diff --git a/pvnet/models/multimodal/multimodal_base.py b/pvnet/models/multimodal/multimodal_base.py
index 0a710ebf..f377f16a 100644
--- a/pvnet/models/multimodal/multimodal_base.py
+++ b/pvnet/models/multimodal/multimodal_base.py
@@ -1,4 +1,6 @@
 """Base model class for multimodal model and unimodal teacher"""
+import copy
+
 from torchvision.transforms.functional import center_crop
 
 from pvnet.models.base_model import BaseModel
@@ -8,26 +10,29 @@ class MultimodalBaseModel(BaseModel):
     """Base model class for multimodal model and unimodal teacher"""
 
     def _adapt_batch(self, batch):
-        """Slice batches into appropriate shapes for model
+        """Slice batches into appropriate shapes for model.
 
+        Returns a new batch dictionary with adapted data, leaving the original batch unchanged.
         We make some specific assumptions about the original batch and the derived sliced batch:
         - We are only limiting the future projections. I.e. we are never shrinking the batch from
           the left hand side of the time axis, only slicing it from the right
         - We are only shrinking the spatial crop of the satellite and NWP data
 
         """
+        # Create a copy of the batch to avoid modifying the original
+        new_batch = {key: copy.deepcopy(value) for key, value in batch.items()}
 
-        if "gsp" in batch.keys():
+        if "gsp" in new_batch.keys():
             # Slice off the end of the GSP data
             gsp_len = self.forecast_len + self.history_len + 1
-            batch["gsp"] = batch["gsp"][:, :gsp_len]
-            batch["gsp_time_utc"] = batch["gsp_time_utc"][:, :gsp_len]
+            new_batch["gsp"] = new_batch["gsp"][:, :gsp_len]
+            new_batch["gsp_time_utc"] = new_batch["gsp_time_utc"][:, :gsp_len]
 
         if self.include_sat:
             # Slice off the end of the satellite data and spatially crop
             # Shape: batch_size, seq_length, channel, height, width
-            batch["satellite_actual"] = center_crop(
-                batch["satellite_actual"][:, : self.sat_sequence_len],
+            new_batch["satellite_actual"] = center_crop(
+                new_batch["satellite_actual"][:, : self.sat_sequence_len],
                 output_size=self.sat_encoder.image_size_pixels,
             )
 
@@ -35,8 +40,8 @@ def _adapt_batch(self, batch):
             # Slice off the end of the NWP data and spatially crop
             for nwp_source in self.nwp_encoders_dict:
                 # shape: batch_size, seq_len, n_chans, height, width
-                batch["nwp"][nwp_source]["nwp"] = center_crop(
-                    batch["nwp"][nwp_source]["nwp"],
+                new_batch["nwp"][nwp_source]["nwp"] = center_crop(
+                    new_batch["nwp"][nwp_source]["nwp"],
                     output_size=self.nwp_encoders_dict[nwp_source].image_size_pixels,
                 )[:, : self.nwp_encoders_dict[nwp_source].sequence_length]
 
@@ -44,8 +49,8 @@ def _adapt_batch(self, batch):
             # Slice off the end of the solar coords data
             for s in ["solar_azimuth", "solar_elevation"]:
                 key = f"{self._target_key}_{s}"
-                if key in batch.keys():
+                if key in new_batch.keys():
                     sun_len = self.forecast_len + self.history_len + 1
-                    batch[key] = batch[key][:, :sun_len]
+                    new_batch[key] = new_batch[key][:, :sun_len]
 
-        return batch
+        return new_batch