management.py

"""MIT License

Copyright (c) 2019, Swiss Federal Institute of Technology (ETH Zurich), Matthias Meyer

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE."""

from ..global_config import get_setting, setting_exists, set_setting
from ..core.graph import StuettNode, configuration
from ..convenience import read_csv_with_store, to_csv_with_store, DirectoryStore

import os

import dask
import logging
import obspy
from obspy.clients.fdsn import Client
from obspy import UTCDateTime
from obsplus import obspy_to_array
from copy import deepcopy
import io

import zarr
import xarray as xr
from PIL import Image
import base64
import re

from pathlib import Path
import warnings

# TODO: revisit the following packages
import numpy as np
import pandas as pd
import datetime as dt
from pathlib import Path


class DataSource(StuettNode):
    def __init__(self, **kwargs):
        super().__init__(kwargs=kwargs)

    def __call__(self, data=None, request=None, delayed=False):
        if data is not None:
            if request is None:
                request = data
            else:
                warnings.warning(
                    "Two inputs (data, request) provided to the DataSource but it can only handle a request. Choosing request. "
                )

        # DataSource only require a request
        # Therefore merge permanent-config and request
        config = self.config.copy()  # TODO: do we need a deep copy?
        if request is not None:
            config.update(request)

        # TODO: change when rewriting for general indices
        if "start_time" in config and config["start_time"] is not None:
            config["start_time"] = pd.to_datetime(
                config["start_time"], utc=True
            ).tz_localize(
                None
            )  # TODO: change when xarray #3291 is fixed
        if "end_time" in config and config["end_time"] is not None:
            config["end_time"] = pd.to_datetime(
                config["end_time"], utc=True
            ).tz_localize(
                None
            )  # TODO: change when xarray #3291 is fixed

        if delayed:
            return dask.delayed(self.forward)(None, config)
        else:
            return self.forward(None, config)

    def configure(self, requests=None):
        """ Default configure for DataSource nodes
            Same as configure from StuettNode but adds is_source flag

        Arguments:
            request {list} -- List of requests

        Returns:
            dict -- Original request or merged requests 
        """
        requests = super().configure(requests)  # merging request here
        requests["requires_request"] = True

        return requests


class GSNDataSource(DataSource):
    def __init__(
        self,
        deployment=None,
        vsensor=None,
        position=None,
        start_time=None,
        end_time=None,
        **kwargs,
    ):
        super().__init__(
            deployment=deployment,
            position=position,
            vsensor=vsensor,
            start_time=start_time,
            end_time=end_time,
            kwargs=kwargs,
        )

    def forward(self, data=None, request=None):
        # code from Samuel Weber
        #### 1 - DEFINE VSENSOR-DEPENDENT COLUMNS ####
        metadata = Path(get_setting("metadata_directory")).joinpath(
            "vsensor_metadata/{:s}_{:s}.csv".format(
                request["deployment"], request["vsensor"]
            )
        )
        # if not metadata.exists():

        colnames = pd.read_csv(metadata, skiprows=0,)
        columns_old = colnames["colname_old"].values
        columns_new = colnames["colname_new"].values
        columns_unit = colnames["unit"].values
        if len(columns_old) != len(columns_new):
            warnings.warn(
                "WARNING: Length of 'columns_old' ({:d}) is not equal length of  'columns_new' ({:d})".format(
                    len(columns_old), len(columns_new)
                )
            )
        if len(columns_old) != len(columns_unit):
            warnings.warn(
                "WARNING: Length of 'columns_old' ({:d}) is not equal length of  'columns_new' ({:d})".format(
                    len(columns_old), len(columns_unit)
                )
            )

        unit = dict(zip(columns_new, columns_unit))
        #### 2 - DEFINE CONDITIONS AND CREATE HTTP QUERY ####

        # Set server
        server = get_setting("permasense_server")

        # Create virtual_sensor
        virtual_sensor = request["deployment"] + "_" + request["vsensor"]

        # Create query and add time as well as position selection
        query = (
            "vs[1]={:s}"
            "&time_format=iso"
            "&timeline=generation_time"
            "&field[1]=All"
            "&from={:s}"
            "&to={:s}"
            "&c_vs[1]={:s}"
            "&c_join[1]=and"
            "&c_field[1]=position"
            "&c_min[1]={:02d}"
            "&c_max[1]={:02d}"
        ).format(
            virtual_sensor,
            pd.to_datetime(request["start_time"], utc=True).strftime(
                "%d/%m/%Y+%H:%M:%S"
            ),
            pd.to_datetime(request["end_time"], utc=True).strftime("%d/%m/%Y+%H:%M:%S"),
            virtual_sensor,
            int(request["position"]) - 1,
            request["position"],
        )

        # query extension for images
        if request["vsensor"] == "binary__mapped":
            query = (
                query
                + "&vs[2]={:s}&field[2]=relative_file&c_join[2]=and&c_vs[2]={:s}&c_field[2]=file_complete&c_min[2]=0&c_max[2]=1&vs[3]={:s}&field[3]=generation_time&c_join[3]=and&c_vs[3]={:s}&c_field[3]=file_size&c_min[3]=2000000&c_max[3]=%2Binf&download_format=csv".format(
                    virtual_sensor, virtual_sensor, virtual_sensor, virtual_sensor
                )
            )

        # Construct url:
        url = server + "multidata?" + query
        # if self.verbose:
        #     print('The GSN http-query is:\n{:s}'.format(url))

        #### 3 - ACCESS DATA AND CREATE PANDAS DATAFRAME ####
        d = []
        d = pd.read_csv(
            url, skiprows=2
        )  # skip header lines (first 2) for import: skiprows=2
        df = pd.DataFrame(columns=columns_new)
        df.time = pd.to_datetime(d.generation_time, utc=True)

        # if depo in ['mh25', 'mh27old', 'mh27new', 'mh30', 'jj22', 'jj25', 'mh-nodehealth']:
        # d = d.convert_objects(convert_numeric=True)  # TODO: rewrite to remove warning
        for k in list(df):
            df[k] = pd.to_numeric(df[k], errors="ignore")

        #        df = pd.DataFrame(columns=columns_new)
        #        df.time = timestamp2datetime(d['generation_time']/1000)
        for i in range(len(columns_old)):
            if columns_new[i] != "time":
                setattr(df, columns_new[i], getattr(d, columns_old[i]))

        df = df.sort_values(by="time")

        # Remove columns with names 'delete'
        try:
            df.drop(["delete"], axis=1, inplace=True)
        except:
            pass

        # Remove columns with only 'null'
        df = df.replace(r"null", np.nan, regex=True)
        isnull = df.isnull().all()
        [df.drop([col_name], axis=1, inplace=True) for col_name in df.columns[isnull]]

        df = df.set_index("time")
        df = df.sort_index(axis=1)

        x = xr.DataArray(df, dims=["time", "name"], name="CSV")

        try:
            unit_coords = []
            for name in x.coords["name"].values:
                # name = re.sub(r"\[.*\]", "", name).lstrip().rstrip()
                u = unit[str(name)]
                u = re.findall(r"\[(.*)\]", u)[0]

                # name_coords.append(name)
                unit_coords.append(u)

            x = x.assign_coords({"unit": ("name", unit_coords)})
        except:
            # TODO: add a warning or test explicitly if units exist
            pass

        return x


class SeismicSource(DataSource):
    def __init__(
        self,
        path=None,
        store=None,
        station=None,
        channel=None,
        start_time=None,
        end_time=None,
        use_arclink=False,
        return_obspy=False,
        **kwargs,
    ):  # TODO: update description
        """ Seismic data source to get data from permasense
            The user can predefine the source's settings or provide them in a request
            Predefined setting should never be updated (to stay thread safe), but will be ignored
            if request contains settings

        Keyword Arguments:
            config {dict}       -- Configuration for the seismic source (default: {{}})
            use_arclink {bool}  -- If true, downloads the data from the arclink service (authentication required) (default: {False})
            return_obspy {bool} -- By default an xarray is returned. If true, an obspy stream will be returned (default: {False})
        """
        super().__init__(
            path=path,
            store=store,
            station=station,
            channel=channel,
            start_time=start_time,
            end_time=end_time,
            use_arclink=use_arclink,
            return_obspy=return_obspy,
            kwargs=kwargs,
        )

    def forward(self, data=None, request=None):
        config = request

        if config["use_arclink"]:
            try:
                arclink = get_setting("arclink")
            except KeyError as err:
                raise RuntimeError(
                    f"The following error occured \n{err}. "
                    "Please provide either the credentials to access arclink or a path to the dataset"
                )

            arclink_user = arclink["user"]
            arclink_password = arclink["password"]
            fdsn_client = Client(
                base_url="http://arclink.ethz.ch",
                user=arclink_user,
                password=arclink_password,
            )
            x = fdsn_client.get_waveforms(
                network="4D",
                station=config["station"],
                location="A",
                channel=config["channel"],
                starttime=UTCDateTime(config["start_time"]),
                endtime=UTCDateTime(config["end_time"]),
                attach_response=True,
            )

            # TODO: potentially resample

        else:  # 20180914 is last full day available in permasense_vault
            # logging.info('Loading seismic with fdsn')
            x = self.get_obspy_stream(
                config,
                config["path"],
                config["start_time"],
                config["end_time"],
                config["station"],
                config["channel"],
            )

        x = x.slice(UTCDateTime(config["start_time"]), UTCDateTime(config["end_time"]))

        # TODO: remove response x.remove_response(output=vel)
        # x = self.process_seismic_data(x)

        if not config["return_obspy"]:
            x = obspy_to_array(x)

            # we assume that all starttimes are equal
            starttime = x.starttime.values.reshape((-1,))[0]
            for s in x.starttime.values.reshape((-1,)):
                if s != starttime:
                    raise RuntimeError(
                        "Please make sure that starttime of each seimsic channel is equal"
                    )

            # change time coords from relative to absolute time
            starttime = obspy.UTCDateTime(starttime).datetime
            starttime = pd.to_datetime(starttime, utc=True).tz_localize(
                None
            )  # TODO: change when xarray #3291 is fixed
            timedeltas = pd.to_timedelta(x["time"].values, unit="seconds")
            xt = starttime + timedeltas
            x["time"] = pd.to_datetime(xt, utc=True).tz_localize(
                None
            )  # TODO: change when xarray #3291 is fixed
            del x.attrs["stats"]

        return x

    def process_seismic_data(
        self,
        stream,
        remove_response=True,
        unit="VEL",
        station_inventory=None,
        detrend=True,
        taper=False,
        pre_filt=(0.025, 0.05, 45.0, 49.0),
        water_level=60,
        apply_filter=True,
        freqmin=0.002,
        freqmax=50,
        resample=False,
        resample_rate=250,
        rotation_angle=None,
    ):
        # author: Samuel Weber
        if station_inventory is None:
            station_inventory = Path(get_setting("metadata_directory")).joinpath(
                "inventory_stations__MH.xml"
            )

        # print(station_inventory)
        inv = obspy.read_inventory(str(station_inventory))
        # st = stream.copy()
        st = stream
        st.attach_response(inv)

        if detrend:
            st.detrend("demean")
            st.detrend("linear")

        if taper:
            st.taper(max_percentage=0.05)

        if remove_response:
            if hasattr(st[0].stats, "response"):
                st.remove_response(
                    output=unit,
                    pre_filt=pre_filt,
                    plot=False,
                    zero_mean=False,
                    taper=False,
                    water_level=water_level,
                )
            else:
                st.remove_response(
                    output=unit,
                    inventory=inv,
                    pre_filt=pre_filt,
                    plot=False,
                    zero_mean=False,
                    taper=False,
                    water_level=water_level,
                )

        if detrend:
            st.detrend("demean")
            st.detrend("linear")

        if taper:
            st.taper(max_percentage=0.05)

        if apply_filter:
            st.filter("bandpass", freqmin=freqmin, freqmax=freqmax)

        if resample:
            st.resample(resample_rate)

        if rotation_angle is None:
            rotation_angle = np.nan
        if not np.isnan(rotation_angle):
            st.rotate("NE->RT", back_azimuth=int(rotation_angle), inventory=inv)

        return st

    def get_obspy_stream(
        self,
        request,
        path,
        start_time,
        end_time,
        station,
        channels,
        pad=False,
        verbose=False,
        fill=0,
        fill_sampling_rate=1000,
        old_stationname=False,
    ):
        """    
        Loads the microseismic data for the given timeframe into a miniseed file.

        Arguments:
            start_time {datetime} -- start timestamp of the desired obspy stream
            end_time {datetime} -- end timestamp of the desired obspy stream
        
        Keyword Arguments:
            pad {bool} -- If padding is true, the data will be zero padded if the data is not consistent
            fill {} -- If numpy.nan or fill value: error in the seismic stream will be filled with the value. If None no fill will be used
            verbose {bool} -- If info should be printed

        Returns:
            obspy stream -- obspy stream with up to three channels
                            the stream's channels will be sorted alphabetically
        """

        if not isinstance(channels, list):
            channels = [channels]

        # We will get the full hours seismic data and trim it to the desired length afterwards
        tbeg_hours = pd.to_datetime(start_time).replace(
            minute=0, second=0, microsecond=0
        )
        timerange = pd.date_range(start=tbeg_hours, end=end_time, freq="H")

        non_existing_files_ts = []  # keep track of nonexisting files

        # drawback of memmap files is that we need to calculate the size beforehand
        stream = obspy.Stream()

        idx = 0
        ## loop through all hours
        for i in range(len(timerange)):
            # start = time.time()
            h = timerange[i]

            st_list = obspy.Stream()

            datayear = timerange[i].strftime("%Y")
            if old_stationname:
                station = (
                    "MHDL" if station == "MH36" else "MHDT"
                )  # TODO: do not hardcode it
            filenames = {}
            for channel in channels:
                # filenames[channel] = Path(station).joinpath(
                #     datayear,
                #     "%s.D/" % channel,
                #     "4D.%s.A.%s.D." % (station, channel)
                #     + timerange[i].strftime("%Y%m%d_%H%M%S")
                #     + ".miniseed",
                # )
                filenames[channel] = [
                    station,
                    datayear,
                    "%s.D" % channel,
                    "4D.%s.A.%s.D." % (station, channel)
                    + timerange[i].strftime("%Y%m%d_%H%M%S")
                    + ".miniseed",
                ]
                # print(filenames[channel])

                # Load either from store or from filename
                if request["store"] is not None:
                    # get the file relative to the store
                    store = request["store"]
                    filename = "/".join(filenames[channel])
                    st = obspy.read(io.BytesIO(store[str(filename)]))
                else:
                    if not Path(path).isdir():
                        # TODO: should this be an error or only a warning. In a period execution this could stop the whole script
                        raise IOError(
                            "Cannot find the path {}. Please provide a correct path to the permasense geophone directory".format(
                                datadir
                            )
                        )
                    datadir = Path(path)
                    if not datadir.joinpath(*filenames[channel]).exists():
                        non_existing_files_ts.append(timerange[i])

                        warnings.warn(
                            RuntimeWarning(
                                "Warning file not found for {} {}".format(
                                    timerange[i], datadir.joinpath(filenames[channel])
                                )
                            )
                        )

                        if fill is not None:
                            # create empty stream of fill value
                            arr = (
                                np.ones(
                                    (
                                        int(
                                            np.ceil(
                                                fill_sampling_rate
                                                * dt.timedelta(hours=1).total_seconds()
                                            )
                                        ),
                                    )
                                )
                                * fill
                            )
                            st = obspy.Stream([obspy.Trace(arr)])
                        else:
                            continue
                    else:
                        filename = str(datadir.joinpath(*filenames[channel]))
                        st = obspy.read(filename)

                st_list += st

            stream_h = st_list.merge(method=0, fill_value=fill)
            start_time_0 = start_time if i == 0 else h
            end_time_0 = (
                end_time if i == len(timerange) - 1 else h + dt.timedelta(hours=1)
            )
            segment_h = stream_h.trim(
                obspy.UTCDateTime(start_time_0),
                obspy.UTCDateTime(end_time_0),
                pad=pad,
                fill_value=fill,
            )
            stream += segment_h

        stream = stream.merge(method=0, fill_value=fill)

        stream.sort(
            keys=["channel"]
        )  # TODO: change this so that the order of the input channels list is maintained

        return stream


class MHDSLRFilenames(DataSource):
    def __init__(
        self,
        base_directory=None,
        store=None,
        method="directory",
        start_time=None,
        end_time=None,
        force_write_to_remote=False,
    ):
        """ Fetches the DSLR images from the Matterhorn deployment, returns the image
            filename(s) corresponding to the end and start time provided in either the
            config dict or as a request to the __call__() function.          
        
        Arguments:
            StuettNode {[type]} -- [description]

        Keyword Arguments:
            base_directory {[type]} -- [description]
            method {str}     -- [description] (default: {'directory'})
        """
        super().__init__(
            base_directory=base_directory,
            store=store,
            method=method,
            start_time=start_time,
            end_time=end_time,
            force_write_to_remote=force_write_to_remote,
        )

    def forward(self, data=None, request=None):
        """Retrieves the images for the selected time period from the server. If only a start_time timestamp is provided, 
          the file with the corresponding date will be loaded if available. For periods (when start and end time are given) 
          all available images are indexed first to provide an efficient retrieval.
        
        Arguments:
            start_time {datetime} -- If only start_time is given the neareast available image is return. If also end_time is provided the a dataframe is returned containing image filenames from the first image after start_time until the last image before end_time.
        
        Keyword Arguments:
            end_time {datetime} -- end time of the selected period. see start_time for a description. (default: {None})
        
        Returns:
            dataframe -- Returns containing the image filenames of the selected period.
        """
        config = request
        methods = ["directory", "web"]
        if config["method"].lower() not in methods:
            raise RuntimeError(
                f"The {config['method']} output_format is not supported. Allowed formats are {methods}"
            )

        if (
            config["base_directory"] is None and config["store"] is None
        ) and output_format.lower() != "web":
            raise RuntimeError("Please provide a base_directory containing the images")

        if config["method"].lower() == "web":  # TODO: implement
            raise NotImplementedError("web fetch has not been implemented yet")

        start_time = config["start_time"]
        end_time = config["end_time"]

        if start_time is None:
            raise RuntimeError("Please provide a least start time")

        # if it is not timezone aware make it
        # if start_time.tzinfo is None:
        #     start_time = start_time.replace(tzinfo=timezone.utc)

        # If there is no tmp_dir we can try to load the file directly, otherwise
        # there will be a warning later in this function and the user should
        # set a tmp_dir
        if end_time is None and (
            not setting_exists("user_dir") or not os.path.isdir(get_setting("user_dir"))
        ):
            image_filename = self.get_image_filename(start_time)
            if image_filename is not None:
                return image_filename

        # If we have already loaded the dataframe in the current session we can use it
        if setting_exists("image_list_df"):
            imglist_df = get_setting("image_list_df")
        else:
            filename = "image_integrity.csv"
            success = False
            # first try to load it from remote via store
            if config["store"] is not None:
                if filename in config["store"]:
                    imglist_df = read_csv_with_store(config["store"], filename)
                    success = True
                elif config["force_write_to_remote"]:
                    # try to reload it and write to remote
                    imglist_df = self.image_integrity_store(config["store"])
                    try:
                        to_csv_with_store(config["store"], filename, imglist_df, dict(index=False))
                        success = True
                    except Exception as e:
                        print(e)

            # Otherwise we need to load the filename dataframe from disk
            if (
                not success
                and setting_exists("user_dir")
                and os.path.isdir(get_setting("user_dir"))
            ):
                imglist_filename = os.path.join(get_setting("user_dir"), "") + filename

                # If it does not exist in the temporary folder of our application
                # We are going to create it
                if os.path.isfile(imglist_filename):
                    # imglist_df = pd.read_parquet(
                    #     imglist_filename
                    # )  # TODO: avoid too many different formats
                    imglist_df = pd.read_csv(imglist_filename)
                else:
                    # we are going to load the full list => no arguments
                    imglist_df = self.image_integrity(config["base_directory"])
                    # imglist_df.to_parquet(imglist_filename)
                    imglist_df.to_csv(imglist_filename, index=False)
            elif not success:
                # if there is no tmp_dir we can load the image list but
                # we should warn the user that this is inefficient
                imglist_df = self.image_integrity(config["base_directory"])
                warnings.warn(
                    "No temporary directory was set. You can speed up multiple runs of your application by setting a temporary directory"
                )

            # TODO: make the index timezone aware
            imglist_df.set_index("start_time", inplace=True)
            imglist_df.index = pd.to_datetime(imglist_df.index, utc=True).tz_localize(
                None
            )  # TODO: change when xarray #3291 is fixed
            imglist_df.sort_index(inplace=True)

            set_setting("image_list_df", imglist_df)

        if end_time is None:
            if start_time < imglist_df.index[0]:
                start_time = imglist_df.index[0]

            return imglist_df.iloc[
                imglist_df.index.get_loc(start_time, method="nearest")
            ]
        else:
            # if end_time.tzinfo is None:
            #     end_time = end_time.replace(tzinfo=timezone.utc)
            if start_time > imglist_df.index[-1] or end_time < imglist_df.index[0]:
                # return empty dataframe
                return imglist_df[0:0]

            if start_time < imglist_df.index[0]:
                start_time = imglist_df.index[0]
            if end_time > imglist_df.index[-1]:
                end_time = imglist_df.index[-1]

            return imglist_df.iloc[
                imglist_df.index.get_loc(
                    start_time, method="bfill"
                ) : imglist_df.index.get_loc(end_time, method="ffill")
                + 1
            ]

    # TODO: write test for image_integrity_store
    def image_integrity_store(
        self, store, start_time=None, end_time=None, delta_seconds=0
    ):
        """ Checks which images are available on the permasense server
        
        Keyword Arguments:
            start_time {[type]} -- datetime object giving the lower bound of the time range which should be checked. 
                                   If None there is no lower bound. (default: {None})
            end_time {[type]} --   datetime object giving the upper bound of the time range which should be checked.
                                   If None there is no upper bound (default: {None})
            delta_seconds {int} -- Determines the 'duration' of an image in the output dataframe.
                                   start_time  = image_time+delta_seconds
                                   end_time    = image_time-delta_seconds (default: {0})
        
        Returns:
            DataFrame -- Returns a pandas dataframe with a list containing filename relative to self.base_directory, 
                         start_time and end_time start_time and end_time can vary depending on the delta_seconds parameter
        """
        """ Checks which images are available on the permasense server

        Arguments:
            start_time:   
            end_time:   
            delta_seconds:  Determines the 'duration' of an image in the output dataframe.
                            start_time  = image_time+delta_seconds
                            end_time    = image_time-delta_seconds
        Returns:
            DataFrame -- 
        """
        if start_time is None:
            # a random year which is before permasense installation started
            start_time = dt.datetime(1900, 1, 1)
        if end_time is None:
            end_time = dt.datetime.utcnow()

        tbeg_days = start_time.replace(hour=0, minute=0, second=0)
        tend_days = end_time.replace(hour=23, minute=59, second=59)

        delta_t = dt.timedelta(seconds=delta_seconds)
        num_filename_errors = 0
        images_list = []

        for key in store.keys():
            try:
                pathkey = Path(key)
                datekey = pathkey.parent.name
                dir_date = pd.to_datetime(str(datekey), format="%Y-%m-%d")
            except:
                # we do not care for files not matching our format
                continue

            if pd.isnull(dir_date):
                continue

            # limit the search to the explicit time range
            if dir_date < tbeg_days or dir_date > tend_days:
                continue

            # print(file.stem)
            start_time_str = pathkey.stem
            try:
                _start_time = pd.to_datetime(start_time_str, format="%Y%m%d_%H%M%S")
                if start_time <= _start_time and _start_time <= end_time:
                    images_list.append(
                        {
                            "filename": str(key),
                            "start_time": _start_time - delta_t,
                            "end_time": _start_time + delta_t,
                        }
                    )
            except ValueError:
                # try old naming convention
                try:
                    start_time = pd.to_datetime(start_time_str, format="%Y%m%d_%H%M%S")

                    if start_time <= _start_time and _start_time <= end_time:
                        images_list.append(
                            {
                                "filename": str(img_file.relative_to(base_directory)),
                                "start_time": _start_time - delta_t,
                                "end_time": _start_time + delta_t,
                            }
                        )
                except ValueError:
                    num_filename_errors += 1
                    warnings.warn(
                        "Permasense data integrity, the following is not a valid image filename and will be ignored: %s"
                        % img_file
                    )
                    continue

        segments = pd.DataFrame(images_list)
        segments.drop_duplicates(inplace=True, subset="start_time")
        segments.start_time = pd.to_datetime(segments.start_time, utc=True)
        segments.end_time = pd.to_datetime(segments.end_time, utc=True)
        segments.sort_values("start_time")

        return segments

    def image_integrity(
        self, base_directory, start_time=None, end_time=None, delta_seconds=0
    ):
        store = DirectoryStore(base_directory)
        return self.image_integrity_store(store, start_time, end_time, delta_seconds)

    def get_image_filename(self, timestamp):
        """ Checks wether an image exists for exactly the time of timestamp and returns its filename

            timestamp: datetime object for which the filename should be returned

        # Returns
            The filename if the file exists, None if there is no file
        """

        datadir = self.base_directory
        new_filename = (
            datadir
            + timestamp.strftime("%Y-%m-%d")
            + "/"
            + timestamp.strftime("%Y%m%d_%H%M%S")
            + ".JPG"
        )
        old_filename = (
            datadir
            + timestamp.strftime("%Y-%m-%d")
            + "/"
            + timestamp.strftime("%Y-%m-%d_%H%M%S")
            + ".JPG"
        )

        if os.path.isfile(new_filename):
            return new_filename
        elif os.path.isfile(old_filename):
            return old_filename
        else:
            return None

    def get_nearest_image_url(self, IMGparams, imgdate, floor=False):
        if floor:
            date_beg = imgdate - dt.timedelta(hours=4)
            date_end = imgdate
        else:
            date_beg = imgdate
            date_end = imgdate + dt.timedelta(hours=4)

        vs = []
        # predefine vs list
        field = []
        # predefine field list
        c_vs = []
        c_field = []
        c_join = []
        c_min = []
        c_max = []

        vs = vs + ["matterhorn_binary__mapped"]
        field = field + ["ALL"]
        # select only data from one sensor (position 3)
        c_vs = c_vs + ["matterhorn_binary__mapped"]
        c_field = c_field + ["position"]
        c_join = c_join + ["and"]
        c_min = c_min + ["18"]
        c_max = c_max + ["20"]

        c_vs = c_vs + ["matterhorn_binary__mapped"]
        c_field = c_field + ["file_complete"]
        c_join = c_join + ["and"]
        c_min = c_min + ["0"]
        c_max = c_max + ["1"]

        # create url which retrieves the csv data file
        url = "http://data.permasense.ch/multidata?"
        url = url + "time_format=" + "iso"
        url = url + "&from=" + date_beg.strftime("%d/%m/%Y+%H:%M:%S")
        url = url + "&to=" + date_end.strftime("%d/%m/%Y+%H:%M:%S")
        for i in range(0, len(vs), 1):
            url = url + "&vs[%d]=%s" % (i, vs[i])
            url = url + "&field[%d]=%s" % (i, field[i])

        for i in range(0, len(c_vs), 1):
            url = url + "&c_vs[%d]=%s" % (i, c_vs[i])
            url = url + "&c_field[%d]=%s" % (i, c_field[i])
            url = url + "&c_join[%d]=%s" % (i, c_join[i])
            url = url + "&c_min[%d]=%s" % (i, c_min[i])
            url = url + "&c_max[%d]=%s" % (i, c_max[i])

        url = url + "&timeline=%s" % ("generation_time")

        # print(url)
        d = pd.read_csv(url, skiprows=2)

        # print(d)

        # print(type(d['#data'].values))
        d["data"] = [s.replace("&amp;", "&") for s in d["data"].values]

        d.sort_values(by="generation_time")
        d["generation_time"] = pd.to_datetime(d["generation_time"], utc=True)

        if floor:
            data_str = d["data"].iloc[0]
            data_filename = d["relative_file"].iloc[0]
            # print(d['generation_time'].iloc[0])
            img_timestamp = d["generation_time"].iloc[0]
        else:
            data_str = d["data"].iloc[-1]
            data_filename = d["relative_file"].iloc[-1]
            # print(d['generation_time'].iloc[-1])
            img_timestamp = d["generation_time"].iloc[-1]

        file_extension = data_filename[-3:]
        base_url = "http://data.permasense.ch"
        # print(base_url + data_str)

        return base_url + data_str, img_timestamp, file_extension


class MHDSLRImages(MHDSLRFilenames):
    def __init__(
        self,
        base_directory=None,
        store=None,
        method="directory",
        output_format="xarray",
        start_time=None,
        end_time=None,
    ):
        super().__init__(
            base_directory=base_directory,
            store=store,
            method=method,
            start_time=start_time,
            end_time=end_time,
        )
        self.config["output_format"] = output_format

    def forward(self, data=None, request=None):
        filenames = super().forward(request=request)

        if request["output_format"] is "xarray":
            return self.construct_xarray(filenames)
        elif request["output_format"] is "base64":
            return self.construct_base64(filenames)
        else:
            output_formats = ["xarray", "base64"]
            raise RuntimeError(
                f"The {request['output_format']} output_format is not supported. Allowed formats are {output_formats}"
            )