Skip to content

Latest commit

 

History

History
93 lines (66 loc) · 3.31 KB

README.md

File metadata and controls

93 lines (66 loc) · 3.31 KB
Raw

Flood Mapping Comparison PH

This repository contains the results from the study: An Intercomparison of Sentinel-1 based Change Detection Algorithms for Flood Mapping.

The comparisons focus on the change detection flood mapping methods using Sentinel-1 SAR data.

All files provided with CC BY 4.0 license.

Methods

The research project aims to compare change detection methods that utilizes time-series parameters derived from Sentinel-1 datacube. Methods compared include:

  • Normalized Scattering Difference Index (NSDI)
  • and its Shannon's Entropy (SNDSI)
  • Standardized Residuals (SR)
  • TU Wien flood mapping method based on Bayes inference (TUWB)

Parameterization

The intercomparisons also shed light on method parameterizations specifically testing various:

No flood reference

no flood reference scenes tests prior image, and datacube computed references.

  • prior image (PRI)
  • harmonic day of year estimate (HDOY)
  • mean (MEAN)

Thresholding

Thresholding

  • Fixed threhold values from literature (FIX)
  • Otsu's (OTSU)
  • KA Illingsworth (KI)

HAND index filtering

Spatial filtering in aid of histogram thresholding can either be performed using the HAND filtering technique (HAND) or with no filtering (NONE).

TUW Bayes Method Parameterization

  • Bayes Inference prior probability

    • non-informed prior or equal 50-50 proir (EQ)
    • HAND based prior (HAND)

  • PDF based exclusion of pixels with low probability of labelling accuracy could either be applied (PDFX) or not applied (NONE).

File naming

The change detection flood maps are labelled as follows:

<Method>_<NoFloodRef>_<Thresolding>_<HANDfiltering>.tif

ex.: NSDI_HDOY_OTSU_HAND.tif, SNSDI_PRI_KI_NONE.tif

TUW Flood maps are labelled as:

<Method>_<NoFloodRef>_<Priors>_<PDFExclusion>.tif

ex.: TUWB_MEAN_HAND_PDFX.tif

Data and File format

Flood maps are encoded as uint8, where:

  • 1 - flood
  • 0 - no flood
  • 255 - no data.

Validation

External Sentinel-1 Reference Flood Maps

Reference flood maps from operation flood mapping activation used in the study were retrieved from:

  • Philippine Department of Science and Techonology (DOST) Advanced Science and Technology Institute (ASTI). Method is based on U-net framework document in this paper.
  • Sentinel Asia access point here. Flood map generated using a change detection workflow with manually selected threshold (*from direct communication). Both flood maps retrieved in shapefile format.

Sentinel-2 Reference

Sentinel-2 flood map was generated using MDWI change detection method. Sentinel-2 data used were:

  • S2A_MSIL2A_20201113T021941_N0214_R003_T51QUV_20201113T055836 (flood)
  • S2B_MSIL2A_20200909T021609_N0214_R003_T51QUV_20200909T065335 (no flood)

Software

Accuracy assessment metrics were performed using ABCRaster software. ABCRaster can accept refrence flood maps in shapefile format, and features random sampling and accuracy metric computations.

Miscellaneous information

Other relevant information:

Grid and Projection

Analysis was done on E058N117T1 tile using Equi7grid tiling and projection system.