#!/usr/bin/env python
# coding: utf-8

# # HRRR Dashboard
# The High Resolution Rapid Refresh (HRRR, pronouned "her") is the highest resolution (2.5km) weather forecast for the entire Continental USA.  Here we investigate the gridded HRRR forecast data products from [Unidata's THREDDS server](http://thredds.ucar.edu) and visualizing the data using the [holoviz](holoviz.org) tools. 

# In[1]:


import xarray as xr


# In[2]:


url = 'http://thredds.ucar.edu/thredds/dodsC/grib/NCEP/HRRR/CONUS_2p5km/Best'
ds = xr.open_dataset(url)


# Find all the data variables that depend on time (and are not time `bounds`)

# In[3]:


time_vars = []
for var in ds.data_vars:
    if len(ds[var].dims) > 0:
        if 'time' in ds[var].dims[0] and not 'bounds' in var:
            time_vars.append(var)


# Read the coordinate reference system (crs) using metpy

# In[4]:


import metpy


# In[5]:


init_var = 'Temperature_height_above_ground'


# In[6]:


ds  = ds.metpy.parse_cf()


# In[7]:


crs = ds[init_var].metpy.cartopy_crs


# Import the [holoviz](https://holoviz.org) tools we need

# In[8]:


from cartopy import crs as ccrs
import hvplot.xarray
import holoviews as hv
from geoviews import tile_sources as gvts
import panel as pn


# Create widget for variable selection

# In[9]:


var_select = pn.widgets.Select(name='HRRR Variables:', options=time_vars, 
                               value=init_var)


# Create widget for basemap selection

# In[10]:


base_map_select = pn.widgets.Select(name='Basemap:', options=gvts.tile_sources, value=gvts.OSM)


# Create a color mesh plot in Lambert Conformal coordinates with `hvplot` and `cartopy`

# The `plot` function below creates the `hvplot` panel layout object.  We specify a basemap, pick the `quadmesh` plot type for the selected variable, and indicate we want to `rasterize` the plot so that we can render massive meshes in the browser. We also specify the `groupby` parameter as the list of dimensions that remains after we remove Y and X: `ds[var].dims[:-2]`, which automatically handles variables with either dimensions [T, Y, X] or [T, Z, Y, X].  We also specify which `bokeh` controls we want to be active by default:  the `wheel_zoom` and `pan` controls.
# 
# We also change the default slider to a selection widget for the `time` dimension so that specific times are easy to select.  See https://stackoverflow.com/a/54912917/2005869

# In[11]:


@pn.depends(var_select, base_map_select)
def plot(var, base_map):
    extra_dims = list(ds[var].dims[:-2])
    mesh = ds[var].hvplot.quadmesh(x='x', y='y', rasterize=True, crs=crs, title=var,
                                   attr_labels=False, project=True,
                                   groupby=extra_dims, cmap='rainbow',
                                   width=600, height=400).opts(alpha=0.7,
                                   data_aspect=None, active_tools=['wheel_zoom', 'pan'])
    return pn.panel(mesh * base_map, widgets={k: pn.widgets.Select for k in extra_dims})


# In[12]:


col = pn.Column(var_select, base_map_select, plot)


# We use `.servable()` below not only to display the panel object, but to makes the panel servable outside the notebook via:  `panel serve HRRR_Dashboard.ipynb`

# In[13]:


col.servable('HRRR Dashboard')