diff --git a/xrspatial/geotiff/_attrs.py b/xrspatial/geotiff/_attrs.py
index df416b5d..37f8615c 100644
--- a/xrspatial/geotiff/_attrs.py
+++ b/xrspatial/geotiff/_attrs.py
@@ -17,6 +17,8 @@
 """
 from __future__ import annotations
 
+import warnings
+
 import numpy as np
 
 from ._coords import (
@@ -206,6 +208,10 @@ def _resolve_nodata_attr(attrs: dict):
     """
     nodata = attrs.get('nodata')
     if nodata is not None:
+        try:
+            float(nodata)
+        except (TypeError, ValueError) as e:
+            raise ValueError(_nodata_attr_non_numeric_msg('nodata', nodata)) from e
         return nodata
 
     vals = attrs.get('nodatavals')
@@ -214,23 +220,38 @@ def _resolve_nodata_attr(attrs: dict):
             seq = list(vals)
         except TypeError:
             seq = [vals]
+        saw_non_numeric = False
         for v in seq:
             if v is None:
                 continue
             try:
                 fv = float(v)
             except (TypeError, ValueError):
+                saw_non_numeric = True
                 continue
             if np.isnan(fv):
                 continue
             return v
+        # A tuple where every entry is non-numeric is almost certainly a
+        # user error (typo, stringified sentinel) rather than a legitimate
+        # "no sentinel" signal. Warn so the caller sees it, but still fall
+        # through to the rest of the resolution chain rather than raising:
+        # the rest of the function's contract is "skip non-numeric entries".
+        if saw_non_numeric:
+            warnings.warn(
+                f"attrs['nodatavals']={vals!r} contained only non-numeric "
+                f"entries; no usable sentinel could be resolved from it. "
+                f"Pass ``nodata=`` explicitly or fix the attr.",
+                UserWarning,
+                stacklevel=2,
+            )
 
     fill = attrs.get('_FillValue')
     if fill is not None:
         try:
             ffv = float(fill)
-        except (TypeError, ValueError):
-            return fill  # non-numeric -- pass through verbatim
+        except (TypeError, ValueError) as e:
+            raise ValueError(_nodata_attr_non_numeric_msg('_FillValue', fill)) from e
         if np.isnan(ffv):
             return None
         return fill
@@ -238,6 +259,20 @@ def _resolve_nodata_attr(attrs: dict):
     return None
 
 
+def _nodata_attr_non_numeric_msg(attr_name: str, value) -> str:
+    """Error string shared by the ``attrs['nodata']`` and ``attrs['_FillValue']``
+    non-numeric branches in ``_resolve_nodata_attr`` (#1973)."""
+    return (
+        f"attrs[{attr_name!r}]={value!r} is not numeric "
+        f"({type(value).__name__}). The writer needs a numeric "
+        f"sentinel to compare against pixel values; passing a "
+        f"non-numeric value would otherwise crash inside "
+        f"``np.isnan`` with an opaque ufunc error. Drop the "
+        f"attr, replace it with a numeric sentinel, or pass "
+        f"``nodata=`` explicitly (issue #1973)."
+    )
+
+
 def _merge_friendly_extra_tags(extra_tags_list, attrs: dict) -> list | None:
     """Combine ``attrs['extra_tags']`` with friendly tag attrs.
 
diff --git a/xrspatial/geotiff/_validation.py b/xrspatial/geotiff/_validation.py
index 790a3ea3..3ed94cba 100644
--- a/xrspatial/geotiff/_validation.py
+++ b/xrspatial/geotiff/_validation.py
@@ -230,3 +230,36 @@ def _validate_predictor_sample_format(predictor, sample_format) -> None:
             f"pair, e.g. `gdal_translate -co PREDICTOR=2` for integers or "
             f"`-co PREDICTOR=1` to drop the predictor."
         )
+
+
+def _validate_nodata_arg(nodata) -> None:
+    """Reject non-numeric ``nodata=`` at the writer entry point (#1973).
+
+    ``None`` (no sentinel) passes through. ``bool`` is rejected with
+    ``TypeError`` so all three writer entry points (eager, GPU, VRT)
+    refuse ``nodata=True`` / ``nodata=False`` the same way the eager
+    path already does for issue #1911 -- ``float(True) == 1.0`` would
+    otherwise slip a bool past the numeric branch on the GPU/VRT paths
+    that do not have their own bool guard. Anything else is run
+    through ``float()``: success means the writer's downstream
+    ``np.isnan(nodata)`` and integer-cast paths will not blow up.
+    Failure raises ``ValueError`` with the offending repr, so users
+    see ``nodata='missing'`` flagged at the boundary instead of an
+    opaque ``ufunc 'isnan' not supported`` TypeError from inside the
+    writer.
+    """
+    if nodata is None:
+        return
+    if isinstance(nodata, (bool, np.bool_)):
+        raise TypeError(
+            f"nodata must be numeric (int or float), got {nodata!r}")
+    try:
+        float(nodata)
+    except (TypeError, ValueError) as e:
+        raise ValueError(
+            f"nodata must be numeric or None, got {nodata!r} "
+            f"(type {type(nodata).__name__}). The writer compares it "
+            f"against pixel values via ``np.isnan`` and casts it to "
+            f"the array dtype; a non-numeric value would otherwise "
+            f"crash inside NumPy with a ufunc TypeError."
+        ) from e
diff --git a/xrspatial/geotiff/_writers/eager.py b/xrspatial/geotiff/_writers/eager.py
index 1915fb84..03fcada0 100644
--- a/xrspatial/geotiff/_writers/eager.py
+++ b/xrspatial/geotiff/_writers/eager.py
@@ -42,6 +42,7 @@
 )
 from .._validation import (
     _validate_3d_writer_dims,
+    _validate_nodata_arg,
     _validate_tile_size_arg,
 )
 from .._writer import write
@@ -267,6 +268,8 @@ def to_geotiff(data: xr.DataArray | np.ndarray,
     if tiled:
         _validate_tile_size_arg(tile_size)
 
+    _validate_nodata_arg(nodata)
+
     # Up-front validation: catch bad compression names before they reach
     # any of the deeper write paths (streaming, GPU, VRT, COG) where the
     # error surfaces from _compression_tag with a less obvious traceback.
@@ -775,6 +778,7 @@ def _write_vrt_tiled(data, vrt_path, *, crs=None, nodata=None,
     This enables streaming dask arrays to disk without materializing the
     full array in RAM.
     """
+    _validate_nodata_arg(nodata)
     # Validate compression_level against codec-specific range
     if compression_level is not None:
         level_range = _LEVEL_RANGES.get(compression.lower())
diff --git a/xrspatial/geotiff/_writers/gpu.py b/xrspatial/geotiff/_writers/gpu.py
index c08ab791..a5de64e0 100644
--- a/xrspatial/geotiff/_writers/gpu.py
+++ b/xrspatial/geotiff/_writers/gpu.py
@@ -27,6 +27,7 @@
 from .._runtime import GeoTIFFFallbackWarning
 from .._validation import (
     _validate_3d_writer_dims,
+    _validate_nodata_arg,
     _validate_tile_size_arg,
 )
 
@@ -259,6 +260,7 @@ def write_geotiff_gpu(data: xr.DataArray | cupy.ndarray | np.ndarray,
     # write_geotiff_gpu is always tiled, so validate tile_size here and
     # keep parity with the public to_geotiff entry point.
     _validate_tile_size_arg(tile_size)
+    _validate_nodata_arg(nodata)
     if max_z_error < 0:
         raise ValueError(
             f"max_z_error must be >= 0, got {max_z_error}")
diff --git a/xrspatial/geotiff/tests/test_nodata_validation_1973.py b/xrspatial/geotiff/tests/test_nodata_validation_1973.py
new file mode 100644
index 00000000..154604e0
--- /dev/null
+++ b/xrspatial/geotiff/tests/test_nodata_validation_1973.py
@@ -0,0 +1,183 @@
+"""Refuse non-numeric ``nodata=`` / ``attrs['_FillValue']`` (#1973).
+
+The writer compares the resolved nodata against pixel values via
+``np.isnan`` and casts it to the array dtype. A non-numeric value
+used to fall through ``_resolve_nodata_attr`` (returned verbatim) or
+the ``nodata=`` kwarg path and then crash inside NumPy with
+``ufunc 'isnan' not supported``. Both the entry point and the attr
+resolution path now refuse non-numeric values up front with a clear
+error.
+"""
+from __future__ import annotations
+
+import importlib.util
+import io
+
+import numpy as np
+import pytest
+import xarray as xr
+
+from xrspatial.geotiff import to_geotiff
+from xrspatial.geotiff._attrs import _resolve_nodata_attr
+from xrspatial.geotiff._validation import _validate_nodata_arg
+
+
+def _gpu_available() -> bool:
+    if importlib.util.find_spec("cupy") is None:
+        return False
+    try:
+        import cupy
+        return bool(cupy.cuda.is_available())
+    except Exception:
+        return False
+
+
+_HAS_GPU = _gpu_available()
+_gpu_only = pytest.mark.skipif(not _HAS_GPU, reason="cupy + CUDA required")
+
+
+def _nan_square():
+    return xr.DataArray(
+        np.full((4, 4), np.nan, dtype=np.float32),
+        coords={'y': np.arange(4.0), 'x': np.arange(4.0)},
+        dims=('y', 'x'),
+    )
+
+
+@pytest.mark.parametrize("bad", ['missing', object(), [1, 2]])
+def test_validate_nodata_arg_rejects_non_numeric(bad):
+    with pytest.raises(ValueError, match="nodata must be numeric"):
+        _validate_nodata_arg(bad)
+
+
+@pytest.mark.parametrize("ok", [None, 0, -9999, 1.5, np.float32(-1), np.int64(0)])
+def test_validate_nodata_arg_accepts_numeric_and_none(ok):
+    _validate_nodata_arg(ok)
+
+
+def test_resolve_nodata_attr_rejects_non_numeric_fillvalue():
+    with pytest.raises(ValueError, match="_FillValue"):
+        _resolve_nodata_attr({'_FillValue': 'missing'})
+
+
+def test_resolve_nodata_attr_rejects_non_numeric_nodata_attr():
+    with pytest.raises(ValueError, match=r"attrs\['nodata'\]"):
+        _resolve_nodata_attr({'nodata': 'missing'})
+
+
+def test_resolve_nodata_attr_skips_non_numeric_in_nodatavals():
+    # nodatavals (rioxarray's per-band tuple) keeps its skip-on-non-numeric
+    # behaviour: those values often come from arbitrary upstream pipelines
+    # and a single bad entry should not block writing.
+    assert _resolve_nodata_attr({'nodatavals': ('NaN ', -9999.0)}) == -9999.0
+
+
+def test_resolve_nodata_attr_still_accepts_numeric_fillvalue():
+    assert _resolve_nodata_attr({'_FillValue': -9999}) == -9999
+
+
+def test_resolve_nodata_attr_returns_none_for_nan_fillvalue():
+    assert _resolve_nodata_attr({'_FillValue': float('nan')}) is None
+
+
+def test_to_geotiff_rejects_non_numeric_nodata_kwarg():
+    buf = io.BytesIO()
+    with pytest.raises(ValueError, match="nodata must be numeric"):
+        to_geotiff(_nan_square(), buf, nodata='missing')
+
+
+def test_to_geotiff_rejects_non_numeric_fillvalue_attr():
+    da = _nan_square()
+    da.attrs['_FillValue'] = 'missing'
+    buf = io.BytesIO()
+    with pytest.raises(ValueError, match="_FillValue"):
+        to_geotiff(da, buf)
+
+
+def test_to_geotiff_vrt_path_rejects_non_numeric_nodata(tmp_path):
+    vrt_path = str(tmp_path / "tmp_1973_vrt.vrt")
+    with pytest.raises(ValueError, match="nodata must be numeric"):
+        to_geotiff(_nan_square(), vrt_path, nodata='missing')
+
+
+def test_to_geotiff_accepts_numeric_nodata_kwarg():
+    buf = io.BytesIO()
+    to_geotiff(_nan_square(), buf, nodata=-9999)
+    assert buf.getbuffer().nbytes > 0
+
+
+# ---------------------------------------------------------------------------
+# Bool rejection: ``nodata=True`` / ``nodata=False`` must raise TypeError at
+# all three writer entry points (eager, GPU, VRT). The eager path already
+# rejected bools for #1911 but the GPU/VRT validators previously routed bool
+# through ``float(True) == 1.0`` and silently coerced. The shared validator
+# now refuses bools so all three paths behave the same.
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.parametrize("bad", [True, False])
+def test_validate_nodata_arg_rejects_bool(bad):
+    with pytest.raises(TypeError, match="nodata must be numeric"):
+        _validate_nodata_arg(bad)
+
+
+def test_validate_nodata_arg_rejects_numpy_bool():
+    with pytest.raises(TypeError, match="nodata must be numeric"):
+        _validate_nodata_arg(np.bool_(True))
+
+
+def test_to_geotiff_eager_rejects_bool_nodata():
+    buf = io.BytesIO()
+    with pytest.raises(TypeError, match="nodata must be numeric"):
+        to_geotiff(_nan_square(), buf, nodata=True)
+
+
+def test_to_geotiff_vrt_rejects_bool_nodata(tmp_path):
+    vrt_path = str(tmp_path / "tmp_1973_bool_vrt.vrt")
+    with pytest.raises(TypeError, match="nodata must be numeric"):
+        to_geotiff(_nan_square(), vrt_path, nodata=True)
+
+
+@_gpu_only
+def test_write_geotiff_gpu_rejects_bool_nodata(tmp_path):
+    import cupy
+
+    from xrspatial.geotiff import write_geotiff_gpu
+
+    da_cpu = _nan_square()
+    da_gpu = da_cpu.copy(data=cupy.asarray(da_cpu.values))
+    out = str(tmp_path / "tmp_1973_bool_gpu.tif")
+    with pytest.raises(TypeError, match="nodata must be numeric"):
+        write_geotiff_gpu(da_gpu, out, nodata=True)
+
+
+# ---------------------------------------------------------------------------
+# All-non-numeric ``attrs['nodatavals']``: warn but still return None and
+# fall through. A tuple where every entry is non-numeric is almost certainly
+# a user error rather than a legitimate "no sentinel" signal.
+# ---------------------------------------------------------------------------
+
+
+def test_resolve_nodata_attr_warns_when_nodatavals_all_non_numeric():
+    with pytest.warns(UserWarning, match="nodatavals"):
+        result = _resolve_nodata_attr({'nodatavals': ('foo', 'bar')})
+    assert result is None
+
+
+def test_resolve_nodata_attr_no_warning_when_nodatavals_has_usable_entry():
+    # First entry is non-numeric, second is a real sentinel. The loop
+    # returns -9999.0 before reaching the warn site, so no warning fires.
+    import warnings as _warnings
+    with _warnings.catch_warnings():
+        _warnings.simplefilter("error")
+        assert _resolve_nodata_attr({'nodatavals': ('foo', -9999.0)}) == -9999.0
+
+
+def test_resolve_nodata_attr_no_warning_when_nodatavals_all_nan():
+    # NaN entries are skipped (they signal "the float NaN is the sentinel",
+    # which doesn't need a GDAL_NODATA tag) but they ARE numeric, so the
+    # all-non-numeric warning must not fire for an all-NaN tuple.
+    import warnings as _warnings
+    with _warnings.catch_warnings():
+        _warnings.simplefilter("error")
+        assert _resolve_nodata_attr({'nodatavals': (float('nan'),)}) is None