fix noise tapering and error handling

pysteps/blending/steps.py

@@ -777,7 +777,7 @@
             self.__params.filter_kwargs = deepcopy(self.__config.filter_kwargs)
 
         if self.__config.noise_kwargs is None:
-            self.__params.noise_kwargs = dict()
+            self.__params.noise_kwargs = {"win_fun": "tukey"}
         else:
             self.__params.noise_kwargs = deepcopy(self.__config.noise_kwargs)
 
@@ -1092,16 +1092,31 @@
 
             self.__precip_models = np.stack(temp_precip_models)
 
+        if self.__params.noise_kwargs["win_fun"] is not None:
+            tapering = utils.tapering.compute_window_function(
+                self.__precip.shape[1],
+                self.__precip.shape[2],
+                self.__params.noise_kwargs["win_fun"],
+            )
+        else:
+            tapering = np.ones((self.__precip.shape[1], self.__precip.shape[2]))
+
+        tapering_mask = tapering == 0.0
+        masked_precip = self.__precip.copy()
+        masked_precip[:, tapering_mask] = np.nanmin(self.__precip)
+        masked_precip_models = self.__precip_models.copy()
+        masked_precip_models[:, :, tapering_mask] = np.nanmin(self.__precip_models)
+
         # Check for zero input fields in the radar and NWP data.
         self.__params.zero_precip_radar = blending.utils.check_norain(
-            self.__precip,
+            masked_precip,
             self.__config.precip_threshold,
             self.__config.norain_threshold,
         )
         # The norain fraction threshold used for nwp is the default value of 0.0,
         # since nwp does not suffer from clutter.
         self.__params.zero_precip_model_fields = blending.utils.check_norain(
-            self.__precip_models,
+            masked_precip_models,
             self.__config.precip_threshold,
             self.__config.norain_threshold,
         )

pysteps/tests/test_nowcasts_steps.py

@@ -7,7 +7,6 @@
 from pysteps import io, motion, nowcasts, verification
 from pysteps.tests.helpers import get_precipitation_fields
 
-
 steps_arg_names = (
     "n_ens_members",
     "n_cascade_levels",
@@ -22,7 +21,7 @@
 steps_arg_values = [
     (5, 6, 2, None, None, "spatial", 3, 1.30),
     (5, 6, 2, None, None, "spatial", [3], 1.30),
-    (5, 6, 2, "incremental", None, "spatial", 3, 7.31),
+    (5, 6, 2, "incremental", None, "spatial", 3, 7.32),
     (5, 6, 2, "sprog", None, "spatial", 3, 8.4),
     (5, 6, 2, "obs", None, "spatial", 3, 8.37),
     (5, 6, 2, None, "cdf", "spatial", 3, 0.60),

pysteps/utils/tapering.py

@@ -82,7 +82,7 @@
         Array of shape (m, n) containing the tapering weights.
     """
     X, Y = np.meshgrid(np.arange(n), np.arange(m))
-    R = np.sqrt((X - int(n / 2)) ** 2 + (Y - int(m / 2)) ** 2)
+    R = np.sqrt(((X / n) - 0.5) ** 2 + ((Y / m) - 0.5) ** 2)
 
     if func == "hann":
         return _hann(R)
@@ -108,26 +108,24 @@
 
 def _hann(R):
     W = np.ones_like(R)
-    N = min(R.shape[0], R.shape[1])
-    mask = R > int(N / 2)
+    mask = R > 0.5
 
     W[mask] = 0.0
-    W[~mask] = 0.5 * (1.0 - np.cos(2.0 * np.pi * (R[~mask] + int(N / 2)) / N))
+    W[~mask] = 0.5 * (1.0 - np.cos(2.0 * np.pi * (R[~mask] + 0.5)))
 
     return W
 
 
 def _tukey(R, alpha):
     W = np.ones_like(R)
-    N = min(R.shape[0], R.shape[1])
 
-    mask1 = R < int(N / 2)
-    mask2 = R > int(N / 2) * (1.0 - alpha)
+    mask1 = R < 0.5
+    mask2 = R > 0.5 * (1.0 - alpha)
     mask = np.logical_and(mask1, mask2)
     W[mask] = 0.5 * (
-        1.0 + np.cos(np.pi * (R[mask] / (alpha * 0.5 * N) - 1.0 / alpha + 1.0))
+        1.0 + np.cos(np.pi * (R[mask] / (alpha * 0.5) - 1.0 / alpha + 1.0))
     )
-    mask = R >= int(N / 2)
+    mask = R >= 0.5
     W[mask] = 0.0
 
     return W