In Training data folder:
Observation-df -- All the training features (including lake nutrient observations) Daily_Observation_df -- Daily training features (Inflow, meteorological data, ice information, themal structure, hydrodynamic features from process-based model)
SST-- surface water temperature (°C) delT -- temperature difference between surface and bottom water (°C) U -- wind speed (m/s) AirT -- air temperature (°C) Humidity -- (0-100 %) CC -- cloud cover fraction (0-1) Prec -- precipitation (mm/day) swr -- short wave radiation (w/m2) inflow -- river inflow (m3/s) outflow -- river outflow (m3/s) Ice_d -- ice duration in the previous winter (days) days from iceoff -- the number of days from previous ice-off date (days) MLD -- mixing layer depth (m) W -- Wedderburn number thermD -- themocline depth
In Lake Erken, trainning environmental factors include daily meteorological data, inflow, thermal structure, ice information, and weekly lake nutrients. To extend the machine learning models to other lakes, the daily meteorological, inflow data, and at least monthly lake nutrientsare required, and thermal structure (observed or modeled), ice information are desired.
Scenario 1: Direct prediction based on observation scenario -- predict the Chl concentrations on the date when all the training features are prepared
Scenario 2: Two-step data-driven models based on pre-generated daily nutrients, observed physical factors, and hydrodynamic features from process-based model.