app.py

# Gradio app that takes seismic waveform as input and marks 2 phases on the waveform as output.

import gradio as gr
import numpy as np
import pandas as pd
from phasehunter.data_preparation import prepare_waveform
import torch
import io

from scipy.stats import gaussian_kde
from bmi_topography import Topography
import earthpy.spatial as es

import obspy
from obspy.clients.fdsn import Client
from obspy.clients.fdsn.header import FDSNNoDataException, FDSNTimeoutException, FDSNInternalServerException
from obspy.geodetics.base import locations2degrees
from obspy.taup import TauPyModel
from obspy.taup.helper_classes import SlownessModelError

from obspy.clients.fdsn.header import URL_MAPPINGS

import matplotlib.pyplot as plt
import matplotlib.dates as mdates
from mpl_toolkits.axes_grid1 import ImageGrid

from glob import glob

def make_prediction(waveform):
    waveform = np.load(waveform)
    processed_input = prepare_waveform(waveform)
    
    # Make prediction
    with torch.inference_mode():
        output = model(processed_input)

    p_phase = output[:, 0]
    s_phase = output[:, 1]

    return processed_input, p_phase, s_phase

def mark_phases(waveform, uploaded_file):

    if uploaded_file is not None:
        waveform = uploaded_file.name

    processed_input, p_phase, s_phase = make_prediction(waveform)

    # Create a plot of the waveform with the phases marked
    if sum(processed_input[0][2] == 0): #if input is 1C
        fig, ax = plt.subplots(nrows=2, figsize=(10, 2), sharex=True)

        ax[0].plot(processed_input[0][0], color='black', lw=1)
        ax[0].set_ylabel('Norm. Ampl.')

    else: #if input is 3C
        fig, ax = plt.subplots(nrows=4, figsize=(10, 6), sharex=True)
        ax[0].plot(processed_input[0][0], color='black', lw=1)
        ax[1].plot(processed_input[0][1], color='black', lw=1)
        ax[2].plot(processed_input[0][2], color='black', lw=1)

        ax[0].set_ylabel('Z')
        ax[1].set_ylabel('N')
        ax[2].set_ylabel('E')

    p_phase_plot = p_phase*processed_input.shape[-1]
    p_kde = gaussian_kde(p_phase_plot)
    p_dist_space = np.linspace( min(p_phase_plot)-10, max(p_phase_plot)+10, 500 )
    ax[-1].plot( p_dist_space, p_kde(p_dist_space), color='r')

    s_phase_plot = s_phase*processed_input.shape[-1]
    s_kde = gaussian_kde(s_phase_plot)
    s_dist_space = np.linspace( min(s_phase_plot)-10, max(s_phase_plot)+10, 500 )
    ax[-1].plot( s_dist_space, s_kde(s_dist_space), color='b')

    for a in ax:
        a.axvline(p_phase.mean()*processed_input.shape[-1], color='r', linestyle='--', label='P')
        a.axvline(s_phase.mean()*processed_input.shape[-1], color='b', linestyle='--', label='S')

    ax[-1].set_xlabel('Time, samples')
    ax[-1].set_ylabel('Uncert., samples')
    ax[-1].legend()

    plt.subplots_adjust(hspace=0., wspace=0.)

    # Convert the plot to an image and return it
    fig.canvas.draw()
    image = np.array(fig.canvas.renderer.buffer_rgba())
    plt.close(fig)
    return image

def bin_distances(distances, bin_size=10):
    # Bin the distances into groups of `bin_size` kilometers
    binned_distances = {}
    for i, distance in enumerate(distances):
        bin_index = distance // bin_size
        if bin_index not in binned_distances:
            binned_distances[bin_index] = (distance, i)
        elif i < binned_distances[bin_index][1]:
            binned_distances[bin_index] = (distance, i)

    # Select the first distance in each bin and its index
    first_distances = []
    for bin_index in binned_distances:
        first_distance, first_distance_index = binned_distances[bin_index]
        first_distances.append(first_distance_index)
    
    return first_distances

def variance_coefficient(residuals):
    # calculate the variance of the residuals
    var = residuals.var()
    # scale the variance to a coefficient between 0 and 1
    coeff = 1 - (var / (residuals.max() - residuals.min()))
    return coeff

def predict_on_section(client_name, timestamp, eq_lat, eq_lon, radius_km, source_depth_km, velocity_model, max_waveforms, conf_thres_P, conf_thres_S):
    distances, t0s, st_lats, st_lons, waveforms, names = [], [], [], [], [], []
    
    taup_model = TauPyModel(model=velocity_model)
    client = Client(client_name)

    window = radius_km / 111.2
    max_waveforms = int(max_waveforms)

    assert eq_lat - window > -90 and eq_lat + window < 90, "Latitude out of bounds"
    assert eq_lon - window > -180 and eq_lon + window < 180, "Longitude out of bounds"

    starttime = obspy.UTCDateTime(timestamp)
    endtime = starttime + 120

    try:
        print('Starting to download inventory')
        inv = client.get_stations(network="*", station="*", location="*", channel="*H*", 
                            starttime=starttime, endtime=endtime, 
                            minlatitude=(eq_lat-window), maxlatitude=(eq_lat+window),
                            minlongitude=(eq_lon-window), maxlongitude=(eq_lon+window), 
                            level='station')
        print('Finished downloading inventory')
        
    except (IndexError, FDSNNoDataException, FDSNTimeoutException, FDSNInternalServerException):
        fig, ax = plt.subplots()
        ax.text(0.5,0.5,'Something is wrong with the data provider, try another')
        fig.canvas.draw();
        image = np.array(fig.canvas.renderer.buffer_rgba())
        plt.close(fig)
        return image
    
    waveforms = []
    cached_waveforms = glob("data/cached/*.mseed")

    for network in inv:
        if network.code == 'SY':
            continue
        for station in network:
            print(f"Processing {network.code}.{station.code}...")
            distance = locations2degrees(eq_lat, eq_lon, station.latitude, station.longitude)

            arrivals = taup_model.get_travel_times(source_depth_in_km=source_depth_km, 
                                                    distance_in_degree=distance, 
                                                    phase_list=["P", "S"])

            if len(arrivals) > 0:

                starttime = obspy.UTCDateTime(timestamp) + arrivals[0].time - 15
                endtime = starttime + 60
                try:
                    filename=f'{network.code}_{station.code}_{starttime}'
                    if f"data/cached/{filename}.mseed" not in cached_waveforms:
                        print(f'Downloading waveform for {filename}')
                        waveform = client.get_waveforms(network=network.code, station=station.code, location="*", channel="*", 
                                                    starttime=starttime, endtime=endtime)
                        waveform.write(f"data/cached/{network.code}_{station.code}_{starttime}.mseed", format="MSEED")
                        print('Finished downloading and caching waveform')
                    else:
                        print('Reading cached waveform')
                        waveform = obspy.read(f"data/cached/{network.code}_{station.code}_{starttime}.mseed")
                        

                except (IndexError, FDSNNoDataException, FDSNTimeoutException, FDSNInternalServerException):
                    print(f'Skipping {network.code}_{station.code}_{starttime}')
                    continue
            
                waveform = waveform.select(channel="H[BH][ZNE]")
                waveform = waveform.merge(fill_value=0)
                waveform = waveform[:3].sort(keys=['channel'], reverse=True)

                len_check = [len(x.data) for x in waveform]
                if len(set(len_check)) > 1:
                    continue

                if len(waveform) == 3:
                    try:
                        waveform = prepare_waveform(np.stack([x.data for x in waveform]))

                        distances.append(distance)
                        t0s.append(starttime)
                        st_lats.append(station.latitude)
                        st_lons.append(station.longitude)
                        waveforms.append(waveform)
                        names.append(f"{network.code}.{station.code}")

                        print(f"Added {network.code}.{station.code} to the list of waveforms")

                    except:
                        continue
                
    
    # If there are no waveforms, return an empty plot
    if len(waveforms) == 0:
        print('No waveforms found')
        fig, ax = plt.subplots()
        ax.text(0.5,0.5,'No waveforms found')
        fig.canvas.draw();
        image = np.array(fig.canvas.renderer.buffer_rgba())
        plt.close(fig)
        output_picks = pd.DataFrame()
        output_picks.to_csv('data/picks.csv', index=False)
        output_csv = 'data/picks.csv'
        return image, output_picks, output_csv
    

    first_distances = bin_distances(distances, bin_size=10/111.2)

    # Edge case when there are way too many waveforms to process
    selection_indexes = np.random.choice(first_distances, 
                                         np.min([len(first_distances), max_waveforms]),
                                         replace=False)

    waveforms = np.array(waveforms)[selection_indexes]
    distances = np.array(distances)[selection_indexes]
    t0s = np.array(t0s)[selection_indexes]
    st_lats = np.array(st_lats)[selection_indexes]
    st_lons = np.array(st_lons)[selection_indexes]
    names = np.array(names)[selection_indexes]

    waveforms = [torch.tensor(waveform) for waveform in waveforms]

    print('Starting to run predictions')
    with torch.no_grad():
        waveforms_torch = torch.vstack(waveforms)
        output = model(waveforms_torch)

    p_phases = output[:, 0]
    s_phases = output[:, 1]

    p_phases = p_phases.reshape(len(waveforms),-1)
    s_phases = s_phases.reshape(len(waveforms),-1)

    # Max confidence - min variance    
    p_max_confidence = p_phases.std(axis=-1).min()
    s_max_confidence = s_phases.std(axis=-1).min()

    print(f"Starting plotting {len(waveforms)} waveforms")
    fig, ax = plt.subplots(ncols=3, figsize=(10, 3))
    
    # Plot topography
    print('Fetching topography')
    params = Topography.DEFAULT.copy()
    extra_window = 0.5
    params["south"] = np.min([st_lats.min(), eq_lat])-extra_window
    params["north"] = np.max([st_lats.max(), eq_lat])+extra_window
    params["west"] = np.min([st_lons.min(), eq_lon])-extra_window
    params["east"] = np.max([st_lons.max(), eq_lon])+extra_window

    topo_map = Topography(**params)
    topo_map.fetch()
    topo_map.load()

    print('Plotting topo')
    hillshade = es.hillshade(topo_map.da[0], altitude=10)
    
    topo_map.da.plot(ax = ax[1], cmap='Greys', add_colorbar=False, add_labels=False)
    topo_map.da.plot(ax = ax[2], cmap='Greys', add_colorbar=False, add_labels=False)
    ax[1].imshow(hillshade, cmap="Greys", alpha=0.5)

    output_picks = pd.DataFrame({'station_name' : [], 
                                'st_lat' : [], 'st_lon' : [],
                                 'starttime' : [], 
                                 'p_phase, s' : [], 'p_uncertainty, s' : [], 
                                 's_phase, s' : [], 's_uncertainty, s' : [],
                                 'velocity_p, km/s' : [], 'velocity_s, km/s' : []})
                        
    for i in range(len(waveforms)):
        print(f"Plotting waveform {i+1}/{len(waveforms)}")
        current_P = p_phases[i]
        current_S = s_phases[i]
        
        x = [t0s[i] + pd.Timedelta(seconds=k/100) for k in np.linspace(0,6000,6000)]
        x = mdates.date2num(x)

        # Normalize confidence for the plot
        p_conf = 1/(current_P.std()/p_max_confidence).item()
        s_conf = 1/(current_S.std()/s_max_confidence).item()

        delta_t = t0s[i].timestamp - obspy.UTCDateTime(timestamp).timestamp

        ax[0].plot(x, waveforms[i][0, 0]*10+distances[i]*111.2, color='black', alpha=0.5, lw=1)

        if (current_P.std().item()*60 < conf_thres_P) or (current_S.std().item()*60 < conf_thres_S):
            ax[0].scatter(x[int(current_P.mean()*waveforms[i][0].shape[-1])], waveforms[i][0, 0].mean()+distances[i]*111.2, color='r', alpha=p_conf, marker='|')
            ax[0].scatter(x[int(current_S.mean()*waveforms[i][0].shape[-1])], waveforms[i][0, 0].mean()+distances[i]*111.2, color='b', alpha=s_conf, marker='|')
        
            velocity_p = (distances[i]*111.2)/(delta_t+current_P.mean()*60).item()
            velocity_s = (distances[i]*111.2)/(delta_t+current_S.mean()*60).item()

            # Generate an array from st_lat to eq_lat and from st_lon to eq_lon
            x = np.linspace(st_lons[i], eq_lon, 50)
            y = np.linspace(st_lats[i], eq_lat, 50)
            
            # Plot the array
            ax[1].scatter(x, y, c=np.zeros_like(x)+velocity_p, alpha=0.1, vmin=0, vmax=8)
            ax[2].scatter(x, y, c=np.zeros_like(x)+velocity_s, alpha=0.1, vmin=0, vmax=8)

        else:
            velocity_p = np.nan
            velocity_s = np.nan
        
        ax[0].set_ylabel('Z')
        print(f"Station {st_lats[i]}, {st_lons[i]} has P velocity {velocity_p} and S velocity {velocity_s}")

        output_picks = output_picks.append(pd.DataFrame({'station_name': [names[i]], 
                                                        'st_lat' : [st_lats[i]], 'st_lon' : [st_lons[i]],
                                                        'starttime' : [str(t0s[i])], 
                                                        'p_phase, s' : [(delta_t+current_P.mean()*60).item()], 'p_uncertainty, s' : [current_P.std().item()*60], 
                                                        's_phase, s' : [(delta_t+current_S.mean()*60).item()], 's_uncertainty, s' : [current_S.std().item()*60],
                                                        'velocity_p, km/s' : [velocity_p], 'velocity_s, km/s' : [velocity_s]}))
        
        
    # Add legend
    ax[0].scatter(None, None, color='r', marker='|', label='P')
    ax[0].scatter(None, None, color='b', marker='|', label='S')
    ax[0].xaxis.set_major_formatter(mdates.DateFormatter('%H:%M:%S'))
    ax[0].xaxis.set_major_locator(mdates.SecondLocator(interval=20))
    ax[0].legend()

    print('Plotting stations')
    for i in range(1,3):
        ax[i].scatter(st_lons, st_lats, color='b', label='Stations')
        ax[i].scatter(eq_lon, eq_lat, color='r', marker='*', label='Earthquake')
        ax[i].set_aspect('equal')
        ax[i].set_xticklabels(ax[i].get_xticks(), rotation = 50)

    fig.subplots_adjust(bottom=0.1, top=0.9, left=0.1, right=0.8,
                    wspace=0.02, hspace=0.02)
    
    cb_ax = fig.add_axes([0.83, 0.1, 0.02, 0.8])
    cbar = fig.colorbar(ax[2].scatter(None, None, c=velocity_p, alpha=0.5, vmin=0, vmax=8), cax=cb_ax)

    cbar.set_label('Velocity (km/s)')
    ax[1].set_title('P Velocity')
    ax[2].set_title('S Velocity')

    for a in ax:
        a.tick_params(axis='both', which='major', labelsize=8)
        
    plt.subplots_adjust(hspace=0., wspace=0.5)
    fig.canvas.draw();
    image = np.array(fig.canvas.renderer.buffer_rgba())
    plt.close(fig)
    output_picks.to_csv(f'data/velocity/{eq_lat}_{eq_lon}_{timestamp}_{len(waveforms)}.csv', index=False)
    output_csv = f'data/velocity/{eq_lat}_{eq_lon}_{timestamp}_{len(waveforms)}.csv'

    return image, output_picks, output_csv

model = torch.jit.load("model.pt")

with gr.Blocks() as demo:
    gr.HTML("""
<div style="padding: 20px; border-radius: 10px;">
    <h1 style="font-size: 30px; text-align: center; margin-bottom: 20px;">PhaseHunter <span style="animation: arrow-anim 10s linear infinite; display: inline-block; transform: rotate(45deg) translateX(-20px);">🏹</span>

<style>
    @keyframes arrow-anim {
        0% { transform: translateX(-20px); }
        50% { transform: translateX(20px); }
        100% { transform: translateX(-20px); }
    }
</style></h1> 
    
    <p style="font-size: 16px; margin-bottom: 20px;">Detect <span style="background-image: linear-gradient(to right, #ED213A, #93291E); 
    -webkit-background-clip: text;
    -webkit-text-fill-color: transparent;
    background-clip: text;">P</span> and <span style="background-image: linear-gradient(to right, #00B4DB, #0083B0); 
    -webkit-background-clip: text;
    -webkit-text-fill-color: transparent;
    background-clip: text;">S</span> seismic phases with <span style="background-image: linear-gradient(to right, #f12711, #f5af19); 
    -webkit-background-clip: text;
    -webkit-text-fill-color: transparent;
    background-clip: text;">uncertainty</span></p>
    <ul style="font-size: 16px; margin-bottom: 40px;">
        <li>Detect seismic phases by selecting a sample waveform or uploading your own waveform in <code>.npy</code> format.</li>
        <li>Select an earthquake from the global earthquake catalogue and PhaseHunter will analyze seismic stations in the given radius.</li>
        <li>Waveforms should be sampled at 100 samples/sec and have 3 (Z, N, E) or 1 (Z) channels. PhaseHunter analyzes the first 6000 samples of your file.</li>
    </ul>
</div>
""")

    with gr.Tab("Try on a single station"):
        with gr.Row(): 
            # Define the input and output types for Gradio
            inputs = gr.Dropdown(
                ["data/sample/sample_0.npy", 
                "data/sample/sample_1.npy", 
                "data/sample/sample_2.npy"], 
                label="Sample waveform", 
                info="Select one of the samples",
                value = "data/sample/sample_0.npy"
            )

            upload = gr.File(label="Or upload your own waveform")

        button = gr.Button("Predict phases")
        outputs = gr.Image(label='Waveform with Phases Marked', type='numpy', interactive=False)
    
        button.click(mark_phases, inputs=[inputs, upload], outputs=outputs)
        
    with gr.Tab("Select earthquake from catalogue"):

        gr.HTML("""
        <div style="padding: 20px; border-radius: 10px; font-size: 16px;">
        <p style="font-weight: bold; font-size: 24px; margin-bottom: 20px;">Using PhaseHunter to Analyze Seismic Waveforms</p>
        <p>Select an earthquake from the global earthquake catalogue (e.g. <a href="https://earthquake.usgs.gov/earthquakes/map">USGS</a>) and the app will download the waveform from the FDSN client of your choice. The app will use a velocity model of your choice to select appropriate time windows for each station within a specified radius of the earthquake.</p>
        <p>The app will then analyze the waveforms and mark the detected phases on the waveform. Pick data for each waveform is reported in seconds from the start of the waveform.</p>
        <p>Velocities are derived from distance and travel time determined by PhaseHunter picks (<span style="font-style: italic;">v = distance/predicted_pick_time</span>). The background of the velocity plot is colored by DEM.</p>
        </div>
        """)
        with gr.Row(): 
            with gr.Column(scale=2):
                client_inputs = gr.Dropdown(
                    choices = list(URL_MAPPINGS.keys()), 
                    label="FDSN Client", 
                    info="Select one of the available FDSN clients",
                    value = "IRIS",
                    interactive=True
                )

                velocity_inputs = gr.Dropdown(
                    choices = ['1066a', '1066b', 'ak135', 
                            'ak135f', 'herrin', 'iasp91', 
                            'jb', 'prem', 'pwdk'], 
                    label="1D velocity model", 
                    info="Velocity model for station selection",
                    value = "1066a",
                    interactive=True
                )

            with gr.Column(scale=2):
                timestamp_inputs = gr.Textbox(value='2019-07-04 17:33:49',
                                    placeholder='YYYY-MM-DD HH:MM:SS',
                                    label="Timestamp",
                                    info="Timestamp of the earthquake",
                                    max_lines=1,
                                    interactive=True)
                
                source_depth_inputs = gr.Number(value=10,
                    label="Source depth (km)",
                    info="Depth of the earthquake",
                    interactive=True)
                
            with gr.Column(scale=2):
                eq_lat_inputs = gr.Number(value=35.766, 
                                label="Latitude", 
                                info="Latitude of the earthquake",
                                interactive=True)
                
                eq_lon_inputs = gr.Number(value=-117.605,
                                    label="Longitude",
                                    info="Longitude of the earthquake",
                                    interactive=True)
                
            with gr.Column(scale=2):
                radius_inputs = gr.Slider(minimum=1, 
                                        maximum=200, 
                                        value=50, 
                                        label="Radius (km)", 
                                        step=10,
                                        info="""Select the radius around the earthquake to download data from.\n 
                                        Note that the larger the radius, the longer the app will take to run.""",
                                        interactive=True)
                
                max_waveforms_inputs = gr.Slider(minimum=1,
                                maximum=100,
                                value=10,
                                label="Max waveforms per section",
                                step=1,
                                info="Maximum number of waveforms to show per section\n (to avoid long prediction times)",
                                interactive=True,
                                )
            with gr.Column(scale=2):
                P_thres_inputs = gr.Slider(minimum=0.01,
                                maximum=1,
                                value=0.1,
                                label="P uncertainty threshold, s",
                                step=0.01,
                                info="Acceptable uncertainty for P picks expressed in std() seconds",
                                interactive=True,
                                )
                S_thres_inputs = gr.Slider(minimum=0.01,
                                maximum=1,
                                value=0.2,
                                label="S uncertainty threshold, s",
                                step=0.01,
                                info="Acceptable uncertainty for S picks expressed in std() seconds",
                                interactive=True,
                                )
            
        button = gr.Button("Predict phases")
        output_image = gr.Image(label='Waveforms with Phases Marked', type='numpy', interactive=False)

        with gr.Row():
            output_picks = gr.Dataframe(label='Pick data', 
                                        type='pandas', 
                                        interactive=False)
            output_csv = gr.File(label="Output File", file_types=[".csv"])

        button.click(predict_on_section, 
                 inputs=[client_inputs, timestamp_inputs, 
                         eq_lat_inputs, eq_lon_inputs, 
                         radius_inputs, source_depth_inputs, 
                         velocity_inputs, max_waveforms_inputs,
                         P_thres_inputs, S_thres_inputs],
                 outputs=[output_image, output_picks, output_csv])

demo.launch()