Merge branch 'main' into visualize-climatology

book/_toc.yml

@@ -27,5 +27,4 @@ parts:
         title: ConvLSTM
         sections:
         - file: notebooks/CHL_prediction_ConvLSTM_.ipynb
-          title: - Fit ConvLSTM
-
+          title: - Fit ConvLSTM

book/notebooks/convlstm.md

@@ -0,0 +1,20 @@
+# ConvLSTM
+
+Our data are a time series of images and it would be good to use that time information and prior days' observations of Chlorophyll-a to help use make predictions. The idea is similar to the simple CNN except that we add the prior days. Mathematically it is more complicated since we do not include the prior days as independent variables; we take into account that the information in consecutive days is correlated.
+
+This type of model is what we will want to use to make predictions when we have the day from prior days available. We want to take that data into account since today is likely to be similar to yesterday. 
+
+![](images/convlstm.jpg)
+
+## Prepping the data
+
+
+## Modeling fitting, validation and testing
+
+The model steps are the same as for the Simple CNN example.
+
+Once we have the data in the right form, we pass our model fitting function the training data sets `X_train` (predictors) and `y_train` (response or what we are trying to learn). During training, we learn the parameters.
+
+During the validation step, we will run a loop to improve our hyperparameters (structure of our model) using the `X_val` and `y_val` that were not used in training. The result is our 'best' model.
+
+Finally, during the test step, we use our best model to make predictions for the days that it has never 'seen' (the test data). It will use the predictors for these test days (`X_test`) to make predictions and then we will compare the predictions to the true values (`y_test`).

book/notebooks/simple_cnn.md

@@ -0,0 +1,24 @@
+# Simple CNN
+
+In this example, we will attempt to use a Convolutional Neural Network trained on sea surface temperature (sst) and salinity to predict Chlorophyl-a. This is a toy example and won't work very well since sst and salinity do not predict chlorphyll-a all that well. But it will show you the steps involved: prep data into xarray objects used for training, validation and testing.
+
+![](images/simple_cnn.jpg)
+
+## Prepping the data
+
+There are many data formats you could use, but we are going to use xarrays (a type of data cube). Technically, it is a numpy array with some meta data. Our arrays with have a bounding box (lat/lon), a time dimension, and variables.
+
+We want to create an xarray for the training, validation, and test data:
+
+- `X_train, X_val, X_test`: the predictor variables of the train/validation/test data
+- `y_train, y_val, y_test`: the response variables of the train/validation/test data
+
+All the xarrays will have the same lat/lon grid. The `_train` sets will have the same days, the `_val` will have the same days, and the `_test` will have the same days.
+
+## Modeling fitting, validation and testing
+
+Once we have the data in the right form, we pass our model fitting function the training data sets `X_train` (predictors) and `y_train` (response or what we are trying to learn). During training, we learn the parameters.
+
+During the validation step, we will run a loop to improve our hyperparameters (structure of our model) using the `X_val` and `y_val` that were not used in training. The result is our 'best' model.
+
+Finally, during the test step, we use our best model to make predictions for the days that it has never 'seen' (the test data). It will use the predictors for these test days (`X_test`) to make predictions and then we will compare the predictions to the true values (`y_test`).