AWS Forecast accuracy

Hello: I have time series data from farming sensors (with temperature, humidity, soil moisture, etc) and I would like to make predictions for each parameter. I 'm trying AWS Forecast, using the console, and using CUSTOM (as the other options seems more suitable for business operations, demand planning, etc) and the DEFAULT Predictor configuration. I actually split the real data (30 days) into the training data set (27 days) and reserved the last 3 days to compare with the predictions. Data frequency is 30 minutes and forecast horizons is 144 (3 days forecast) . I added geolocation, with TimeZone and weather index as well for better accurracy (I tried the same without these additions and accuracy was poor as well).

The results given in P10, P50 and P90 shows a huge variation of course, but so far, I don't see a good accuracy between the predicted values and the real ones... There are many values outside the P10 ~ P90 range which I would not expect... As an example, this is chart comparing forecasted values (p10, p50 and p90) vs the real values:

As we can see, the real values (yellow) are higher than predicted values...

I'm not sure if I'm doing something wrong here, but so far, these results don't seem to be very useful in a real life case... Am I missing something?

UPDATE: I trained the model using much more data, and now it looks very good:

Thanks in advance; SOS