diff --git a/fme/downscaling/aggregators/main.py b/fme/downscaling/aggregators/main.py
index 1d1b4b138..65479867f 100644
--- a/fme/downscaling/aggregators/main.py
+++ b/fme/downscaling/aggregators/main.py
@@ -534,6 +534,8 @@ def __init__(
         self._name = ensure_trailing_slash(name)
         self._mean_target = Mean(batch_mean)
         self._mean_prediction = Mean(batch_mean)
+        # corresponds to 40 x 80 deg at 3km resolution, in comparison CONUS is 28x65 deg
+        self.max_img_upload_size = 3276800
 
     @torch.no_grad()
     def record_batch(self, target: TensorMapping, prediction: TensorMapping) -> None:
@@ -583,19 +585,20 @@ def get_relative_mean(target, prediction):
         metrics = {}
         spectra = {}
         for var_name in target.keys():
-            gap = torch.full(
-                (target[var_name].shape[-2], self.gap_width),
-                float(target[var_name].min()),
-                device=target[var_name].device,
-            )
-            maps[f"maps/{self._name}full-field/{var_name}"] = torch.cat(
-                (prediction[var_name], gap, target[var_name]), dim=1
-            )
-            maps[f"maps/{self._name}log10_relative_mean/{var_name}"] = relative[
-                var_name
-            ]
             error = prediction[var_name] - target[var_name]
-            maps[f"maps/{self._name}error/{var_name}"] = error
+            if target[var_name].nelement() < self.max_img_upload_size:
+                gap = torch.full(
+                    (target[var_name].shape[-2], self.gap_width),
+                    float(target[var_name].min()),
+                    device=target[var_name].device,
+                )
+                maps[f"maps/{self._name}full-field/{var_name}"] = torch.cat(
+                    (prediction[var_name], gap, target[var_name]), dim=1
+                )
+                maps[f"maps/{self._name}log10_relative_mean/{var_name}"] = relative[
+                    var_name
+                ]
+                maps[f"maps/{self._name}error/{var_name}"] = error
             metrics[f"metrics/{self._name}bias/{var_name}"] = error.mean()
 
             spectra_prefix = ensure_trailing_slash(f"power_spectrum_of_{self._name}")