GISCI 241 Chap.10 Time Series and Disturbance Signals
Time Series and Disturbance Signals
A time series uses successive observations to reveal baseline behaviour, seasonal cycles, long-term trends, abrupt disturbances and recovery. This chapter shows why two endpoints can conceal event timing and trajectory. It also explains how cloud, missing observations, classification flicker and cross-season comparison create false temporal signals.
Dense temporal segmentation can identify breaks and fitted trends, but those segments remain representations of spectral or index behaviour that require spatial context and independent evidence for environmental attribution.
What this chapter covers
- 01
Successive observations and temporal baselines
- 02
Seasonality, phenology and comparable timing
- 03
Gradual trends and abrupt event-scale changes
- 04
Recovery trajectories and endpoint compression
- 05
Data gaps, cloud contamination and classification flicker
- 06
Per-pixel temporal segmentation and validation
Worked example · free
Read a disturbance and recovery sequence
- +1The first four observations form a stable baseline between 0.63 and 0.66.
- +1The fifth observation drops from 0.65 to 0.31, an abrupt change of negative 0.34.
- +1Later values rise steadily, showing partial recovery rather than a second stable low state.
- +1Last minus first equals 0.47 minus 0.64, giving negative 0.17.
- +1The endpoint difference understates the disturbance and omits timing and recovery.
Key terms
- Time series
- A sequence of observations of the same quantity or location across successive times.
- Baseline
- The characteristic pre-event level and variability against which later observations are compared.
- Seasonality
- A repeating temporal pattern associated with annual or other regular environmental cycles.
- Temporal trend
- A sustained directional change in a measured quantity across time.
- Abrupt disturbance
- A rapid departure from prior behaviour, such as a sharp spectral-index break.
- Temporal segmentation
- Representing a series as connected intervals with fitted trends and breakpoints.
- Persistence
- The continuation of a detected change across later valid observations rather than one isolated anomaly.
Time Series and Disturbance Signals FAQ
Why is a time series better than two dates?
Additional observations reveal baseline variation, event timing, seasonality, trend and recovery. Two endpoints can miss a sharp event followed by partial recovery or mistake a seasonal difference for long-term change. The time series adds context, provided acquisition and preprocessing remain comparable.
How can seasonality cause false change?
Vegetation, water, snow and moisture vary predictably through the year. Comparing observations from different seasons can produce a large difference without unusual disturbance. Compare like seasons, model the cycle or use multiple years to establish expected timing and amplitude before calling the change persistent.
What does a temporal breakpoint prove?
A breakpoint indicates that a fitted representation of the spectral or index series changes at that time. It does not by itself identify fire, harvest, flood or another cause. Check data quality, neighbouring pixels, maps, external event records and persistence before attributing a process.
How precisely can a time series date a disturbance?
Timing precision is bounded by observation spacing and data quality. The event occurred after the last reliable pre-event view and before the first reliable post-event view; an annual composite rarely supplies an exact day. Mark that interval, show gaps and quality masks, and use external event records only when they genuinely narrow the window.
Exam move
Draw a time-series line and label baseline, seasonal cycle, gradual trend, abrupt break and recovery. Practise identifying which component dominates short synthetic sequences. Compare first-to-last difference with the largest single break so you see how endpoints compress history. For every anomaly, check masks, neighbouring dates and persistence before environmental interpretation.
When classification maps flicker, inspect class probabilities or direct spectral trajectories rather than counting every label change as real. In a project figure, show enough pre-event and post-event observations, mark data gaps and keep axes consistent. Close the interpretation with the measured temporal pattern, its likely explanation and an explicit alternative.
Create three short sequences with the same first and last values but different internal paths: a temporary dip, an abrupt loss followed by recovery, and a steady decline. Explain why endpoint subtraction cannot distinguish them. Add a repeating seasonal curve and compare like seasons before judging change.
For every candidate breakpoint, record the last reliable pre-event observation and first reliable post-event observation; their interval is the honest timing precision. Inspect masks, raw bands and neighbouring pixels when one low value appears alone. For segmentation, sketch fitted line pieces over the observations and state that the fit summarises spectral behaviour rather than naming a land-cover cause.
Compare direct index trajectories with annual classified maps to expose label flicker near decision boundaries. Build a three-panel project figure containing the time series, spatial footprint and quality evidence. Keep axes, units and gaps explicit. Finish the caption with the observed break or trend, the degree of persistence, the leading explanation and the independent event or field evidence required for attribution.