Long data helps to forecast more accurately. You should use at least 70% of your data for model building and the rest of the 30% for validation of the model.

You can try reading these:

1. https://www.sciencedirect.com/science/article/pii/B9780128206737000068

2. https://koreascience.kr/article/JAKO201719363528990.page

3.

For better forecasting, we must have as long as long data to be considered, and 30 years is the minimum we would consider from a climate change point of view.

But I can say there is no thumb rule.

I would caution you to make sure you have appropriate metadata about the dataset as you decide what to parts to use. Did methods change during the 50 years? Are those changes well documented? Quality and quantity are important. Otherwise, predictions based upon faulty or questionable data are sometimes worse than no predictions at all. Good luck!

Variables that have more variability should be longer in length so that the prediction results have the lowest amount of error. But as a standard, the length of the time series of your data should be more than 30 years.

Good luck

على الاقل 30 سنة.

Use for a period of 50 years with the adoption of the highest rate of annual precipitation.