diff --git a/parcels/application_kernels/advection.py b/parcels/application_kernels/advection.py
index 82cb6c970..ffc5aa5ce 100644
--- a/parcels/application_kernels/advection.py
+++ b/parcels/application_kernels/advection.py
@@ -16,7 +16,7 @@
 
 def AdvectionRK4(particle, fieldset, time):  # pragma: no cover
     """Advection of particles using fourth-order Runge-Kutta integration."""
-    dt = particle.dt / np.timedelta64(1, "s")  # noqa TODO improve API for converting dt to seconds
+    dt = particle.dt
     (u1, v1) = fieldset.UV[particle]
     lon1, lat1 = (particle.lon + u1 * 0.5 * dt, particle.lat + v1 * 0.5 * dt)
     (u2, v2) = fieldset.UV[time + 0.5 * particle.dt, particle.depth, lat1, lon1, particle]
@@ -30,7 +30,7 @@ def AdvectionRK4(particle, fieldset, time):  # pragma: no cover
 
 def AdvectionRK4_3D(particle, fieldset, time):  # pragma: no cover
     """Advection of particles using fourth-order Runge-Kutta integration including vertical velocity."""
-    dt = particle.dt / np.timedelta64(1, "s")  # noqa TODO improve API for converting dt to seconds
+    dt = particle.dt
     (u1, v1, w1) = fieldset.UVW[particle]
     lon1 = particle.lon + u1 * 0.5 * dt
     lat1 = particle.lat + v1 * 0.5 * dt
@@ -53,7 +53,7 @@ def AdvectionRK4_3D_CROCO(particle, fieldset, time):  # pragma: no cover
     """Advection of particles using fourth-order Runge-Kutta integration including vertical velocity.
     This kernel assumes the vertical velocity is the 'w' field from CROCO output and works on sigma-layers.
     """
-    dt = particle.dt / np.timedelta64(1, "s")  # noqa TODO improve API for converting dt to seconds
+    dt = particle.dt
     sig_dep = particle.depth / fieldset.H[time, 0, particle.lat, particle.lon]
 
     (u1, v1, w1) = fieldset.UVW[time, particle.depth, particle.lat, particle.lon, particle]
@@ -97,7 +97,7 @@ def AdvectionRK4_3D_CROCO(particle, fieldset, time):  # pragma: no cover
 
 def AdvectionEE(particle, fieldset, time):  # pragma: no cover
     """Advection of particles using Explicit Euler (aka Euler Forward) integration."""
-    dt = particle.dt / np.timedelta64(1, "s")  # noqa TODO improve API for converting dt to seconds
+    dt = particle.dt
     (u1, v1) = fieldset.UV[particle]
     particle_dlon += u1 * dt  # noqa
     particle_dlat += v1 * dt  # noqa
@@ -113,7 +113,7 @@ def AdvectionRK45(particle, fieldset, time):  # pragma: no cover
     Time-step dt is halved if error is larger than fieldset.RK45_tol,
     and doubled if error is smaller than 1/10th of tolerance.
     """
-    dt = min(particle.next_dt / np.timedelta64(1, "s"), fieldset.RK45_max_dt)  # noqa TODO improve API for converting dt to seconds
+    dt = min(particle.next_dt, fieldset.RK45_max_dt)
     c = [1.0 / 4.0, 3.0 / 8.0, 12.0 / 13.0, 1.0, 1.0 / 2.0]
     A = [
         [1.0 / 4.0, 0.0, 0.0, 0.0, 0.0],
@@ -178,7 +178,7 @@ def AdvectionAnalytical(particle, fieldset, time):  # pragma: no cover
 
     tol = 1e-10
     I_s = 10  # number of intermediate time steps
-    dt = particle.dt / np.timedelta64(1, "s")  # TODO improve API for converting dt to seconds
+    dt = particle.dt
     direction = 1.0 if dt > 0 else -1.0
     withW = True if "W" in [f.name for f in fieldset.fields.values()] else False
     withTime = True if len(fieldset.U.grid.time) > 1 else False
diff --git a/parcels/kernel.py b/parcels/kernel.py
index 56ae8b2f9..e9de1a5e0 100644
--- a/parcels/kernel.py
+++ b/parcels/kernel.py
@@ -308,7 +308,7 @@ def execute(self, pset, endtime, dt):
         """Execute this Kernel over a ParticleSet for several timesteps."""
         pset._data["state"][:] = StatusCode.Evaluate
 
-        if abs(dt) < np.timedelta64(1000, "ns"):  # TODO still needed?
+        if abs(dt) < 1e-6:
             warnings.warn(
                 "'dt' is too small, causing numerical accuracy limit problems. Please chose a higher 'dt' and rather scale the 'time' axis of the field accordingly. (related issue #762)",
                 RuntimeWarning,
diff --git a/parcels/particleset.py b/parcels/particleset.py
index 1776efdb0..e93dd9211 100644
--- a/parcels/particleset.py
+++ b/parcels/particleset.py
@@ -9,6 +9,7 @@
 from tqdm import tqdm
 
 from parcels._core.utils.time import TimeInterval
+from parcels._core.utils.time import is_compatible as time_is_compatible
 from parcels._reprs import particleset_repr
 from parcels.application_kernels.advection import AdvectionRK4
 from parcels.basegrid import GridType
@@ -108,14 +109,7 @@ def __init__(
             depth = convert_to_flat_array(depth)
         assert lon.size == lat.size and lon.size == depth.size, "lon, lat, depth don't all have the same lenghts"
 
-        if time is None or len(time) == 0:
-            time = type(fieldset.U.data.time[0].values)(
-                "NaT", "ns"
-            )  # do not set a time yet (because sign_dt not known)
-        elif type(time[0]) in [np.datetime64, np.timedelta64]:
-            pass  # already in the right format
-        else:
-            raise TypeError("particle time must be a datetime, timedelta, or date object")
+        time = _get_release_times_from_interval_start(time, self.fieldset.time_interval)
         time = np.repeat(time, lon.size) if time.size == 1 else time
 
         assert lon.size == time.size, "time and positions (lon, lat, depth) do not have the same lengths."
@@ -142,7 +136,7 @@ def __init__(
             "lat": lat.astype(lonlatdepth_dtype),
             "depth": depth.astype(lonlatdepth_dtype),
             "time": time,
-            "dt": np.timedelta64(1, "ns") * np.ones(len(trajectory_ids)),
+            "dt": np.ones(len(trajectory_ids), dtype=np.float64),
             # "ei": (["trajectory", "ngrid"], np.zeros((len(trajectory_ids), len(fieldset.gridset)), dtype=np.int32)),
             "state": np.zeros((len(trajectory_ids)), dtype=np.int32),
             "lon_nextloop": lon.astype(lonlatdepth_dtype),
@@ -735,37 +729,42 @@ def execute(
         if output_file:
             output_file.metadata["parcels_kernels"] = self._kernel.name
 
-        if (dt is not None) and (not isinstance(dt, np.timedelta64)):
-            raise TypeError("dt must be a np.timedelta64 object")
-        if dt is None or np.isnat(dt):
+        if dt is None:
             dt = np.timedelta64(1, "s")
+        if not isinstance(dt, np.timedelta64):
+            raise TypeError("dt must be a np.timedelta64 object")
+
+        dt /= np.timedelta64(1, "s")
+
         self._data["dt"][:] = dt
         sign_dt = np.sign(dt).astype(int)
         if sign_dt not in [-1, 1]:
             raise ValueError("dt must be a positive or negative np.timedelta64 object")
 
         if self.fieldset.time_interval is None:
-            start_time = np.timedelta64(0, "s")  # For the execution loop, we need a start time as a timedelta object
+            start_time = 0  # For the execution loop, we need a start time as a timedelta object
             if runtime is None:
                 raise TypeError("The runtime must be provided when the time_interval is not defined for a fieldset.")
 
             else:
                 if isinstance(runtime, np.timedelta64):
-                    end_time = runtime
+                    end_time = runtime / np.timedelta64(1, "s")
                 else:
                     raise TypeError("The runtime must be a np.timedelta64 object")
 
         else:
-            if not np.isnat(self._data["time_nextloop"]).any():
+            if not np.isnan(self._data["time_nextloop"]).any():
                 if sign_dt > 0:
                     start_time = self._data["time_nextloop"].min()
                 else:
                     start_time = self._data["time_nextloop"].max()
             else:
                 if sign_dt > 0:
-                    start_time = self.fieldset.time_interval.left
+                    start_time = 0.0
                 else:
-                    start_time = self.fieldset.time_interval.right
+                    start_time = (
+                        self.fieldset.time_interval.right - self.fieldset.time_interval.left
+                    ) / np.timedelta64(1, "s")
 
             if runtime is None:
                 if endtime is None:
@@ -786,11 +785,16 @@ def execute(
                         end_time = max(endtime, self.fieldset.time_interval.left)
                 else:
                     raise TypeError("The endtime must be of the same type as the fieldset.time_interval start time.")
+                end_time = (end_time - self.fieldset.time_interval.left) / np.timedelta64(1, "s")
             else:
+                if isinstance(runtime, np.timedelta64):
+                    runtime = runtime / np.timedelta64(1, "s")
+                elif isinstance(runtime, (int, float)):
+                    raise TypeError("The runtime must be a np.timedelta64 object")
                 end_time = start_time + runtime * sign_dt
 
         # Set the time of the particles if it hadn't been set on initialisation
-        if np.isnat(self._data["time"]).any():
+        if np.isnan(self._data["time"]).any():
             self._data["time"][:] = start_time
             self._data["time_nextloop"][:] = start_time
 
@@ -801,7 +805,7 @@ def execute(
             logger.info(f"Output files are stored in {output_file.fname}.")
 
         if verbose_progress:
-            pbar = tqdm(total=(end_time - start_time) / np.timedelta64(1, "s"), file=sys.stdout)
+            pbar = tqdm(total=(end_time - start_time), file=sys.stdout)
 
         next_output = outputdt if output_file else None
 
@@ -824,7 +828,7 @@ def execute(
                         next_output += outputdt
 
             if verbose_progress:
-                pbar.update((next_time - time) / np.timedelta64(1, "s"))
+                pbar.update(next_time - time)
 
             time = next_time
 
@@ -832,6 +836,21 @@ def execute(
             pbar.close()
 
 
+def _get_release_times_from_interval_start(time: np.ndarray | None, time_interval: TimeInterval | None) -> np.ndarray:
+    if time is None or len(time) == 0:
+        return np.array([np.nan])  # do not set a time yet (because sign_dt not known)
+
+    if time_interval is None:
+        return (time - np.min(time)) / np.timedelta64(1, "s")
+
+    if time_is_compatible(time[0], time_interval.left):
+        raise TypeError(
+            f"Release times must be of a compatible type with the fieldset's time interval. Got {time[0]=!r}, but time interval is {time_interval=!r}."
+        )
+
+    return (time - time_interval.left) / np.timedelta64(1, "s")
+
+
 def _warn_outputdt_release_desync(outputdt: float, starttime: float, release_times: Iterable[float]):
     """Gives the user a warning if the release time isn't a multiple of outputdt."""
     if any((np.isfinite(t) and (t - starttime) % outputdt != 0) for t in release_times):
@@ -845,7 +864,7 @@ def _warn_outputdt_release_desync(outputdt: float, starttime: float, release_tim
 
 
 def _warn_particle_times_outside_fieldset_time_bounds(release_times: np.ndarray, time: TimeInterval):
-    if np.isnat(release_times).all():
+    if np.isnan(release_times).all():
         return
 
     if isinstance(time.left, np.datetime64) and isinstance(release_times[0], np.timedelta64):
diff --git a/tests/v4/test_particleset.py b/tests/v4/test_particleset.py
index 78049442d..440221e09 100644
--- a/tests/v4/test_particleset.py
+++ b/tests/v4/test_particleset.py
@@ -109,7 +109,7 @@ def test_pset_custominit_on_pclass(fieldset, pset_override):
 @pytest.mark.parametrize(
     "time, expectation",
     [
-        (np.timedelta64(0, "s"), does_not_raise()),
+        (np.timedelta64(0, "s"), pytest.raises(TypeError)),
         (np.datetime64("2000-01-02T00:00:00"), does_not_raise()),
         (0.0, pytest.raises(TypeError)),
         (timedelta(seconds=0), pytest.raises(TypeError)),
@@ -127,6 +127,22 @@ def test_pset_create_outside_time(fieldset):
         ParticleSet(fieldset, pclass=Particle, lon=[0] * len(time), lat=[0] * len(time), time=time)
 
 
+def test_pset_invalid_release_times(fieldset):
+    # define release times to be floats, and make sure that execution errors out informatively
+    ...
+
+
+def test_pset_incompatible_release_times(fieldset):
+    # define release times to be incompatible with the fieldset time interval, and make sure that execution errors out informatively
+    ...
+
+
+def test_get_release_times_from_interval_start(): ...  # test inputs and returns
+
+
+def test_get_release_times_from_interval_start_time_interval_none(): ...
+
+
 @pytest.mark.parametrize(
     "dt, expectation",
     [
@@ -148,7 +164,7 @@ def test_pset_starttime_not_multiple_dt(fieldset):
     pset = ParticleSet(fieldset, lon=[0] * len(times), lat=[0] * len(times), pclass=Particle, time=datetimes)
 
     def Addlon(particle, fieldset, time):  # pragma: no cover
-        particle_dlon += particle.dt / np.timedelta64(1, "s")  # noqa
+        particle_dlon += particle.dt  # noqa
 
     pset.execute(Addlon, dt=np.timedelta64(2, "s"), runtime=np.timedelta64(8, "s"), verbose_progress=False)
     assert np.allclose([p.lon_nextloop for p in pset], [8 - t for t in times])
diff --git a/tests/v4/test_particleset_execute.py b/tests/v4/test_particleset_execute.py
index b1994d4d4..e2a0c7fb1 100644
--- a/tests/v4/test_particleset_execute.py
+++ b/tests/v4/test_particleset_execute.py
@@ -47,11 +47,11 @@ def test_pset_stop_simulation(fieldset):
     pset = ParticleSet(fieldset, lon=0, lat=0, pclass=Particle)
 
     def Delete(particle, fieldset, time):  # pragma: no cover
-        if time >= fieldset.time_interval.left + np.timedelta64(4, "s"):
+        if time >= 4:
             return StatusCode.StopExecution
 
     pset.execute(Delete, dt=np.timedelta64(1, "s"), runtime=np.timedelta64(21, "s"))
-    assert pset[0].time == fieldset.time_interval.left + np.timedelta64(4, "s")
+    assert pset[0].time == 4
 
 
 @pytest.mark.parametrize("with_delete", [True, False])
@@ -74,36 +74,36 @@ def AddLat(particle, fieldset, time):  # pragma: no cover
 
 @pytest.mark.parametrize(
     "starttime, endtime, dt",
-    [(0, 10, 1), (0, 10, 3), (2, 16, 3), (20, 10, -1), (20, 0, -2), (5, 15, None)],
+    [(0, 10, 1), (0, 10, 3), (2, 16, 3), (20, 10, -1), (20, 0, -2), (5, 15, 1)],
 )
 def test_execution_endtime(fieldset, starttime, endtime, dt):
     starttime = fieldset.time_interval.left + np.timedelta64(starttime, "s")
-    endtime = fieldset.time_interval.left + np.timedelta64(endtime, "s")
+    endtime_date = fieldset.time_interval.left + np.timedelta64(endtime, "s")
     dt = np.timedelta64(dt, "s")
     pset = ParticleSet(fieldset, time=starttime, lon=0, lat=0)
-    pset.execute(DoNothing, endtime=endtime, dt=dt)
-    assert abs(pset.time_nextloop - endtime) < np.timedelta64(1, "ms")
+    pset.execute(DoNothing, endtime=endtime_date, dt=dt)
+    assert abs(pset.time_nextloop - endtime) < 1e-3
 
 
 @pytest.mark.parametrize(
     "starttime, runtime, dt",
-    [(0, 10, 1), (0, 10, 3), (2, 16, 3), (20, 10, -1), (20, 0, -2), (5, 15, None)],
+    [(0, 10, 1), (0, 10, 3), (2, 16, 3), (20, 10, -1), (20, 0, -2), (5, 15, 1)],
 )
 def test_execution_runtime(fieldset, starttime, runtime, dt):
-    starttime = fieldset.time_interval.left + np.timedelta64(starttime, "s")
-    runtime = np.timedelta64(runtime, "s")
+    starttime_date = fieldset.time_interval.left + np.timedelta64(starttime, "s")
+    runtime_date = np.timedelta64(runtime, "s")
     sign_dt = 1 if dt is None else np.sign(dt)
     dt = np.timedelta64(dt, "s")
-    pset = ParticleSet(fieldset, time=starttime, lon=0, lat=0)
-    pset.execute(DoNothing, runtime=runtime, dt=dt)
-    assert abs(pset.time_nextloop - starttime - runtime * sign_dt) < np.timedelta64(1, "ms")
+    pset = ParticleSet(fieldset, time=starttime_date, lon=0, lat=0)
+    pset.execute(DoNothing, runtime=runtime_date, dt=dt)
+    assert abs(pset.time_nextloop - starttime - runtime * sign_dt) < 1e-3
 
 
 def test_execution_fail_python_exception(fieldset, npart=10):
     pset = ParticleSet(fieldset, lon=np.linspace(0, 1, npart), lat=np.linspace(1, 0, npart))
 
     def PythonFail(particle, fieldset, time):  # pragma: no cover
-        if particle.time >= fieldset.time_interval.left + np.timedelta64(10, "s"):
+        if particle.time >= 10:
             raise RuntimeError("Enough is enough!")
         else:
             pass
@@ -111,8 +111,8 @@ def PythonFail(particle, fieldset, time):  # pragma: no cover
     with pytest.raises(RuntimeError):
         pset.execute(PythonFail, runtime=np.timedelta64(20, "s"), dt=np.timedelta64(2, "s"))
     assert len(pset) == npart
-    assert pset.time[0] == fieldset.time_interval.left + np.timedelta64(10, "s")
-    assert all([time == fieldset.time_interval.left + np.timedelta64(0, "s") for time in pset.time[1:]])
+    assert pset.time[0] == 10
+    assert all([time == 0 for time in pset.time[1:]])
 
 
 @pytest.mark.parametrize("verbose_progress", [True, False])