Simple-Track

Simple-Track is a data-agnostic, threshold-based feature tracking algorithm for 2D data.

Features are tracked between consecutive frames of data by projecting feature fields onto common timeframes and matching between them based on the degree of overlap. Matched features retain the same identification between all tracked fields, while new features are assigned a unique label. Simple-Track compiles comprehensive information about feature merging, splitting, accretion, initiation and dissipation with an easy to use interface.

Installation

Simple-Track can be installed using PyPi or conda-forge:

python3 -m pip install simple-track

conda install conda-forge::simple-track

Coming soon to uv

Alternatively, Simple-Track can be run on meteorological data in CSET using the feature.track operator. See CSET docs for more information.

User Guide

This section describes the main methods of running Simple-Track. More details can be found in the docs

Input Requirements

While Simple-Track is designed to accept a wide range of input data, certain requirements must be met for the tool to function as intended:

• The input data must be gridded and contain a consistent spatial domain and resolution between frames.

• The features of interest must be defined by a threshold value, and these features must translate as a result of a spatially consistent background flow.

• The time between frames should be sufficiently short such that features can be reasonably expected to persist between frames. This is not a strict requirement since the tool includes an artificial advection step that projects data onto a common time, but it is likely that longer time steps will lead to more errors in feature matching and therefore less accurate tracking statistics.

Running Simple-Track

Simple-Track can be run in two ways:

1. Running Simple-Track from the Command Line

• Simple-Track can be run from the command line with a config file as an additional argument:

   simpletrack my_config.yaml 
  

• The my_config.yaml file contains the parameters for running Simple-Track. The required parameters are shown below:

   INPUT:
   	path: /path_to_folder_containing_data/*.data
   	loader: /path_to_file_containing_function|function_name # See next section
   FEATURE:
   	threshold: 1 # Threshold used for defining a feature

• Other parameters, such as experiment_name, output_path and save_data, along with more technical options, can also be set in this config file. See All Simple-Track Parameters for a full list.

• A valid loader function is required for pre-processing input data before tracking. See Loading Data for more information.

• Any number of config files can be provided as additional arguments, Simple-Track will iterate over each one in turn.

2. Importing Simple-Track to a python file

• Simple-Track can be run by importing the Tracker class from the simpletrack module. A config can be input either using a path to a yaml file, or by passing a dict when instantiating the object:

   from simpletrack import Tracker
  
   my_config = {
   	INPUT: {
   		path: "/path_to_folder_containing_data/*.data",
   		loader: "/path_to_file_containing_function|function_name" # See next section
   	},
   	FEATURE: {
   		threshold: 1, # Threshold used for defining a feature
   	}
   }
  
   timeline = Tracker(my_config).run()
  
   # Alternatively, if these parameters are saved in a config file, the path to this config can also be set as input
   timeline = Tracker("./my_config.yaml").run()

• Other parameters, such as experiment_name, output_path and save_data, along with more technical options, can also be set in this config. See All Simple-Track Parameters for a full list.

• If loader is included as a config input, the specified function is used for pre-processing input data before tracking. Alternatively, valid pre-processed data may be passed to the Tracker.run() method, bypassing the use of a separate function, and eliminating the need for the INPUT config section. See Loading Data for more information.

• Tracker.run() returns a Timeline object which is used to store all tracking and feature data. This can be inspected and analysed beyond the outputs that are saved as part of standard operation.

Loading Data

Each Simple-Track input must contain two sets of data:

1. A datetime object specifying the time that the data is valid for
2. A numpy.array object containing the data to track

There are three methods of providing these data pairs to Simple-Track:

1. Loading through config options

• Simple-Track will load all data matching the structure given in "INPUT": "path" config section. This input supports wildcard matching (i.e., using "./path_to_data/*.data" would load all files with the .data suffix).

• Each file contains one input to Simple-Track (see here for more information)

• Since Simple-Track is a data-agnostic tool, there may be any number of bespoke tools for loading and pre-processing data before it is suitable for tracking. This functionality can be contained in a custom loader function that will perform these actions before passing the compatible data to the main processing workflow.

• An example of a custom loader function is shown below:

   def user_definable_load(self, filename):
   	import iris # Import any required libraries here
  
   	# Get 2D data from input file as a numpy array
   	cube = iris.load_cube(filename, "precipitation_flux")
   	data = cube.data
  
   	# Additional data pre-processing can be performed here too!
  
   	# Get time from input file, in datetime format
   	tcoord = cube.coord("time")
   	time = tcoord.units.num2pydate(tcoord.points)[0]
  
   	# Method must return a tuple of 
   	# (datetime.datetime, numpy.NDArray), where the 
   	# first element is the time the data is valid for
   	# and second element is the 2D array of data to be tracked
   	return time, data

• This loader function is then specified in the "INPUT": "loader" config using the ./path_to_file.py|func_name format. So in this case, the config option would be ./path_to_file.py|user_definable_load.

• Loading via the config can be used whether Simple-Track is being run from the command line or from a python file.

2. Loading through the Command Line

• The same "INPUT" config sections mentioned above can also be input from the command line

   simpletrack my_config.yaml -i /path_to_folder/*.data -l ./path_to_file.py|func_name
  

• Each file contains one input to Simple-Track (see here for more information)

3. Passing a dict directly to Tracker.run()

• If SimpleTrack is being run from a python file and a suitable set of data has already been loaded, this data can be passed directly to Tracker.run() as a dict, with the datetime object as the key and a numpy.array object as the value. For example:

   import datetime as dt
   import numpy as np
   from simpletrack import Tracker
  
   time1 = dt.datetime(year=2000, month=1, day=1, hour=10, minute=5)
   time2 = time1 + dt.timedelta(minutes=5)
  
   data1 = np.array(...)
   data2 = np.array(...)
  
   st_input = {
   	time1: data1,
   	time2: data2,
   }
  
   my_config = {...}
  
   Tracker(my_config).run(st_input)

• Any number of time:data pairs can be passed to Tracker.run() and the code will iterate over the ordered dict.

• Passing data into Tracker.run() via this method will bypass any "INPUT":"loader" or "INPUT":"path" inputs specified in the corresponding config file.

Outputs

For each frame of data, Simple-Track compiles a set of fields and tracked feature properties. Each 2D field is of the same shape as the input fields, and contains an overview of feature properties across the space.

Fields (.field files):

• Feature field: 2D array of positive integers showing unique feature id present at each location (zero indicating no feature present)
• Lifetime field: 2D array of positive integers showing lifetime of the feature present at each location (zero indicating no feature present)
• x-flow, y-flow: 2D array of floats containing the x- and y-components of the motion vectors at each location that translate features from the previous frame to the current frame

Features (.csv or .txt files):

• ID: Unique feature identifier that persists between frames (i.e., a feature retains the same id across all frames that it is tracked).
• centroid: (y, x) tuple containing central location of feature.
• size: Number of pixels spanned by the feature.
• dydx: (dy, dx) tuple containing motion vector that translated feature to its location in the current frame from the previous frame.
• max: Maximum value contained within the feature in the input data.
• lifetime: Number of timesteps the feature has existed for.
• accreted: List of IDs of features that were accreted by this feature, if applicable.
• parent: ID of parent feature that this feature split from, if applicable
• children: List of IDs of features that split from this feature, if applicable

It it also possible to perform further analysis of tracking statistics using the data structures and tools of Simple-Track. This can be done using the Timeline object returned by Tracker.run(), which contains Frame and Feature data and built-in methods for easily accessing relevant data.

Alternatively, the data that is output by Simple-Track can be read back in to a Timeline object using the LoadOutput class in frame_output.py. This object only requires a path to the stored Simple-Track data. The LoadOutput.load_to_timeline() method will return a Timeline object containing all of the loaded data in the same data structures that Simple-Track stores its data. See User Guide in docs for more details.

All Simple-Track Parameters

A complete list of parameters and their default values are given below. For a more thorough explanation of each parameter, refer to the docs.

INPUT:
  path: ./path_to_input_data/*.data
  loader: /path_to_file_containing_function|function_name
  iterate_over_array: False # Whether to iterate a single array or multiple files
  iterating_dim: 0 # If iterate_over_array flag is enabled, this sets the dimension to iterate over

OUTPUT:
  path: ./output
  experiment_name: Simple-Track Experiment # Name of experiment to add to output files 
  save_data: true # Whether to save data to output
  skip_tracking: false # Whether to skip tracking and just output feature properties
  output_raw_data: true # Whether to save a copy of the input data to the simple-track output folder. This is then used when re-loading data into a Timeline using the LoadOutput object. This ensures loaded Timelines are functionally identical to freshly generated timelines. If set to False, LoadOutput will only load data generated by Simple-Track. To make these timelines functionally identical to generated ones, the raw data will then need to be loaded separately into each Frame. 

FEATURE:
  threshold: 1 # Threshold used for defining a feature
  under_threshold: false # Whether features are defined above or below the threshold
  min_size: 4 # Minimum size of feature to be tracked (in pixels)

FLOW_SOLVER:
  overlap_threshold: 0.3 # Minimum fraction of overlap between features for use in flow_solver
  subdomain_size: 100 # Size in pixels of individual squares to run fft for (dy, dx) displacement. Must divide (y,x) lengths of the array. Defaults to domain size / 5
  min_fractional_coverage: 0.01  # Minimum fractional cover of objects required for fft to obtain (dy, dx) displacement
  subdomain_tolerance: 3.0  # Maximum difference in displacement values between adjacent squares (to remove spurious values)
  apply_tukey_filtering: True # Apply a 2D Tukey window to each subdomain before phase cross-correlation

TRACKING:
  overlap_nbhood: 5 # Radius of halo in pixels for orphan storms - big halo assumes storms may spawn "children" at a distance multiple pixels away
  overlap_threshold: 0.3 # Minimum fraction of overlap 
  retain_lifetime_on_split: True # If a child Feature splits from its parent feature, this determines whether the child Feature should carry over the lifetime from the parent or whether its lifetime should be set to 1