hydrotools.events 1.1.5

Various methods to support event-based evaluations.

OWPHydroTools :: Events

This subpackage implements methods to partition hydrometric times series into discrete "events." See the Events Documentation for a complete list and description of the currently available methods. To report bugs or request new methods, submit an issue through the OWPHydroTools Issue Tracker on GitHub.

Installation

In accordance with the python community, we support and advise the usage of virtual environments in any workflow using python. In the following installation guide, we use python's built-in venv module to create a virtual environment in which the tool will be installed. Note this is just personal preference, any python virtual environment manager should work just fine (conda, pipenv, etc. ).

# Create and activate python environment, requires python >= 3.8
$ python3 -m venv venv
$ source venv/bin/activate
$ python3 -m pip install --upgrade pip

# Install events
$ python3 -m pip install hydrotools.events

Usage

The following example demonstrates how one might use hydrotools.events.event_detection to extract hydrological events from a time series of real streamflow measurements. This example requires the hydrotools.nwis_client package.

Code

# Import tools to retrieve data and detect events
from hydrotools.nwis_client.iv import IVDataService
from hydrotools.events.event_detection import decomposition as ev

# Use pandas to resample the data
from pandas import Grouper

# Use a helper function to detect events at multiple sites
def list_events_helper(mi_series, level, halflife, window):
    """Reduce multi-index series before applying event detection."""
    return ev.list_events(mi_series.droplevel(level), halflife, window)

# Retrieve streamflow observations for two sites
service = IVDataService()
observations = service.get(
    sites='02146750,0214676115,09489000', 
    startDT='2019-10-01', 
    endDT='2020-09-30'
    )

# Drop extra columns to be more efficient
observations = observations[[
    'usgs_site_code', 
    'value_date', 
    'value'
    ]]

# Check for duplicate time series, keep first by default
observations = observations.drop_duplicates(
    subset=['usgs_site_code', 'value_date']
    )

# Resample to hourly, keep first measurement in each 1-hour bin
observations = observations.groupby([
    'usgs_site_code',
    Grouper(key='value_date', freq='H')
    ]).first().ffill()

# Detect events
events = observations['value'].groupby(
    level='usgs_site_code').apply(
        list_events_helper, 
        level='usgs_site_code', 
        halflife='6H', 
        window='7D'
    )

# Print event list    
print(events)

Output

                                start                 end
usgs_site_code                                           
02146750       0  2019-10-13 22:00:00 2019-10-15 04:00:00
               1  2019-10-16 09:00:00 2019-10-18 02:00:00
               2  2019-10-19 03:00:00 2019-10-23 19:00:00
               3  2019-10-27 05:00:00 2019-10-28 18:00:00
               4  2019-10-30 01:00:00 2019-11-02 18:00:00
...                               ...                 ...
09489000       34 2020-09-16 00:00:00 2020-09-16 19:00:00
               35 2020-09-17 00:00:00 2020-09-17 15:00:00
               36 2020-09-18 04:00:00 2020-09-18 10:00:00
               37 2020-09-27 01:00:00 2020-09-27 03:00:00
               38 2020-09-28 05:00:00 2020-09-28 11:00:00

Event Detection Tips

The evaluation_tools.events.event_detection.decomposition method has two main parameters: halflife and window. These parameters are passed directly to underlying filters used to remove noise and model the trend in a streamflow time series (also called baseflow). Significant contiguous deviations from this trend are flagged as "events". This method was originally conceived to detect rainfall-driven runoff events in small watersheds from records of volumetric discharge or total runoff. Before using decomposition you will want to have some idea of the event timescales you hope to detect.

In general you'll want to consider the following:

1. Ensure your time series is monotonically increasing with a frequency significantly less than the frequency of events and fill missing data
2. Use a pandas.Timedelta compatible str to specify halflife and window. See the pandas docs
3. Specify a halflife larger than the expected frequency of noise, but smaller than the event timescale
4. Specify a window larger than the event timescale, but at least 4 to 5 times smaller than the entire length of the time series
5. Filter the final list of events to remove false positive events, particularly in noisy signals. You could filter on peak event flow, event duration, or other event characteristics