Lecture slides for GIS/MEA582

Dynamic spatial phenomena

Multitemporal Data Handling in TGRASS

Helena Mitasova, Anna Petrasova, Vaclav Petras

GIS714 Geosimulations NCSU

Learning objectives

observation time series and multitemporal data
dynamic simulations
managing geospatial multitemporal data
dynamic visualization

Observation time series and multitemporal data

Definitions:

systematic, consistent monitoring: time series
data collected over the same location but for different purpose and/or using different technologies: multitemporal data

Observation time series and multitemporal data

Objectives:

gain understanding of dynamics / change / evolution of studied phenomenon or system
provide input for diagnostic or predictive modeling

Your observation data

Describe your time series or multitemporal data:

source:
in situ point measurements, satellite, airborne/UAS imagery, polygon-based data ...
spatial and temporal distribution, resolution, scale:
image(s) at regular time step, event-based acquisition, static points, moving objects,...
management of your data:
spreadsheet, spatio-temporal database
processing, analysis and visualization

Monitoring dynamic processes

How fast do your variables change?
Do your data capture the full dynamics or snapshots?
Is the process multiscale?
What temporal scales are captured by your data?
What is driving the studied process dynamics?
What data you have to capture the driving forces?

Discuss temporal properties of your data and answer these questions in your project proposal

Dynamic models from monitoring data:
time series and multitemporal examples

Observation time series

Annual dynamics of 30yr monthly precipitation in South America
$p_i = f_i(x,y), i=1, ... 12$
bi-variate interpolation from 130 point data

aggregated over each month, averaged over 30 years

Observation time series

Annual dynamics of monthly precipitation in North Carolina for the years 2004,2005,2006

aggregated over each month, subset from 2000-2012 time series used in the assignemnt

Model and Observation time series

Real-time atmospheric and ocean data visualization, combines observations and models earth

Beccario, C: Visualization of global weather conditions forecast by supercomputers updated every three hours

Observation multitemporal data

Multitemporal point clouds with variable point distribution, densities and accuracies

Data sources: digitized contours, photogrammetry mass points, lidar, GPS

Observation multitemporal data

Jockey's Ridge 1974 - 2017:
40 years of southward migration and landform change
$z_i = f_i(x,y), i=1, ... n$:
bi-variate interpolation of DEMs from point cloud snapshots

Note that the visualization does not respect the variable time interval

Multitemporal DEM 1974 - 2017

Multitemporal Planet imagery

Satellite imagery (3m resolution), Sep. 2017 - July 2018

Planet: constellation of Earth-imaging micro satellites providing daily observations for entire Earth at 1m resolution

Observation multivariate time series

Chesapeake bay monthly N concentration measurements in 1993 - 3D distribution of sampling sites
$w = f(x,y,z,t)$: multivariate (3D space-time) interpolation reveals seasonal trends, influx of N in spring and decrease in summer

Observation multitemporal data

Sampling well sites: scattered in space and time

Observation multitemporal data

groundwater pollution over 10 years $w = f(x,y,z,t)$, space-time interpolation of chemical concentrations data
using scattered space-time point data to create a continuous space-time model of plume evolution

Dynamic models from simulations

Results of numerical modeling
Well defined time step: depends on application, input data and numerical method
Challenge: coupled models of processes with different time steps

Solar radiation

Simulation of solar radiation during one day (summer solstice) with 30 minute time step at 1m spatial resolution

Ultra-high resolution water flow

Time step depends on flow velocity and resolution
UAS-based surveys: cm-resolution DEMs
Captures impact of microtopography, time step is fraction of a second

Jeziorska et al., 2016, Overalnd flow analysis using time series of UAS-derived elevation models

Fire spread

Simulation of fire spread at 1m spatial resolution with time step dependent on wind speed, generally in minutes

Managing time series and multitemporal data

Temporal data framework in GRASS GIS

designed for massive series of satellite imagery or modeling outputs
efficient processing, management and analysis of space-time data sets
space-time dataset is a set of maps registered in a temporal database
individual maps with assigned time-stamp represent the state of a dynamic variable at a given time

Gebbert S., Pebesma E. 2014. A temporal GIS for field based environmental modeling. EMS 53, 1-12.

Gebbert S., Pebesma, E. 2017. The GRASS GIS temporal framework. IJGIS, pp. 1273-1292

Veronica Andreo, 2021, Analyzing space-time satellite data for ecological applications with GRASS and R (presentation)

Time stamp type

Time stamp assigns time to an individual map in the space-time data set

time instant - snapshot at a given time expressed as absolute time:
2013-10-15 13:00:00 (date time format)
time interval - defined by start and end time: time period of a day, a month, or a year
example: a single UAS survey represents a snapshot (e.g. elevation at the time of survey) which can be agreggated into time intervals (e.g. monthly average elevation)
relative time expressed as value (e.g. years 1, 2, 3 or interval 3 years, useful also for non-temporal series)

Time stamp type

Timeline illustrates different types of time stamp and the temporal relationship between the data sets

time instant - snapshot is shown as points
time interval - defined by start and end time, shown as thick line, examples show intervals with gaps and overlapping intervals

Space-Time dataset types

space-time raster data set: strds
space-time vector data set (points, lines or areas): stvds
space-time 3D raster data set: str3ds

subset of the vector (points) and raster (DEMs) space-time data sets for Jockey's Ridge

Working with a space-time data set

Given a set of GRASS GIS maps with assigned time:

create empty space-time dataset
register the existing maps in this space-time dataset
check/verify the properties of the dataset
process, analyze, and visualize the space-time data

Register maps and check data timeline

Given a set of DEMs for Jockey's Ridge dunes:
- photogrammetry (1974, 1995, 1998),
- lidar (1999, 2001, 2007, 2008, 2009, 2012, 2015, 2017),
- structure from motion from UAS imagery (2016, 2018)
register the raster maps as strds and check time line, spatial extents, per-cell elevation change

Mapping common spatial extent

Analyze spatial coverage of DEMs in the strds:

Count map: number of maps where a grid cell has non-null value
Intersection map: grid cells with non-null values from all maps in strds

Corrections for systematic errors

Using stable features (e.g. roads) to identify potential systematic errors and apply corrections

Critical for computing coastal erosion but also for growth of vegetation and other applications

Visualize DEMs

Visualize the Jockey's Ridge harmonized strds as a 3D animation masked by the intersection map:

Note that the visualization does not respect the variable time interval

Basic multitemporal DEM analysis

Per cell statistics computed for each cell over time:

mean and standard deviation
min, max value and range
time at minimum, time at maximum
linear regression: slope, offset, regr. coefficient

Starek MJ, et al. 2011, Modeling and analysis of landscape evolution using airborne, terrestrial, and laboratory laser scanning, Geosphere, 7(6), p. 1340-1356, doi:10.1130/GES00699.1

Mitasova, H. et al. 2009, New spatial measures of terrain dynamics derived from time series of lidar data, Proc. 17th Int. Conf. Geoinformatics, Fairfax, VA.

Hardin, E., et al, 2014, GIS-based Analysis of Coastal Lidar Time-Series, Springer Briefs in Computer Science

Basic multitemporal DEMs analysis

Given $n$ DEM snaphots $z(i,j,t_k) \quad k=1, ..., n$, per-cell operation can be used to derive:

Core surface: $z_{core}(i,j)=\min_k z(i,j,t_k) \quad k=1, ..., n$
Envelope surface: $z_{env}(i,j)=\max_k z(i,j,t_k) \quad k=1, ..., n$
Time of min: $t_{min} (i,j) = t_p, \quad {\rm where} \quad z(i,j,t_p) = z_{core}(i,j)$
Time of max: $t_{max} (i,j) = t_l, \quad {\rm where} \quad z(i,j,t_l) = z_{env}(i,j) $

Core surface is the minimum elevation measured at each cell over the given time period, core defines the volume of sand that has not moved,

Envelope surface as the maximum elevation measured at each cell over the given time period, envelope defines the space within which the dune surface evolved.

Basic multitemporal DEM analysis: core, envelope

East Jockey's Ridge dune: (a) time of elevation maximum map, (b) core, envelope and DEM surfaces

Core and envelope animation

Impact of storms: Changes in structures

Credit: Weather news, House relocation hookedonhouses.net

Multitemporal DEMs: analysis

Discrete changes - buildings lost or new:
$z_{env}(i_c,j_c) - z_{core}(i_c,j_c) > h_b$

Lost structure: $t_{max} (i_l,j_l) < t_{min} (i_l,j_l)$
New structure: $t_{max} (i_l,j_l) > t_{min} (i_l,j_l)$
Shoreline band - shoreline from core and envelope

Basic time series analysis: regression

applies to systematic monitoring with longer time series, with rate of change close to linear
per cell linear regression analysis: map of regression slope (rate of change)