Enable observed impacts to be overwritten with NaN

This is useful where we have a structure or rock wall and want to specify that we don't know what the observed storm regime is. In this case, we'll put a 'unknown' string in the ./data/raw/profile_features observed storm impact csv field and overwrite it with a NaN in our pandas dataframe.

src/analysis/observed_storm_impacts.py

@@ -218,9 +218,21 @@ def overwrite_impacts(df_observed_impacts, df_raw_features):
     :param df_raw_profile_features:
     :return:
     """
-    df_observed_impacts.update(
-        df_raw_features.rename(columns={"observed_storm_regime": "storm_regime"})
-    )
+
+    # Get manually specified impacts from the profile features ./data/raw/ folder. Note that sites which need to be
+    # overwritten with a NaN, use the string 'none' in the csv. This is because when we use the df.update() command,
+    # it doesn't overwrite NaN values. So we'll put in the 'none' string, then overwrite that with the NaN.
+
+    df_overwritten_impacts = df_raw_features.rename(
+        columns={"observed_storm_regime": "storm_regime"}
+    ).storm_regime.to_frame()
+
+    df_observed_impacts.update(df_overwritten_impacts)
+
+    # Replace 'none' with nan
+    df_overwritten_impacts.loc[
+        df_overwritten_impacts.storm_regime == "unknown", "storm_regime"
+    ] = np.nan
     return df_observed_impacts