fix: robust snow PoE fetch by scanning available asnow*gt*sfc vars; read each var individually to avoid pydap buffer issue; auto-generate labels

app.py

@@ -92,18 +92,8 @@ def run_forecast(lat, lon, hours=24):
     # Try to fetch snow-related probabilities akin to NBM Viewer
     try:
         t_idx, prob_map = fetch_point_probabilities(dataset_url, lat, lon, hours=hours)
-        label_map = {
-            "asnow762gtaccsfc": ">= 0.3 in",
-            "asnow254gtaccsfc": ">= 0.1 in",
-            "asnow1270gtsfc": ">= 0.5 in",
-            "asnow381gtsfc": ">= 1.5 in",
-            "asnow508gtsfc": ">= 2.0 in",
-            "asnow635gtsfc": ">= 2.5 in",
-            "asnow1016gtsfc": ">= 4.0 in",
-            "apcp254gtsfc": "P(precip > 10mm)",
-        }
         if len(prob_map) > 0:
-            snow_prob_fig = make_snow_prob_fig(t_idx, prob_map, label_map)
+            snow_prob_fig = make_snow_prob_fig(t_idx, prob_map)
     except Exception as e:
         print(f"Probability fetch/plot error: {e}")
         snow_prob_fig = None

nbm_client.py

@@ -257,39 +257,48 @@ def fetch_point_probabilities(
     ilat = _nearest_index(lat_vals, lat)
     ilon = _nearest_index(lon_vals, lon)
 
-    # Default set focused on snow thresholds commonly used in NBM viewer
-    default_vars = [
-        "asnow254gtaccsfc",  # > 0.1 in
-        "asnow381gtsfc",     # > 1.5 in (0.0381 m)
-        "asnow508gtsfc",     # > 2.0 in (0.0508 m)
-        "asnow635gtsfc",     # > 2.5 in (0.0635 m)
-        "asnow762gtaccsfc",  # > 0.3 in (0.00762 m)
-        "asnow1270gtsfc",    # > 0.5 in (0.0127 m)
-        "asnow1016gtsfc",    # > 4.0 in (0.1016 m)
-        "apcp254gtsfc",      # > 10 mm liquid (approx 0.4 in) example precip prob
-    ]
-    sel_vars = variables if variables else default_vars
-    present = [v for v in sel_vars if v in ds.variables]
-    if not present:
-        logger.warning("No requested probability variables present; returning empty")
+    # Build list of snow probability-of-exceedance vars available in this dataset
+    # Accept common thresholds; names vary between hours (sometimes include 'acc')
+    desired_threshold_ids = {127, 254, 381, 508, 635, 762, 1016}
+    candidate = [v for v in ds.variables if v.startswith("asnow") and v.endswith("sfc") and "gt" in v]
+    def _extract_id(v: str) -> Optional[int]:
+        import re as _re
+        m = _re.search(r"asnow(\d+)", v)
+        return int(m.group(1)) if m else None
+    filtered: List[str] = []
+    for v in candidate:
+        vid = _extract_id(v)
+        if vid in desired_threshold_ids:
+            filtered.append(v)
+
+    # If user provided explicit list, use intersection
+    if variables:
+        filtered = [v for v in variables if v in ds.variables]
+
+    if not filtered:
+        logger.warning("No snow probability variables found; returning empty")
         return pd.DatetimeIndex([]), {}
 
-    subset = ds[present].isel(lat=ilat, lon=ilon)
+    # Use time coordinate from any available variable
+    tvar = filtered[0]
+    subset = ds[tvar].isel(lat=ilat, lon=ilon)
     full_index = _to_datetime_index(subset["time"])  # full timeline
     step_hours = _infer_step_hours(full_index)
     n_req = int(np.ceil(max(1, float(hours)) / step_hours))
     n = min(len(full_index), n_req)
     t_index = full_index[:n]
-    subset = subset.isel(time=slice(0, n)).load()
-
     out: Dict[str, pd.Series] = {}
-    for v in present:
-        vals = subset[v].values
-        # Many probability fields are 0-1 or 0-100; try to normalize to percent
-        # Heuristic: if max <= 1.0 -> scale by 100.
-        mx = float(np.nanmax(vals)) if np.isfinite(vals).any() else 0.0
-        if mx <= 1.0:
-            vals = vals * 100.0
-        out[v] = pd.Series(np.round(vals, 1), index=t_index)
+    # Read each variable individually to avoid pydap multi-variable buffer bugs
+    for v in filtered:
+        try:
+            arr = ds[v].isel(lat=ilat, lon=ilon, time=slice(0, n)).load()
+            vals = np.asarray(arr.values)
+            # Normalize to percent if needed
+            mx = float(np.nanmax(vals)) if np.isfinite(vals).any() else 0.0
+            if mx <= 1.0:
+                vals = vals * 100.0
+            out[v] = pd.Series(np.round(vals.astype(float), 1), index=t_index)
+        except Exception as ex:
+            logger.warning(f"Skipping {v} due to read/convert error: {ex}")
 
     return t_index, out

plot_utils.py

@@ -116,8 +116,29 @@ def make_snow_prob_fig(
         "#bcbd22",
         "#17becf",
     ]
+    def _auto_label(k: str) -> str:
+        # Try to decode asnow threshold to inches
+        import re as _re
+        m = _re.search(r"asnow(\d+)", k)
+        if m:
+            val = int(m.group(1))
+            # Known special labels
+            special = {127: 0.5, 254: 0.1, 381: 1.5, 508: 2.0, 635: 2.5, 762: 0.3, 1016: 4.0}
+            if val in special:
+                inc = special[val]
+            else:
+                # Fallback: interpret as meters with 1e4 divisor -> inches
+                inc = round((val / 10000.0) / 0.0254, 2)
+            return f">= {inc:g} in"
+        # apcp threshold label
+        m2 = _re.search(r"apcp(\d+)", k)
+        if m2:
+            mm = int(m2.group(1)) / 10.0  # rough fallback
+            return f"P(precip > {mm:g} mm)"
+        return k
+
     for i, key in enumerate(order):
-        label = label_map.get(key, key) if label_map else key
+        label = label_map.get(key, _auto_label(key)) if label_map else _auto_label(key)
         fig.add_trace(
             go.Bar(
                 x=time_index,
@@ -135,4 +156,3 @@ def make_snow_prob_fig(
         legend=dict(orientation="h", yanchor="bottom", y=1.02, xanchor="right", x=1),
     )
     return fig
-