Introduction to MazamaTimeSeries

Mazama Science

2024-03-08

Background

This package supports data management activities associated with environmental time series collected at fixed locations in space. The motivating fields include both air and water quality monitoring where fixed sensors report at regular time intervals.

Data Model

The most compact format for time series data collected at fixed locations is a list including two tables. MazamaTimeSeries stores time series measurements in a data table where each row is a synoptic record containing all measurements associated with a particular UTC time stamp and each column contains data measured by a single sensor (aka “device”). Any time invariant metadata associated with a sensor at a known location (aka a “device-deployment”) is stored in a separate meta table. A unique deviceDeploymentID connects the two tables. In the language of relational databases, this “normalizes” the database and can greatly reduce the disk space and memory needed to store and work with the data.

Single Time Series

Time series data from a single environmental sensor typically consists of multiple parameters measured at successive times. This data is stored in an R list containing two dataframes. The package refers to this structure as an sts object for SingleTimeSeries:

sts$meta – 1 row = unique device-deployment; cols = device/location metadata

sts$data – rows = UTC times; cols = measured parameters (plus an additional datetime column)

sts objects can support the following types of time series data:

Raw, “engineering data” containing uncalibrated measurements, instrument voltages and QC flags may be stored in this format. This format is also appropriate for processed and QC’ed data whenever multiple parameters are measured by a single device.

Note: The sts object time axis specified in data$datetime reflects device measurement times and is not required to have uniform spacing. (It may be regular but it need not be.) It is guaranteed to be monotonically increasing.

Multiple Time Series

Working with timeseries data from multiple sensors at once is often challenging because of the amount of memory required to store all the data from each sensor. However, a common situation is to have time series that share a common time axis – e.g. hourly measurements. In this case, it is possible to create single-parameter data dataframes that contain all data for all sensors for a single parameter of interest. In air quality applications, common parameters of interest include PM2.5 and Ozone.

Multi-sensor, single-parameter time series data is stored in an R list with two dataframes. The package refers to this structure as an mts object for MultipleTimeSeries:

mts$meta – N rows = unique device-deployments; cols = device/location metadata

mts$data – rows = UTC times; N cols = device-deployments (plus an additional datetime column)

A key feature of mts objects is the use of the deviceDeploymentID as a “foreign key” that allows sensor data columns to be mapped onto the associated spatial and sensor metadata in a meta row. The following will always be true:

identical(names(mts$data), c('datetime', mts$meta$deviceDeploymentID))

mts objects can support the following types of time series data:

Each column of mts$data represents a timeseries associated with a particular device-deployment while each row represents a synoptic snap shot of all measurements made at a particular time.

In this manner, software can create both timeseries plots and maps from a single mts object in memory.

Note: The mts object time axis specified in data$datetime is guaranteed to be a regularly spaced, monotonic axis with no gaps.

Example Usage

See usage examples in the function documentation.


Best wishes for efficient and productive analysis of time series data!


This R package was created with funding from the USFS AirFire Research Team.