Possible use of machine learning in hydrological time series?

Hi everyone! I want to say that i'm not so expertised in machine learning but for my curiosity i'd like to try to apply some basic concept to an hydrological series.
I have hourly meteorological data (rainfall, temperature, solar radiation, relative humidity...) + hourly discharge data for a 40 years time period. I would like to apply machine learning on this dataset, i don't know what can i do in order to find some hydrological conclusions... any suggestion? I hope that my question is clear. Thank you!!

Search rseek.org with the term time series machine learning for resources on this application. Before doing this, conduct an exploratory data analysis and thoroughly cleanse the data and review the principles of time series analysis with a text such as Hyndman. This is because it's necessary to understand how properly to deal with autocorrelation—observations of natural phenomenon such as weather or discharge on an hourly basis usually are correlated with the immediately preceding and successive observations and likely will show such correlation at longer intervals reflecting seasons.

Following that preparation, I recommend the {tsibble} package to prepare a tslm model (and perhaps other models) to provide a benchmark for your machine learning results. The tsibble package also provides facilities for aggregating the hourly data into daily, weekly, monthly, quarterly and annual bases. This will likely be necessary to prepare training and test sets that permit useful forecast horizons. Also, be aware that as forecast horizons increase, confidence bands may quickly encompass negative or other unrealistic values. Also, consider limiting training sets to around 270 observations, as a rule of thumb, although that should be tested.

Finally, be alert for changes in ground truth, such as levees, dikes, revetments, weirs and dams.