import numpy as np import pandas as pd import panel as pn import xarray as xr import param import hvplot.xarray import matplotlib.pyplot as plt import cartopy.crs as ccrs import cartopy.feature as cfeature from datatree import open_datatree from xsadcp import get_range, load_csv, load_bathymetry, load_zarr, load_file, filter_df, filter_data, quiver_depth_filtered, bathy_uship_vship_bottom_depth, corsen_data, vectors_plot, fix_time, transform_netCDF, get_info, get_file_names, open_ds class SADCP_Viewer(param.Parameterized): """ A parameterized class for viewing SADCP data. This class provides widgets for selecting data parameters, updating data based on selections, and generating plots to visualize the SADCP data. Available functions: - update_name_options: Update dropdown options and slider ranges based on selected years and file. - update_plots: Update plots based on selected data and parameters. """ # Load data and initialize widgets df = load_csv() bathy = load_bathymetry() tree=load_zarr() file_names = df["file_name"].tolist() years = sorted(df["year"].unique()) # Widgets for selecting data parameters year_slider = pn.widgets.IntRangeSlider(name="Year Range", start=df["year"].min(), end=df["year"].max()) file_dropdown = pn.widgets.Select(name="File Selector") longitude_slider = pn.widgets.RangeSlider(name="Longitude Range", start=-180, end=180, step=1) latitude_slider = pn.widgets.RangeSlider(name="Latitude Range", start=-90, end=90, step=1) depth_range_slider = pn.widgets.IntRangeSlider(start=100, end=300, value=(100, 300), step=1, name="Depth Range") depth_2_checkbox = pn.widgets.Checkbox(value=False, name="Depth 2 Checkbox") depth_3_checkbox = pn.widgets.Checkbox(value=False, name="Depth 3 Checkbox") depth_2_range_slider = pn.widgets.IntRangeSlider(start=100, end=300, value=(100, 300), step=1, name="Depth 2 Range") depth_3_range_slider = pn.widgets.IntRangeSlider(start=100, end=300, value=(100, 300), step=1, name="Depth 3 Range") num_vectors_slider = pn.widgets.IntSlider(start=40, end=800, step=1, value=100, name="Number of Vectors") scale_factor_slider = pn.widgets.FloatSlider(start=0.1, end=1, step=0.1, value=0.5, name="Scale Factor") bathy_checkbox = pn.widgets.Checkbox(value=False, name="Bathy Checkbox") data_table = pn.widgets.Tabulator(width=400, height=200) metadata_table = pn.widgets.Tabulator(width=600, height=800) # Download button is not working : TODO download_button = pn.widgets.Button(name="Download", button_type="primary") def __init__(self, **params): """ Initialize the SADCP_Viewer class. Parameters: **params: Additional parameters to be passed to the superclass. """ super(SADCP_Viewer, self).__init__(**params) self.file_dropdown.objects = self.file_names self.file_dropdown.value = ( self.file_dropdown.objects[0] if self.file_dropdown.objects else None ) self.update_name_options() @param.depends("year_slider.value", "file_dropdown.value", watch=True) def update_name_options(self): """ Update dropdown options and slider ranges based on selected years and file. This function updates the dropdown options and slider ranges based on the selected years and file. It also loads the selected file's data and adjusts slider ranges accordingly. """ # Extract selected start and end years start_year, end_year = self.year_slider.value # Filter DataFrame based on selected years and sort by year mask = (self.df["year"] >= start_year) & (self.df["year"] <= end_year) sorted_df = self.df[mask].sort_values(by="year") # Get unique file names files = sorted_df["file_name"].unique().tolist() # Update file dropdown options self.file_dropdown.options = files if files: selected_file = self.file_dropdown.value # Set default selected file if not selected or not in options if not selected_file or selected_file not in files: selected_file = files[0] self.file_dropdown.value = selected_file # Update data table and metadata table based on selected file self.data_table.value, self.metadata_table.value = filter_df(sorted_df, selected_file) # Load selected file's data self.ds = load_file(self.tree,selected_file) # Update slider ranges for longitude, latitude, and depth for slider, coord in zip([self.longitude_slider, self.latitude_slider, self.depth_range_slider, self.depth_2_range_slider, self.depth_3_range_slider], [self.ds.LONGITUDE, self.ds.LATITUDE, self.ds.PROFZ,self.ds.PROFZ,self.ds.PROFZ]): coord_range = get_range(coord) slider.start, slider.end, slider.value = coord_range[0], coord_range[1], coord_range # Close dataset to free up resources # self.ds.close() @param.depends( "year_slider.value", "file_dropdown.value", "depth_range_slider.value", "depth_2_checkbox.value", "depth_3_checkbox.value", "depth_2_range_slider.value", "depth_3_range_slider.value", "longitude_slider.value", "latitude_slider.value", "num_vectors_slider.value", "scale_factor_slider.value", "bathy_checkbox.value", watch=False,) def update_plots(self): """ This function updates the plots based on the selected data and parameters. The function filters the data, generates additional plots, and updates the main vector plot based on the selected parameters. Returns: pn.Row: A Panel row containing the updated map plot and additional plots. """ # Filter the data self.ds_filtered = filter_data(self.ds,self.longitude_slider.value,self.latitude_slider.value) # Prepare the plots shown in left # Update vector plots vector_plot = vectors_plot(self.ds_filtered, self.bathy, self.longitude_slider.value, self.latitude_slider.value, self.depth_range_slider.value, self.depth_2_range_slider.value, self.depth_3_range_slider.value, self.scale_factor_slider.value,self.num_vectors_slider.value, depth_2_checkbox= self.depth_2_checkbox.value, depth_3_checkbox= self.depth_3_checkbox.value, bathy_checkbox=self.bathy_checkbox.value, ) # Generate plots which will be plotted on the left row. self.plot_left = pn.Column( # Here adjust the style option later TODO # https://panel.holoviz.org/how_to/styling/matplotlib.html pn.pane.Matplotlib(vector_plot, dpi=144), # Add here the hvplot block of contour TODO sizing_mode="stretch_both") # Generate additional plots which will be plotted on the right row. other_plots = bathy_uship_vship_bottom_depth(self.ds_filtered) self.plot_right = pn.Column( *(pn.pane.HoloViews(plot, width=400, height=200) for plot in other_plots), sizing_mode="stretch_width" ) # Return a Panel row containing the updated map plot and additional plots return pn.Row(self.plot_left, self.plot_right, sizing_mode="stretch_both") pn.extension("tabulator") pn.config.theme = 'dark' explorer = SADCP_Viewer() # Instantiate the SADCP_Viewer class and create a template tabs = pn.Tabs( ("Plots", pn.Column(explorer.update_plots)), ( "Metadata", pn.Column( explorer.metadata_table, explorer.download_button, height=500, margin=10 ), ), ) sidebar = [ pn.panel('./EuroGO-SHIP_logo_wide_tagline_1.2.png',width=300 ), """This application, developed in the frame of Euro Go Shop, helps to interactively visualise and download ship ADCP data.""", explorer.year_slider, explorer.file_dropdown, explorer.longitude_slider, explorer.latitude_slider, explorer.bathy_checkbox, explorer.depth_range_slider, explorer.depth_2_checkbox, explorer.depth_3_checkbox, explorer.depth_2_range_slider, explorer.depth_3_range_slider, explorer.num_vectors_slider, explorer.scale_factor_slider, explorer.data_table, """You can consult detailed information on this data in the metadata tab shown on the right. To download full dataset, please go to https://cdi.seadatanet.org/search and search with LOCAL_CDI_ID indicated above.""", #pn.panel('https://upload.wikimedia.org/wikipedia/commons/4/47/PNG_transparency_demonstration_1.png', ), #width=10) ] template = pn.template.FastListTemplate( title="SADCP data Viewer", logo='https://avatars.githubusercontent.com/u/123177533?s=200&v=4', sidebar=sidebar, main=[tabs] ) template.servable()