Merge pull request #1130 from mraspaud/feature-datatype-note

Add note about datatype in custom reader documentation

doc/source/dev_guide/custom_reader.rst

@@ -506,6 +506,25 @@ a convenience and are not required to read these formats. In many cases using
 the :func:`xarray.open_dataset` function in a custom file handler is a much
 better idea.
 
+.. note::
+   Be careful about the data types of the datasets your reader is returning.
+   It is easy to let the data be coerced into double precision floats (`np.float64`). At the
+   moment, satellite instruments are rarely measuring in a resolution greater
+   than what can be encoded in 16 bits. As such, to preserve processing power,
+   please consider carefully what data type you should scale or calibrate your
+   data to.
+
+   Single precision floats (`np.float32`) is a good compromise, as it has 23
+   significant bits (mantissa) and can thus represent 16 bit integers exactly,
+   as well as keeping the memory footprint half of a double precision float.
+
+   One commonly used method in readers is :meth:`xarray.DataArray.where` (to
+   mask invalid data) which can be coercing the data to `np.float64`. To ensure
+   for example that integer data is coerced to `np.float32` when
+   :meth:`xarray.DataArray.where` is used, you can do::
+
+     my_float_dataarray = my_int_dataarray.where(some_condition, np.float32(np.nan))
+
 One way of implementing a file handler is shown below:
 
 .. code:: python