diff --git a/fink_broker/hbaseUtils.py b/fink_broker/hbaseUtils.py
index e3a20b81..3879910a 100644
--- a/fink_broker/hbaseUtils.py
+++ b/fink_broker/hbaseUtils.py
@@ -44,7 +44,7 @@ def load_fink_cols():
     --------
     >>> fink_cols, fink_nested_cols = load_fink_cols()
     >>> print(len(fink_cols))
-    17
+    19
 
     >>> print(len(fink_nested_cols))
     18
@@ -66,7 +66,9 @@ def load_fink_cols():
         'tracklet': {'type': 'string', 'default': ''},
         'vsx': {'type': 'string', 'default': 'Unknown'},
         'x3hsp': {'type': 'string', 'default': 'Unknown'},
-        'x4lac': {'type': 'string', 'default': 'Unknown'}
+        'x4lac': {'type': 'string', 'default': 'Unknown'},
+        'lc_features_g': {'type': 'string', 'default': '[]'},
+        'lc_features_r': {'type': 'string', 'default': '[]'},
     }
 
     fink_nested_cols = {}
@@ -97,7 +99,7 @@ def load_all_cols():
     >>> root_level, candidates, images, fink_cols, fink_nested_cols = load_all_cols()
     >>> out = {**root_level, **candidates, **images, **fink_cols, **fink_nested_cols}
     >>> print(len(out))
-    146
+    148
     """
     fink_cols, fink_nested_cols = load_fink_cols()
 
@@ -308,7 +310,7 @@ def load_ztf_index_cols():
     --------
     >>> out = load_ztf_index_cols()
     >>> print(len(out))
-    70
+    72
     """
     common = [
         'objectId', 'candid', 'publisher', 'rcid', 'chipsf', 'distnr',
@@ -326,7 +328,7 @@ def load_ztf_index_cols():
         'vsx',
         'snn_snia_vs_nonia', 'snn_sn_vs_all', 'rf_snia_vs_nonia',
         'classtar', 'drb', 'ndethist', 'rf_kn_vs_nonkn', 'tracklet',
-        'anomaly_score', 'x4lac', 'x3hsp'
+        'anomaly_score', 'x4lac', 'x3hsp', 'lc_features_g', 'lc_features_r'
     ]
 
     mangrove = [
@@ -801,8 +803,11 @@ def push_full_df_to_hbase(df, row_key_name, table_name, catalog_name):
     catalog_name: str
         Name for the JSON catalog (saved locally for inspection)
     """
+    # Cast feature columns
+    df_casted = cast_features(df)
+
     # Check all columns exist, fill if necessary, and cast data
-    df_flat, cols_i, cols_d, cols_b = bring_to_current_schema(df)
+    df_flat, cols_i, cols_d, cols_b = bring_to_current_schema(df_casted)
 
     # Assign each column to a specific column family
     # This is independent from the final structure
@@ -832,6 +837,42 @@ def push_full_df_to_hbase(df, row_key_name, table_name, catalog_name):
         catfolder=catalog_name
     )
 
+def cast_features(df):
+    """ Cast feature columns into string of array
+
+    Parameters
+    ----------
+    df: Spark DataFrame
+        DataFrame of alerts
+
+    Returns
+    ----------
+    df: Spark DataFrame
+
+    Examples
+    ----------
+    # Read alert from the raw database
+    >>> df = spark.read.format("parquet").load(ztf_alert_sample_scidatabase)
+
+    >>> df = cast_features(df)
+    >>> assert 'lc_features_g' in df.columns, df.columns
+
+    >>> a_row = df.select('lc_features_g').limit(1).toPandas().values[0][0]
+    >>> assert isinstance(a_row, str), a_row
+    """
+    if ('lc_features_g' in df.columns) and ('lc_features_r' in df.columns):
+        df = df.withColumn(
+            'lc_features_g',
+            F.array('lc_features_g.*').astype('string')
+        )
+
+        df = df.withColumn(
+            'lc_features_r',
+            F.array('lc_features_r.*').astype('string')
+        )
+
+    return df
+
 
 if __name__ == "__main__":
     """ Execute the test suite with SparkSession initialised """