diff --git a/lidar_pipeline/visualizations.py b/lidar_pipeline/visualizations.py
index e8863ef..6eb5bd2 100644
--- a/lidar_pipeline/visualizations.py
+++ b/lidar_pipeline/visualizations.py
@@ -391,7 +391,8 @@ def generate_mslrm(dem_file, basename, vis_dir, resolution):
             lrm_stack.append(lrm_norm.astype(np.float32))
 
         lrm_array = np.array(lrm_stack)
-        mslrm = np.sqrt(np.nanmean(lrm_array ** 2, axis=0))
+        with np.errstate(invalid='ignore'):
+            mslrm = np.sqrt(np.nanmean(lrm_array ** 2, axis=0))
         mslrm[nan_mask] = np.nan
         _save_tif(output, mslrm.astype(np.float32), transform, crs)
         logger.info(f"  ✓ MSRM terminé ({time.time()-t0:.1f}s){gpu_tag}")
@@ -658,7 +659,8 @@ def generate_wavelet(dem_file, basename, vis_dir, resolution):
             response = response / std_val
             wavelet_stack.append(response)
 
-        combined = np.sqrt(np.nanmean(np.array(wavelet_stack) ** 2, axis=0))
+        with np.errstate(invalid='ignore'):
+            combined = np.sqrt(np.nanmean(np.array(wavelet_stack) ** 2, axis=0))
         combined[nan_mask] = np.nan
 
         _save_tif(output, combined.astype(np.float32), transform, crs)
@@ -714,7 +716,8 @@ def generate_texture(dem_file, basename, vis_dir, resolution):
 
         # Entropy — compute bin-by-bin to avoid large 3D allocation
         n_bins = 16
-        img_uint8 = np.clip(img * 255, 0, 255).astype(np.uint8)
+        img_clean = np.nan_to_num(img, nan=0.0)
+        img_uint8 = np.clip(img_clean * 255, 0, 255).astype(np.uint8)
         quantized = (img_uint8 // (256 // n_bins)).astype(np.int32)
         entropy = np.zeros_like(img, dtype=np.float64)
         win_area = max(window * window, 1)