diff --git a/packetraven/model.py b/packetraven/model.py
index 7883661d..3f6677d9 100644
--- a/packetraven/model.py
+++ b/packetraven/model.py
@@ -1,4 +1,12 @@
+from dateutil.parser import parse as parse_date
+from matplotlib import pyplot, quiver
 import numpy
+import pyproj
+from pyproj import CRS, Transformer
+import xarray as xarray
+
+WGS84 = CRS.from_epsg(4326)
+WEB_MERCATOR = CRS.from_epsg(3857)
 
 # `dh/dt` based on historical flight data
 FREEFALL_DESCENT_RATE = lambda altitude: -5.8e-08 * altitude ** 2 - 6.001
@@ -8,3 +16,314 @@
 # integration of `(1/(dh/dt)) dh` based on historical flight data
 # TODO make this model better with ML
 FREEFALL_SECONDS_TO_GROUND = lambda altitude: 1695.02 * numpy.arctan(9.8311e-5 * altitude)
+from datetime import datetime, timedelta
+
+
+class VectorField:
+    """
+    Vector field of (u, v) values.
+    """
+
+    def __init__(self, time_deltas: [timedelta], crs: CRS = None):
+        """
+        Build vector field of (u, v) values.
+
+        :param time_deltas: list of time deltas
+        :param crs: native CRS of field
+        """
+
+        if crs is None:
+            crs = WGS84
+        if (
+            not isinstance(time_deltas, numpy.ndarray)
+            or time_deltas.dtype != numpy.timedelta64
+        ):
+            time_deltas = numpy.array(time_deltas, dtype='timedelta64[s]')
+        self.time_deltas = time_deltas
+        self.crs = crs
+
+        time_delta = numpy.nanmean(self.time_deltas).item()
+        if type(time_delta) is int:
+            time_delta *= 1e-9
+        elif type(time_delta) is timedelta:
+            time_delta = int(time_delta / timedelta(seconds=1))
+
+        self.delta_t = timedelta(seconds=time_delta)
+
+    def u(self, point: numpy.array, time: datetime) -> float:
+        """
+        u velocity in m/s at coordinates
+
+        :param point: coordinates in linear units
+        :param time: time
+        :return: u value at coordinate in m/s
+        """
+
+        pass
+
+    def v(self, point: numpy.array, time: datetime) -> float:
+        """
+        v velocity in m/s at coordinates
+
+        :param point: lon / lat coordinates
+        :param time: time
+        :return: v value at coordinate in m/s
+        """
+
+        pass
+
+    def velocity(self, point: numpy.array, time: datetime) -> float:
+        """
+        absolute velocity in m/s at coordinate
+
+        :param point: lon / lat coordinates
+        :param time: time
+        :return: magnitude of uv vector in m/s
+        """
+
+        return numpy.sqrt(self.u(point, time) ** 2 + self.v(point, time) ** 2)
+
+    def direction(self, point: numpy.array, time: datetime) -> float:
+        """
+        angle of uv vector
+
+        :param point: lon / lat coordinates
+        :param time: time
+        :return: radians from east of uv vector
+        """
+
+        return numpy.arctan2(self.u(point, time), self.v(point, time))
+
+    def plot(self, time: datetime, axis: pyplot.Axes = None, **kwargs) -> quiver.Quiver:
+        """
+        Plot vector field at the given time.
+
+        :param time: time at which to plot
+        :param axis: pyplot axis on which to plot
+        :return: quiver plot
+        """
+
+        pass
+
+    def __getitem__(self, position: (numpy.array, datetime)) -> numpy.array:
+        """
+        velocity vector (u, v) in m/s at coordinates
+
+        :param point: coordinates in linear units
+        :param time: time
+        :return: (u, v) vector
+        """
+
+        point, time = position
+        vector = numpy.array([self.u(point, time), self.v(point, time)])
+        vector[numpy.isnan(vector)] = 0
+        return vector
+
+    def __repr__(self) -> str:
+        return f'{self.__class__.__name__}({repr(self.delta_t)})'
+
+
+class RankineVortex(VectorField):
+    """
+    Time-invariant circular vector field moving in solid-body rotation (like a turntable).
+    """
+
+    def __init__(
+        self,
+        center: (float, float),
+        radius: float,
+        period: timedelta,
+        time_deltas: numpy.array,
+    ):
+        """
+        Construct a 2-dimensional solid rotating disk surrounded by inverse falloff of tangential velocity.
+
+        :param center: tuple of geographic coordinates
+        :param radius: radius of central solid rotation in meters
+        :param period: rotational period
+        :param time_deltas: time differences
+        """
+
+        transformer = Transformer.from_crs(WGS84, WEB_MERCATOR)
+        self.center = numpy.array(transformer.transform(*center))
+        self.radius = radius
+        self.angular_velocity = 2 * numpy.pi / (period / timedelta(seconds=1))
+
+        super().__init__(time_deltas)
+
+    def u(self, point: numpy.array, time: datetime) -> float:
+        return -self.velocity(point, time) * numpy.cos(numpy.arctan2(*(point - self.center)))
+
+    def v(self, point: numpy.array, time: datetime) -> float:
+        return self.velocity(point, time) * numpy.sin(numpy.arctan2(*(point - self.center)))
+
+    def velocity(self, point: numpy.array, time: datetime) -> float:
+        radial_distance = numpy.sqrt(numpy.sum((point - self.center) ** 2))
+
+        if radial_distance <= self.radius:
+            return self.angular_velocity * radial_distance
+        else:
+            return self.angular_velocity * self.radius ** 2 / radial_distance
+
+    def plot(self, axis: pyplot.Axes = None, **kwargs) -> quiver.Quiver:
+        if axis is None:
+            axis = pyplot.axes()
+
+        points = []
+        radii = numpy.linspace(1, self.radius * 2, 20)
+
+        for radius in radii:
+            num_points = 50
+            points.extend(
+                [
+                    (
+                        numpy.cos(2 * numpy.pi / num_points * point_index) * radius
+                        + self.center[0],
+                        numpy.sin(2 * numpy.pi / num_points * point_index) * radius
+                        + self.center[1],
+                    )
+                    for point_index in range(0, num_points + 1)
+                ]
+            )
+
+        vectors = [self[point, datetime.now()] for point in points]
+        points = list(zip(*pyproj.transform(WEB_MERCATOR, WGS84, *zip(*points))))
+
+        quiver_plot = axis.quiver(*zip(*points), *zip(*vectors), units='width', **kwargs)
+        axis.quiverkey(
+            quiver_plot, 0.9, 0.9, 1, r'$1 \frac{m}{s}$', labelpos='E', coordinates='figure'
+        )
+
+        return quiver_plot
+
+
+class VectorDataset(VectorField):
+    """
+    Vector field with time component using xarray observation.
+    """
+
+    def __init__(
+        self,
+        dataset: xarray.Dataset,
+        u_name: str = None,
+        v_name: str = None,
+        x_name: str = None,
+        y_name: str = None,
+        z_name: str = None,
+        t_name: str = None,
+        coordinate_system: CRS = None,
+    ):
+        """
+        Create new velocity field from given observation.
+
+        :param dataset: xarray observation containing velocity data (u, v)
+        :param u_name: name of u variable
+        :param v_name: name of v variable
+        :param x_name: name of x coordinate
+        :param y_name: name of y coordinate
+        :param z_name: name of z coordinate
+        :param t_name: name of time coordinate
+        :param coordinate_system: coordinate system of observation
+        """
+
+        if coordinate_system is None:
+            coordinate_system = WGS84
+
+        self.coordinate_system = coordinate_system
+
+        if u_name is None:
+            u_name = 'u'
+        if v_name is None:
+            v_name = 'v'
+        if x_name is None:
+            x_name = 'lon'
+        if y_name is None:
+            y_name = 'lat'
+        if z_name is None:
+            z_name = 'z'
+        if t_name is None:
+            t_name = 'time'
+
+        self.u_name = u_name
+        self.v_name = v_name
+        self.x_name = x_name
+        self.y_name = y_name
+        self.z_name = z_name
+        self.t_name = t_name
+
+        self.dataset = dataset
+
+        self.delta_x = numpy.mean(numpy.diff(self.dataset[self.x_name]))
+        self.delta_y = numpy.mean(numpy.diff(self.dataset[self.x_name]))
+
+        super().__init__(numpy.diff(self.dataset[self.t_name].values))
+
+    def _interpolate(
+        self, variable: str, point: numpy.array, time: datetime
+    ) -> xarray.DataArray:
+        if not isinstance(point, numpy.ndarray):
+            point = numpy.array(point)
+        if isinstance(time, str):
+            time = parse_date(time)
+        elif isinstance(time, timedelta):
+            time += self.dataset[self.t_name][0]
+
+        x_range = slice(
+            self.dataset[self.x_name]
+            .sel({self.x_name: numpy.min(point[0]) - 1}, method='bfill')
+            .values.item(),
+            self.dataset[self.x_name]
+            .sel({self.x_name: numpy.max(point[0]) + 1}, method='ffill')
+            .values.item(),
+        )
+        y_range = slice(
+            self.dataset[self.y_name]
+            .sel({self.y_name: numpy.min(point[1]) - 1,}, method='bfill')
+            .values.item(),
+            self.dataset[self.y_name]
+            .sel({self.y_name: numpy.max(point[1]) + 1,}, method='ffill')
+            .values.item(),
+        )
+        time_range = slice(
+            self.dataset['time'].sel(time=time, method='bfill').values,
+            self.dataset['time'].sel(time=time, method='ffill').values,
+        )
+
+        if time_range.start == time_range.stop:
+            time_range = time_range.start
+
+        cell = self.dataset[variable].sel(
+            {'time': time_range, self.x_name: x_range, self.y_name: y_range,}
+        )
+
+        if len(point.shape) > 1:
+            cell = cell.interp({'time': time}) if 'time' in cell.dims else cell
+            return xarray.concat(
+                [
+                    cell.interp({self.x_name: location[0], self.y_name: location[1],})
+                    for location in point.T
+                ],
+                dim='point',
+            )
+        else:
+            cell = cell.interp({self.x_name: point[0], self.y_name: point[1],})
+            return cell.interp({'time': time}) if 'time' in cell.dims else cell
+
+    def u(self, point: numpy.array, time: datetime) -> float:
+        return self._interpolate(self.u_name, point, time).values
+
+    def v(self, point: numpy.array, time: datetime) -> float:
+        return self._interpolate(self.v_name, point, time).values
+
+
+class VectorGFS(VectorDataset):
+    opendap_url = (
+        'http://nomads.ncep.noaa.gov:80/dods/gfs_0p25_1hr/gfs20210413/gfs_0p25_1hr_12z'
+    )
+
+    def __init__(self):
+        dataset = xarray.open_dataset(self.opendap_url)
+
+        super().__init__(
+            dataset=dataset, u_name='ugrdprs', v_name='vgrdprs', z_name='lev',
+        )
diff --git a/setup.py b/setup.py
index 1611c012..af081d6d 100644
--- a/setup.py
+++ b/setup.py
@@ -88,6 +88,7 @@
     install_requires=[
         'aprslib',
         'humanize',
+        'netcdf4',
         'numpy>=1.20.0',
         'pandas',
         'pyserial',
@@ -101,6 +102,7 @@
         'sshtunnel',
         'tablecrow>=1.3.6',
         'teek',
+        'xarray',
     ],
     extras_require={
         'testing': ['pytest', 'pytest-cov', 'pytest-xdist', 'pytz'],