Skip to content

Temporal Raster Plot

A nested-time grid places observations by interval and position within that interval, encoding each cell's value to compare recurring temporal structure.

Version
v1 · 2026-10-04 · History
Domain-specific #
13776
Domain group
Formal Sciences
Origin domain
Experimental Design & Statistics
Subdomain
Data Visualization → Experimental Design & Statistics
Aliases
Temporal carpet plot

Core Idea

A temporal raster plot reorganizes a time series into a grid with two temporal coordinates: one identifies a repeated outer interval, and the other identifies position within that interval. A cell then encodes the observation or summary at that time by color or intensity. Days can form columns and times of day rows; water years can form rows and days or months within each year columns. This nested-time organization puts corresponding phases of successive intervals next to one another for comparison.[1][2]

The plot is a visual transformation, not a statistical proof of periodicity. Repeated operations may form stripes, seasonal shifts may form gradients, and isolated events may appear as patches, but the appearance depends on the chosen interval, alignment, cell aggregation, color scale and missing data. A calendar day is a convenient slice even if the underlying process is not strictly daily periodic. Repeating the view at another interval can test whether a pattern persists or was an artifact of the fold.[3]

The word raster here means a grid of visual cells, not a satellite or geographic raster data cube. The distinctive feature is that both axes are time, at different nested resolutions. A generic heat map with one spatial and one temporal axis lacks that structure, and a run chart keeps time on one linear axis.

Structural Signature

  1. Time-indexed input: measurements or events span multiple intervals.
  2. Outer-interval rule: a calendar day, week, water year or other repeatable reporting unit partitions time.
  3. Within-interval coordinate: each observation is placed by hour, day-of-year or an analogous position in its interval.[1][2]
  4. Cell mapping: a value, count or summary is encoded as color or intensity.
  5. Alignment and binning: observations may require aggregation, interpolation or empty cells; interval length can vary.
  6. Comparison task: aligned cells and bands make cross-interval similarities, shifts and exceptions inspectable, subject to scale and data-quality checks.

Condensed: time series + repeated interval + within-interval position + cell value encoding = temporal raster plot.

Sig role-phrases: time-indexed observations; repeated outer interval; within-interval temporal position; value-to-cell encoding; alignment and missingness policy.

What It Is Not

  • Not a generic heat map. Both axes represent nested time coordinates rather than arbitrary categories or geography.
  • Not a run chart. A run chart preserves a single continuous chronological axis; the raster folds it into intervals.
  • Not proof that the data are periodic. Any time series can be partitioned by a calendar unit; whether the underlying phenomenon repeats is a separate empirical question.
  • Not restricted to perfectly regular sampling. Irregular observations can be binned, but the aggregation rule and missing-cell marks must be explicit.
  • Not automatically anomaly detection. A bright isolated cell may be a true event, sensor error, changed scale or missing-data artifact.
  • Not a geographic raster time series. That term describes a sequence of spatial images, not necessarily a two-time-axis plot.
  • Not always “carpet plot” without qualification. That name is used for other chart types; temporal carpet plot is the safer alias.

Scope of Application

In building energy analysis, day-by-time-of-day is a plausible temporal-raster arrangement for measured or simulated loads. The accessible LBNL/OSTI abstract verifies only that SEE IT generates time-series, scatter and carpet plots independently of whether data are measured or simulated, to compare building performance. It does not disclose its carpet-plot axes or case-study observations. The day/hour illustration below is therefore author-constructed and cannot be attributed as the exact SEE IT implementation.[1]

In hydrology, the USGS raster seasonality plot places water year against day within that year and colors daily mean streamflow. The USGS's Minnesota River at Mankato example, streamgage 05325000, is Figure 16 in its report on water years 1921–2020. The report describes higher flow as blue, middle flow as green, lower flow as tan, and no data as white. Its text interprets a shift: before 1960 larger flows are generally April to mid-July; afterward they extend from mid-March to about September. A viewer can therefore distinguish a long-run change in where high-color cells occur, rather than merely saying “there is a pattern.” The source's interpretation is an observed visual contrast, not proof that the plot alone established its cause.[2]

In general visualization, TimeElide research concerns non-contiguous time slices, not necessarily periodic day/year folds. Its Figure 1 contrasts 2D heatmaps using absolute versus normalized within-slice timing. This is a source-backed neighboring design problem: normalizing can make unequal-length slices occupy the same visual width but can hide duration differences. It supports the need to declare an alignment convention, not a claim that every TimeElide view is a temporal raster plot in this narrower sense.[3]

Clarity

Take hourly electricity use for a year. Give each day one column and each hour a row; color each cell by usage. A weekday schedule might appear as repeated daily bands, while holidays may differ from neighboring days. If the hour changes under daylight saving time, however, a local-time day can have 23 or 25 hours. The plot must either preserve that irregularity, map to UTC, or define a binning rule. Otherwise an apparent stripe break can be a clock convention rather than an operational change.

For a stream gauge, place years as rows and days of the water year as columns. High-flow seasons may appear at similar horizontal positions; a flood could be a saturated patch. But the color palette may compress high values, and empty measurements must not be colored as zero flow. The grid helps identify where to investigate, not what caused the event.[2]

Manages Complexity

A long one-dimensional series can overwhelm a line chart. Folding it into comparable intervals compresses many observations into a compact field while preserving a coarse sense of phase and long-run progression. It makes within-period regularity and across-period change visible at once. The price is loss of exact timestamp readability, adjacency across interval boundaries, and potentially value precision. An analyst should pair the overview with scale, missingness and drill-down to the original series.

Abstract Reasoning

Choose the interval based on the question—day for diurnal behavior, water year for within-year flow—not just because the software offers it. Define how each observation maps to an outer interval and within-interval coordinate. Specify bin aggregation, time zone, variable-length intervals, leap days and missingness. Set a color range that makes comparisons honest across the full chart. Inspect apparent patterns, then test them against the raw series and alternative folds or scales.[3][2]

The diagnostic question is: Is this pattern in the measurements, or was it created by the interval, alignment, binning or color mapping?

Knowledge Transfer

The general idea is to arrange repeated temporal slices for comparison. Its visualization identity is specific: the two axes are nested time coordinates and cells encode observed values. Analogies to a geographic raster or arbitrary matrix visualization lose that temporal structure.

Examples

Constructed daily building load

Consider an author-constructed four-cell load series: Monday 09:00 = 10 kW, Monday 10:00 = 14 kW, Tuesday 09:00 = 12 kW, Tuesday 10:00 = 8 kW. Place day in columns, hour in rows, and use one fixed color scale for kW. The 09:00 row then compares 10 with 12 across days; the 10:00 row compares 14 with 8. If a single unchanging scale is used, Tuesday's 10:00 cell is lower than Monday's despite Tuesday's higher 09:00 cell. No measured building behavior is inferred from these invented values. The LBNL abstract establishes carpet-plot use for measured/simulated comparison but not these axes or numbers.[1]

Mapped back: four time-indexed observations, day as outer interval, hour as within-day coordinate, kW as value color, and row-wise cross-day comparison; all bins and missingness are explicit in this toy 2×2 grid.

Minnesota River flow, USGS Figure 16

USGS Figure 16 displays the Minnesota River at Mankato, streamgage 05325000, over water years 1921–2020. Read across a pre-1960 row: the reported larger-flow blue region is generally concentrated from April to mid-July. Read across a post-1960 row: larger-flow colors extend from mid-March to about September. Read downward through years: the report notes greener and bluer cells by the late 2000s and fewer tan low-flow cells. White denotes missing daily mean flow, not zero. These are report-described contrasts; no single colored cell is given an exact cfs value in the accessible text.[2]

Mapped back: daily mean flow is input, water year is outer row, day/month of year is horizontal phase, tan/green/blue encode low/middle/high reported flows, white is no data, and the before/after-1960 location of high-color cells is the cross-interval comparison. The source's hydrologic interpretation remains separate from the visual encoding itself.

TimeElide absolute/normalized boundary case

TimeElide Figure 1(f–g) puts the same non-contiguous slices into two 2D heatmap variants, one with absolute and one with normalized within-slice timing. A feature at the midpoint of a short slice can line up visually with a feature at the midpoint of a long slice under normalization, although their elapsed durations differ. This is a design-space comparison, not a claim that the specific figure contains a measured event at those positions. It tests what the within-slice coordinate means.[3]

Mapped back: outer slices and within-slice position exist, but the choice of absolute versus normalized coordinate changes what aligned cells can legitimately be compared; periodic repetition is not guaranteed, so TimeElide is a neighbor rather than automatically this named plot.

One-axis line chart near miss

The same hourly load shown as a single year-long line preserves exact chronological adjacency but does not align 9 a.m. across successive days in a grid.

Mapped back: time-indexed input and value encoding are present, but the repeated outer interval and aligned within-interval coordinate are absent.

Structural Tensions

Compact pattern versus exact chronology. Folding many years into rows exposes the Minnesota plot's shift in high-flow season at a glance; the same folding separates September 30 from October 1 onto opposite ends of adjacent rows. A line chart preserves that temporal adjacency but makes a century of seasonal comparisons harder to scan. A flood crossing the seam could appear as two patches unless exact dates are recovered from the original series. Diagnostic: does an apparent pair of edge events become one continuous event when unfolded?[2]

Alignment versus duration fidelity. Normalized within-slice time aligns phases of unequal slices, easing visual comparison; it can make intervals of different actual duration appear equivalent. Absolute time preserves elapsed duration, but corresponding phases of unequal slices may no longer line up. TimeElide Figure 1(f–g) directly exhibits the competing encodings. Diagnostic: would the conclusion change if the same slices were viewed with absolute rather than normalized within-slice time?[3]

Visual contrast versus faithful missingness. A strong palette makes high- and low-flow seasons legible across a century; coarse color bands lose exact cfs precision and can visually overstate a borderline change. Treating white no-data cells as tan low flow would falsely extend a dry period, while reserving a distinct missing state reduces available visual contrast for measured values. The USGS figure's white no-data convention is therefore an interpretive safeguard, not decoration. Diagnostic: do inferred low-flow regions persist after excluding no-data cells and checking the original daily values?[2]

Structural–Framed Character

The temporal raster plot sits toward the structural side of a mixed structural–framed spectrum. Its invariant is the mapping \(t\mapsto(\text{outer interval},\text{position within interval})\), followed by value-to-cell encoding. No ethical approval or institutional authority is built into that mapping. Evaluative weight is lower than in a moral rule, but not zero: “useful pattern” and an honest color scale depend on the analyst's question and display choices. Human practice enters through calendar conventions, water-year definitions, time zones, binning, missing-data codes and viewers' visual interpretation; the cells do not decide whether a hydrologic shift is causal.

There is no single institutional origin established for every temporal raster. The checked records show an LBNL building-energy carpet-plot tool, a USGS hydrologic raster seasonality plot and UBC visualization research on time slices. The name and visual vocabulary travel across these settings, while the time axis, measurement and reading task change. A scientist may import a familiar “carpet plot” technique into hydrology, but an independently designed year-by-day colored grid can be recognized as the same structure without borrowing that label. TimeElide's non-contiguous slices show where a superficially similar heatmap may not meet the narrower repeated-interval identity. Its character: a relatively structural visual encoding with practice-bound alignment and interpretation, not a claim of periodicity or causation.[1][2][3]

Structural Core vs. Domain Accent

The portable skeleton is a sequence factored into repeated outer intervals and aligned inner positions, with values encoded in corresponding cells. Hydrology uses water years and daily flow; a building example can use days and hours of load. The domain mechanism is visualization of time-indexed measurements with a declared interval, alignment, units, palette and missingness policy. Remove the nested temporal coordinates and an ordinary heatmap remains; remove the encoded observations and one has only a calendar grid.

The named plot fails the prime bar because its identity remains a specific visual representation and its validity depends on chart conventions. Run Chart preserves one chronological axis and is a sibling/contrast, not a strict parent. Visualization (Graphics) is the strict genus: this plot is a particular graphic encoding of time-indexed values. A broad Heat Map node is not asserted as a further parent. A future substrate-neutral prime for “fold and align repeated slices” would need non-visual transfer evidence, which these display sources do not provide.

This entry is a kind of Visualization (graphics).

  • Periodization: time is divided into repeated reporting intervals.
  • Alignment: within-interval positions are made comparable across intervals.
  • Encoding: color or intensity stands for measured value.

Visualization (Graphics) is the strict genus. Periodization, Alignment and Encoding remain conceptual links, not additional parent edges.

Relationships to Other Abstractions

Local relationship map for Temporal Raster PlotParents appear above the current abstraction, mutual partners to the right, and children below. Node labels state whether each abstraction is prime or domain-specific; colors identify relation types.Temporal Raster PlotDOMAINDomain-specific abstraction: Visualization (graphics) — is a kind ofVisualization(graphics)DOMAIN

Current abstraction Temporal Raster Plot Domain-specific

Parents (1) — more general patterns this builds on

  • Temporal Raster Plot is a kind of Visualization (graphics) Domain-specific

    A temporal raster plot specializes visualization by mapping repeated time intervals and within-interval positions to value-coded cells.

Hierarchy path (1) — routes to 1 parentless root

Neighborhood in Abstraction Space

Temporal Raster Plot sits in a sparse region of the domain-specific corpus (98th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (2551 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-08

Not to Be Confused With

Run Chart uses one chronological axis. Heat Map is a broader visual encoding and need not have two time axes. Calendar Heat Map may arrange dates in calendar weeks but not always display a within-period phase as a second temporal coordinate. Raster Hydrograph is a hydrology-specific instance. Geographic Raster Time Series is a sequence of spatial grids and a different data object.[2]

References

[1] Maile et al., “A Software Tool to Compare Measured and Simulated Building Energy Performance Data,” LBNL/OSTI (2011). Accessible official abstract verifies carpet-plot use, not axis details or case results. registry ↩a ↩b ↩c ↩d ↩e

[2] Tara Williams-Sether and Chris Sanocki (2025), U.S. Geological Survey, “Peak streamflow trends in Minnesota and their relation to changes in climate, water years 1921–2020,” Figure 16 and “Raster Seasonality Plots”, DOI 10.3133/sir20235064E, streamgage 05325000, color/no-data legend and described pre/post-1960 seasonal contrast. Publisher-indexed text describes the figure; this entry makes no claim about individual cell values. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j

[3] Oppermann, Liu and Munzner, “TimeElide: Visual Analysis of Non-Contiguous Time Series Slices” (IEEE VIS 2021), original visualization research and absolute/normalized within-slice heatmap variants. registry ↩a ↩b ↩c ↩d ↩e ↩f