Prelim results simple LSTM - one segment

USGS-R / river-dl

Deep learning model for predicting environmental variables on river systems

Creative Commons Zero v1.0 Universal

22 stars 14 forks source link

Prelim results simple LSTM - one segment #1

Closed jsadler2 closed 4 years ago

jsadler2 commented 4 years ago

I ran the simple_streamflow_lstm.py script to get these results. This is a model trained only on one segment. These are the predictions for segment 2038:

With these predictors: ['seg_tave_air', 'seg_rain', 'seg_tave_water', 'seg_upstream_inflow']

Without 'seg_upstream_inflow'

Run info: Epochs: 40 Optimizer: Adam Loss: MAE Learning rate: 0.001

jsadler2 commented 4 years ago

So looks like I still have some things to figure out about

jsadler2 commented 4 years ago

Even in my training it's performing quite poorly:

jsadler2 commented 4 years ago

It's getting seasonal cycles, but none of the events pretty much:

jsadler2 commented 4 years ago

Maybe I just don't have enough training data and I should try to use all of the segments. But if that's the case, I should probably use the segment (catchment) attributes too.