Automated Spatio-Temporal Analysis of Remotely Sensed Imagery for Water Resources Management

BACKGROUNDBACKGROUNDBACKGROUNDBACKGROUND Since 2012, California faces an extreme drought, impacting water supply in many ways.

OBJECTIVESOBJECTIVESOBJECTIVESOBJECTIVES Use of latest time series analysis capabilities to examine surface water in reservoirs located along the western flank of the Sierra Nevada region of California.

Automation via the object-oriented and IDL-based ENVITask API for operational use.

RESOURCESRESOURCESRESOURCESRESOURCES Landsat images (L-5 TM, L-7 ETM+, L-8 OLI) 1999-2015 (October acquisitions).ENVI, IDL, and ArcGIS®.

Thomas Bahr

Harris Geospatial, Gilching, Germany – thomas.bahr@harris.com

EGU General Assembly 2016

Vienna, Austria | April 17 – 22

EGU2016EGU2016EGU2016EGU2016----2843284328432843

• Based on a time series of satellite imagery, the results of this case study verify the drastic decrease of the amount of surface water in the AOI, indicative of the major drought that is pervasive throughout California.

• The simple implementation in online services with the technology of HARRIS and ESRI makes it possible to calculate and retrieve tailored, individual analyses for user-defined areas on-the-fly.

DATASETSDATASETSDATASETSDATASETS 17 Landsat images, georeferenced (UTM Zone 10N WGS84), radiometricallyuncorrected.

METHODOLOGYMETHODOLOGYMETHODOLOGYMETHODOLOGY The following ENVITasks were used to pre-process the Landsat images:

• Triangulation based gap-filling for the SLC-off Landsat-7 ETM+ images.

• Spatial subsetting to the same geographic extent.

• Radiometric correction to top-of-atmosphere (TOA) reflectance.

• Atmospheric correction using QUAC®, which determines atmospheric correction parameters

directly from the observed pixel spectra in a scene, without ancillary information.

DATASETSDATASETSDATASETSDATASETS 17 Landsat images, spatially subsetted, georeferenced, radiometrically corrected.

METHODOLOGYMETHODOLOGYMETHODOLOGYMETHODOLOGY Classification was executed with the following ENVITasks:

• Creation of Modified Normalized Difference Water Index images (MNDWI, XU 2006) to enhance open water features while suppressing noise from built-up land, vegetation,

and soil.

• Threshold based classification of the water index images to extract the water features.

• Classification aggregation as a post-classification cleanup process.

DATASETSDATASETSDATASETSDATASETS 17 classification images derived from MNDWI images.

METHODOLOGYMETHODOLOGYMETHODOLOGYMETHODOLOGY Analysis and dissemination consisted of the following steps:

• Plotting the time series of water surface area [km2/y] and correlation with the daily reservoir elevations of the Don Pedro reservoir (station DNP, operated by CDEC).

• Animation of the classification series and export to a common video format.

• Export of the respective water classes to vector layers for further evaluation in a GIS.

This automated spatio-temporal analysis can be embedded in virtually any existing geospatial workflow for operational applications.

Tested integration options are:

• Deployment within service-based information systems, notably ArcGIS®. This allows to make the application available to organizations through their Portal for ArcGIS® website.

• Publishing the spatio-temporal analysis tasks as web-based services via the ENVI Services Engine (ESE) to existing enterprise infrastructures or cloud solutions.

• General implementation in geospatial workflows using the IDL bi-directional Python bridge.

Location of the Landsat images in California.(ENVI Reference Link, powered by Esri)

Landsat-8 OLI color composite (RGB=753), acquired 22 Sept 2015. The box indicates the Don Pedro reservoir.

Temporal and spatial subset of the Landsat time series (L-5 TM, L-7 ETM+, L-8 OLI) 1999-2015 (October acquisitions), showing the Don Pedro Reservoir.

Modified Normalized Difference Water Index image (MNDWI, XU 2006) to enhance open water features (subset).

Time series of water surface area per year.Correlation of the time series with daily reservoir elevations of the Don Pedro reservoir (station DNP, operated by CDEC).

IDL script (above) for implementation of the radiometric calibration and atmospheric correction in the ArcGIS® ModelBuilder (right) for an automated spatio-temporal analysis of remotely sensed imagery.

PREPROCESSING

CLASSIFICATION

ANALYSIS &

DISSEMINATION

AUTOMATED & OPERATIONAL APPLICATION

CONCLUSION

INTRODUCTION

www.harrisgeospatial.de

CLASSIFICATIONCLASSIFICATIONCLASSIFICATIONCLASSIFICATION

PREPREPREPRE----PROCESSINGPROCESSINGPROCESSINGPROCESSING

ANALYSIS ANALYSIS ANALYSIS ANALYSIS & & & & DISSEMINATIONDISSEMINATIONDISSEMINATIONDISSEMINATION

Image StackSample

SelectionGeocoded Time Series

Landsat ETM+

Gap Filling

Subsetted

Image Stack

Radiometric

Calibration

Calibrated

Image Stack

Classification

Aggregation

Raster to Vector

Conversion

Classified

Image Stack

Time Series

(*.shp)

Atmospheric

Correction

MNDWI ImagesThreshold

based

Classification

Classification

Images

Spectral Index

MNDWI

Pre-processed

Image Stack

Build

Time Series

Video

Animation

IDL Barplot

Time Series

(plotted)

PRO radcal_Landsat, filelist

COMPILE_OPT IDL2

e = ENVI(/HEADLESS)

; Get the radiometric calibration task

RCTask = ENVITask('RadiometricCalibration')

RCTask.Output_Data_Type = 'Float'

RCTask.Calibration_Type = 'Top-of-Atmosphere Reflectance'

; Get the quick atmospheric correction task

QUACTask = ENVITask('QUAC')

QUACTask.Sensor = 'Landsat TM/ETM/OLI'

; Loop over every file

FOREACH file, filelist DO BEGIN

raster = e.OpenRaster(file)

; Calibrate to top-of-atmosphere reflectance

RCTask.input_raster = raster

RCTask.output_raster_uri = e.GetTemporaryFilename()

RCTask.Execute

; Perform QUAC atmospheric correction

QUACTask.input_raster = RCTask.output_raster

QUACTask.output_raster_uri = e.GetTemporaryFilename()

QUACTask.Execute

ENDFOREACH

END

Classified water features on the 2015 Landsat color composite (RGB=543).

1999

2015201420132012

200820052002

2015