat Das . I'm just a computer scientist so I thought that because the rainfall depend on some others meteorological parameters, it would be necessary to use them in addition to the rain. I will read the state of art.

Thanks Akpan Jimmy Essien too. So you mean that Machine learning (AI) could be better than the work done meteorological departments when it comes to forecast?

Abdoulaye Kama

That is not what I said or mean. For machine-learning (or AI) to be effective, it must operate on trends that are well-established on principles that are more or less permanent. But you cannot meet those conditions for rainfall unless you can control rain. My answer stated very clearly that meteorological departments are well equipped with sophisticated gadgets in the effort to predict the weather (i.e. rainfall). Yet, they fail time and time again because they have no control over the weather. If, for instance, you have a dog on a lead, you can, to some extent, control the movements of the dog. But if the dog is not on a lead, it can go wherever it wants even if you keep calling it. In like manner, no human has any control over the amount of rain that can fall at anytime, anywhere. One year it is flooding, another year it is drought. Therefore, one can only study developments in real time and predict for the next few days or so. Even then, because the situation may change at anytime regardless of what the forecaster has predicted, predictions are bound to be erratic. Many computer scientists predict natural phenomena such as eclipses and can calculate the position of any planet at anytime. This is possible because planets move with unmatched precision, and they do so for billions of years. If you store such information in your computer, you are assured of perfect results every time. It is a far cry from what we know of rainfall. Your AI on this matter will no doubt encounter disappointments, however huge and precise your data might be.

Thank you Akpan Jimmy Essien for the response. It's very clear. Our idea was to combine hydrological model with machine learning for rainfall forecasting and maybe after spatio-temporal variability. Ok I'm goind to focus firstly in machine learning.