From 43c45f99c1bac668315fd804bfe3ada4a0b4d289 Mon Sep 17 00:00:00 2001
From: Brendan Collins <brendancol@gmail.com>
Date: Mon, 30 Mar 2026 17:17:45 -0700
Subject: [PATCH] Propagate NaN from curve_number in curve_number_runoff
 (#1104)

The NaN fixup only checked np.isnan(p), missing NaN in the curve
number raster. When CN is NaN, P > NaN is False, so np.where wrote
0.0 instead of NaN. Now checks both p and cn in all three backends.
---
 xrspatial/flood.py            |  8 ++++----
 xrspatial/tests/test_flood.py | 26 ++++++++++++++++++++++++++
 2 files changed, 30 insertions(+), 4 deletions(-)

diff --git a/xrspatial/flood.py b/xrspatial/flood.py
index efc983ab..e0adb94a 100644
--- a/xrspatial/flood.py
+++ b/xrspatial/flood.py
@@ -328,8 +328,8 @@ def _cn_runoff_numpy(p, cn):
     s = (25400.0 / cn) - 254.0
     ia = 0.2 * s
     q = np.where(p > ia, (p - ia) ** 2 / (p + 0.8 * s), 0.0)
-    # propagate NaN from rainfall
-    q = np.where(np.isnan(p), np.nan, q)
+    # propagate NaN from rainfall or curve number
+    q = np.where(np.isnan(p) | np.isnan(cn), np.nan, q)
     return q
 
 
@@ -340,7 +340,7 @@ def _cn_runoff_cupy(p, cn):
     s = (25400.0 / cn) - 254.0
     ia = 0.2 * s
     q = cp.where(p > ia, (p - ia) ** 2 / (p + 0.8 * s), 0.0)
-    q = cp.where(cp.isnan(p), cp.nan, q)
+    q = cp.where(cp.isnan(p) | cp.isnan(cn), cp.nan, q)
     return q
 
 
@@ -349,7 +349,7 @@ def _cn_runoff_dask(p, cn):
     s = (25400.0 / cn) - 254.0
     ia = 0.2 * s
     q = _da.where(p > ia, (p - ia) ** 2 / (p + 0.8 * s), 0.0)
-    q = _da.where(_da.isnan(p), np.nan, q)
+    q = _da.where(_da.isnan(p) | _da.isnan(cn), np.nan, q)
     return q
 
 
diff --git a/xrspatial/tests/test_flood.py b/xrspatial/tests/test_flood.py
index e3c7304e..b30f4e91 100644
--- a/xrspatial/tests/test_flood.py
+++ b/xrspatial/tests/test_flood.py
@@ -362,6 +362,32 @@ def test_numpy_equals_dask_cupy(self):
                               expected_results=result_np.data)
 
 
+def test_cn_runoff_nan_curve_number_1104():
+    """NaN in curve_number should produce NaN output, not 0.
+
+    Regression test for #1104: P > NaN is always False, so np.where
+    took the else-branch and wrote 0.0 instead of NaN.
+    """
+    rainfall = xr.DataArray(
+        np.array([[100.0, 100.0, 100.0]], dtype=np.float64)
+    )
+    cn_data = np.array([[80.0, np.nan, 90.0]], dtype=np.float64)
+    cn_raster = xr.DataArray(cn_data)
+
+    result = curve_number_runoff(rainfall, curve_number=cn_raster)
+    data = result.data
+    if hasattr(data, 'compute'):
+        data = data.compute()
+    data = np.asarray(data)
+
+    # Cell 0 (CN=80): valid runoff
+    assert np.isfinite(data[0, 0]) and data[0, 0] > 0
+    # Cell 1 (CN=NaN): must be NaN, not 0
+    assert np.isnan(data[0, 1]), f"expected NaN, got {data[0, 1]}"
+    # Cell 2 (CN=90): valid runoff
+    assert np.isfinite(data[0, 2]) and data[0, 2] > 0
+
+
 # ===================================================================
 # travel_time
 # ===================================================================