ARD Preparation and Integration in Taiwan Data Cube

1This document contains proprietary and controlled information of National Space Organization (NSPO) of Taiwan and shall not be duplicated in whole or in part for any purposes without permission from NSPO.

Associate Researcher

Li-Yu Chang

ARD Preparation and Integration in

Taiwan Data Cube


Content

Pre-processing of remote sensing data

– Top of Atmospheric Reflectance of FORMOSAT-2/5 optical image

– Surface Reflectance of Sentinel-2 optical image

– Radar backscatter coefficient of Sentinel-1 SAR image

Taiwan Data Cube (TWDC) integration with ARD data

– FORMOSAT-2/5 ARD

– Sentinel-1/2 ARD

TWDC data export

– Custom mosaic of data cube application

– Jupyter envrionment

Application of TWDC

– Paddy rice detection using FORMOSAT-5 Images

– Flooding area detection using Sentinel-1 SAR Data

3

Pre-processing ofremote sensing data

Top of Atmospheric Reflectance of FORMOSAT-2/5 optical image


Process of optical remote sensing

Parameters that affect the received

radiance of optical satellite

– Reflectance

– Solar incident angle

– Sensor’s view angle

– Atmosphere related parameters

• Atmospheric transmission

• Path radiance

• Diffused radiance

Reflectance estimation of ground

object

– Surface Reflectance, SR

– Top of Atmosphere Reflectance, TOAR

• A reasonable approximation


Estimation of TOAR

DN (for Digital Numbers) to spectral radiance (or

TOA Radiance) L

L(b)=DN(b)/gain(b)+bias(b)

(in W/m2/steradians/micrometers)

TOA Radiance to TOA Reflectance

TOAR(b) = (π*L(b)*d*d) / (ESUN(b)*cos(θ))

L(b): the spectral radiance for band b

d: earth-sun distance (in astronomical units)

ESUN(b): mean TOA solar irradiance (or solar illumination) in W/m2/micrometers

θ : solar incident angle in degrees

θ

θ


Why we need reflectance

Reflectance as a normalized

observations for ground information

retrieval

– Multi-temporal image application

– Multi-sensor image application


DN、TOAR and NDVI obtained by

FORMOSAT-5 and SPOT-6

FORMOSAT-5 SPOT-6

2017/12/20 02:40

(UTC)

2017/12/20 02:15

(UTC)

Temporal information of used images

Location of Study Area

FORMOSAT-5

SPOT-6

Study area is 9km x 9km

NDVI of FORMOSAT-5

NDVI of SPOT-6


Comparison of FORMOSAT-5 and

SPOT-6：DN

DN of SPOT-6 Images

DN

of

FO

RM

OS

AT

-5 I

ma

ges

DN of SPOT-6 Images

DN

of

FO

RM

OS

AT

-5 I

ma

ges

DN of SPOT-6 Images

DN

of

FO

RM

OS

AT

-5 I

ma

ges

DN of SPOT-6 Images

DN

of

FO

RM

OS

AT

-5 I

ma

ges

Band 1 Band 2

Band 3 Band 4



SPOT-6：TOAR

TOAR of SPOT-6 Images

TO

AR

of

FO

RM

OS

AT

-5 I

ma

ge

s



TO

AR

of

FO

RM

OS

AT

-5 I

ma

ges


Band 1 Band 2

Band 3 Band 4

TO

AR

of

FO

RM

OS

AT

-5 I

ma

ge

sT

OA

R o

f F

OR

MO

SA

T-5

Im

ag

es

10This document contains proprietary and controlled information of National Space Organization (NSPO) of Taiwan and shall not be duplicated in whole or in part for any purposes without permission from NSPO.This document contains proprietary and controlled information of National Space Organization (NSPO) of Taiwan and shall not be duplicated in whole or in part for any purposes without permission from NSPO.


SPOT-6： NDVI

NDVI obtained

from DN

NDVI =IR − R

IR + R

NDVI obtained

from TOAR

NDVI of SPOT-6 Images

ND

VI

of

FO

RM

OS

AT

-5 I

ma

ges

NDVI of SPOT-6 Images

ND

VI

of

FO

RM

OS

AT

-5 I

ma

ges


Multi-temporal and Multi-sensor rice

phenology observation by TOAR

2017/07/20

RapidEye

2017/08/28

RapidEye

2017/08/06

SPOT-7

2017/10/01

RapidEye

2017/09/19

FORMOSAT-5

2017/12/14

RapidEye

2017/11/10

SPOT-6

2017/12/20

FORMOSAT-5


Multi-Temporal NDVI Composite for

rice observation

R: 2017/12/14 NDVI by RapidEye

G: 2017/09/19 NDVI by FORMOSAT-5

B: 2017/08/06 NDVI by SPOT-7


Rice phenology fitting by

double logistic functionSite 1

Rice Field

Site 2

Rice Field

Site 3

Building

Site 4

Forest

Red lines are fitted by

Double Logistics Function:

14


Surface Reflectance of Sentinel-2 optical image

15

Sentinel-2 Optical Data

https://sentinel.esa.int/web/sentinel/missions/sentinel-2/data-products

Buriram

Study Area:

Near by region

around Buriram

Multi-Temporal Remote Sensing Data of OCT-NOV, 2019

Paddy

fields

SENTINEL-2

(2019/10/21)

Maximum

Biomass

SENTINEL-2

(2019/11/20)

Harvested

Estimated paddy rice area:

14477.7(HA)



16

Sentinel-2 Optical Data

Level-2A Source data in Copernicus Open Access Hub is SR already

Once downloaded, no further pre-processing needed

17


Radar backscatter coefficient of Sentinel-1 SAR image


Sentinel-1 SAR Data

Acquisition modes (C-band, 5.404 GHz,

693KM)

– Stripmap (SM)

– Interferometric Wide Swath (IW)

– Extra Wide Swath (EW)

– Wave (WV)

ModeResolution

rg x az

Number of

looks

SM 1.7x4.3 m to 3.6x4.9 m 1x1

IW 2.7x22 m to 3.5x22 m 1x1

EW 7.9x43 m to 15x43 m 1x1

WV 2.0x4.8 m and 3.1x4.8 m 1x1

https://sentinel.esa.int/web/sentinel/technical-guides/sentinel-1-sar/sar-instrument

https://sentinel.esa.int/web/sentinel/technical-guides/sentinel-1-sar/sar-instrument


Interferometric Wide Swath

IW mode captures

three sub-swaths

using Terrain

Observation with

Progressive Scans

SAR (TOPS)


Product Types and Processing

Levels

L0 RAW

L1 SLC

L1 GRD• Full Resolution

• High Resolution

• Medium Resolution

L2 OCN• Ocean Wind Field

• Ocean Swell Spectra

• Surface Radial Velocity

L0 RAW

L1 SLC

L0 RAW

L1 SLC L1 SLC

L1 GRD• High Resolution


L1 GRD• High Resolution








L1 SLC


Product Types and Processing

Levels Level-0

– The SAR Level-0 products consist of compressed and unfocused SAR

raw data. Level-0 products are the basis from which all other high level

products are produced.

Level-1

– Level-1 focused data are the products intended for most data users.

The Level-0 product (raw data) is transformed into a Level-1 product by

the Instrument Processing Facility (IPF)


Why we need σ0 ?

Normalization of SAR imagery is

required for a meaningful analysis of

multi-sensor or even single-sensor

multi-track data

Radar backscatter coefficient σ0 is a

normalized estimation to enable more

robust use of the retrieved SAR data

values in applications


Radar backscattering coefficient σ0

Radar backscatter is the effects of terrain on

the radar signal that we are most interested

in, i.e. the amount of radar cross-section, σ,

reflected back to the receiver, per unit area

on the ground

σ

σ

Small, D. (2011). Flattening gamma: Radiometric terrain correction for SAR

imagery. IEEE Transactions on Geoscience and Remote Sensing, 49(8), 3081-3093.


Processing of σ0 for Sentinel-1 SAR Data

Manual procedure

– Thermal noise removal

– Calibration to radar

backscattering coefficient

– Deburst and merge for TOPS

data

– Multilooking

– Speckle filtering

– Slant Range to Ground

Range conversion

– dB Conversion

– Topographic Correction

Processing with ESA SNAP

(Sentinel Application Platform)


Processing of σ0 for Sentinel-1 SAR Data

Automatic procedure– Using Graphic Builder under ESA

SNAP to generate XML of

processing procedure

– Applying GPT (Graph Processing

Tool)、GDAL (Geospatial Data

Abstraction Library) and

customized code to create SAR

σ0 and corresponding Quick-Look

image

• Different IW mode data are

processed separately to increase

efficiency

– Building a batch process to

automate whole workflow

SNAP

Graphic Builder

Batch for

Workflow Control

ESA SNAP XML for GPT

26

Taiwan Data Cube (TWDC) integration with ARD data

27This document contains proprietary and controlled information of National Space Organization (NSPO) of Taiwan and shall not be duplicated in whole or in part for any purposes without permission from NSPO.Credit to Alex Held (CSIRO, Australia) @ workshop in Taiwan, 2017


Surface reflectance and radar

backscatter as ARD datasets

Truckenbrodt, John, et al. "Towards Sentinel-1 SAR Analysis-Ready

Data: A Best Practices Assessment on Preparing Backscatter Data for

the Cube." Data 4.3 (2019): 93.

http://ceos.org/ard/index.html#slide1

http://ceos.org/ard/index.html#slide1


Reflectance and radar backscatter

as ARD datasets in TWDC

Data Cube integration with

FORMOSAT-2/5 and Sentinel-1 data

– FORMOSAT-2/5: TOAR with quality Index

– Sentinel-1: Back scatter coefficient

– Sentinel-2: Surface reflectance with quality

Index

30

TWDC Integration of FORMOSAT-2/5 ARD


TWDC Integration of FORMOSAT-2/5

ARD

Pre-processing and orthoimage

generation

ARD preparation

– TOAR

– Quality index (QI)

ARD import


Pre-processing and rthorectification

By integrated processing of DMS (Data Management System) and

DPS (Data Processing System) in NSPO, the received FORMOSAT-

2/5 data can be processed in streamline and the orthoimage cab be

generated automatically

Product Generation Framework


TOAR ARD

product

Ortho-rectified

product

FORMOSAT-2/5 ARD import

Generation of FORMOSAT-2/5 ARD

including TOAR and QI is part of

processing procedure of Formosat-

2/5 image production line

ARD is sent to NCHC by FTP and

import to TWDC

– In NCHC, needed .PY and .YAML

scripts are prepared for importing

ARD to TWDC

Import to DC

QITOAR


Cloud coverage detection

Stage 1: Check if image cloud free

– Classify image by k-means

– Check cloud existence by spectral

thresholds

Stage 2: Extract cloud and potential

cloud

– Segment image by region growing

– Cluster segment patch by k-means

– Extract cloud and potential cloud with

different spectral thresholds

Stage 3: Cloud refinement

– Merge neighboring potential cloud by using

extracted cloud as seed

– Filter small potential clouds with area

threshold

– Remove broken parts and gaps in clouds by

morphology closing

FORMOSAT-2/5 Level4

MS TOAR Image

k-means clustering

Image Segmentation

Cloud Mask

Cloud free or

not

Cloud free image

Cloud Refinement

Image with Cloud

Higher

Spectral

Threshold as

Clouds

Lower

Spectral

Threshold as

Potential

Clouds

35

TWDC Integration of Sentinel-1/2 ARD


Integration of Sentinel-1 ARD

Procedure for automatic processing of

Sentinel-1 ARD

– Follow the steps suggested in

Copernicus Open Access Hub to

download Sentinel-1 data

• Obtain download List

• Download the data according to list

– Check integrity of downloaded file

– Radar backscattering coefficient ARD

generation (As describe above)

– Post processing for ARD

Prepare needed .PY and .YAML scripts

for importing ARD to DC

Above processing procedures are

triggered periodically for data updating

Period and Coverage Setting

Search Download List from

Copernicus Data Hub

Download SLC Products

Import σ0 ARD to DC

Generate σ0 Using SNAP GPT

and Perform Post-Processing

σ0 ARD Product in DC

Product

Integrity

Check

Success

Fail


Integration of Sentinel-2 ARD

Procedure for automatic processing

of Sentinel-2 ARD

– Follow the steps suggested in

Copernicus Open Access Hub to

download Sentinel-2 data

• Obtain download List

• Download the data according to list

– Check integrity of downloaded file

Prepare needed .PY and .YAML

scripts for importing ARD to DC

Above processing procedures are

triggered periodically for data

updating

Once L2A product is not available,

L1C product will be downloaded and

“Sen2Cor” process will be followed

to generate L2A product

Period and Coverage Setting

Search Download List from

Copernicus Data Hub

Download L2A Products

Product

Integrity

Check

Fail

L2A ProductL1C Product

Sen2Cor

L2A Product

38

TWDC data export

Custom mosaic of data cube application

39

Exporting data by custom mosaic

40

Single date data: 2018/11/21

Cloud Mask

41

Single date data: 2018/12/17

No Data Area

42

Multiple date data mosaic：2018/11/21+2018/12/17

Better Result

43

TWDC data export

Jupyter enviroment


TWDC data export with Jupyter


TWDC data export with Jupyter

FS-2 Image export for

Thailand Area

FS-2 Image export for

Taiwan Area

S-1 Image export for

Taiwan Area

46

Application of TWDC

Paddy rice detection using FORMOSAT-5 Images


Paddy rice detection using

FORMOSAT-5 Images Change detection for rice growing area in Taiwan

– Farm fields prepared in winter and filled with water in spring for rice

planting

– Some farmers also grow green manure or cover crops in winter to

increase land fertility or income

Source Data

– FORMOSAT-5 images in winter (2018/12/17) and spring (2019/03/01)

Applied approaches

– Change detection by Change Vector Analysis (CVA)

• Using CVA to detect changed area

– Normalized Difference Vegetation Index (NDVI)

• Delineating vegetation and bare soil areas in prior image

– Normalized Difference Water Index (NDWI)

• Extracting water area in later image


Workflow of paddy rice detection

Prior Period ARD Image Later Period ARD Image

Obtaining NDVI Obtaining NDWI

Change Vector Analysis

(CVA)

Paddy Rice

with green manure

Is Land

Changed?

(CVA > TCVA)

Is Prior Period

Vegetation?

(NDVI > TNDVI)

Is Later Period

Water?

(NDWI>TNDWI)

No Paddy RicePaddy Rice

without green manure

YES

YES

YES

NO

NO

NO


Paddy rice detection with ERDAS

Imagine modeler (Post-processing)


Paddy rice detection with Jupyter

(Online processing)


Source FORMOSAT-5 images

2018/12/17 2019/03/01


Analyzed result

Paddy rice with green manure Paddy rice without green manure

2018/12/17 2019/03/01


Map publishing for rice field map

54

Application of TWDC

Flooding area detection using Sentinel-1 SAR Data


Flooding area detection using SAR

Images

Radar backscatter of water is small due to

water’s specular reflection property

If an area was changed from non-water to

water, the corresponding backscatter

coefficients should be decreased

𝜎𝐿𝑎𝑡𝑒𝑟0 − 𝜎𝑃𝑟𝑖𝑜𝑟

0 <δ

In addition, following criteria should be

satisfied as well

𝜎𝑃𝑟𝑖𝑜𝑟0 > 𝜎𝑊𝑎𝑡𝑒𝑟

0

𝜎𝐿𝑎𝑡𝑒𝑟0 ≤ 𝜎𝑊𝑎𝑡𝑒𝑟

0


Workflow of flooding area detection

Prior Period SAR

Backscatter Image

Later Period SAR

Backscatter Image

Multi-Temporal

Backscatter Differencing

Flooded Area

Is Difference

Smaller Than

Threshold?

Is Later Period

Water?(𝜎𝐿𝑎𝑡𝑒𝑟

0 ≤ 𝜎𝑊𝑎𝑡𝑒𝑟0 )

Is Prior Period

Non-Water?

(𝜎𝑃𝑟𝑖𝑜𝑟0 > 𝜎𝑊𝑎𝑡𝑒𝑟

0 )

Not Flooded Area

YES

YES

YES

NO

NO

NO


Flooding area detection with ERDAS

Imagine modeler (Post-processing)


Flooding area detection with Jupyter

(Online processing)


Flooding over south Taiwan due to

heavy rain on 2018/8/23

Sentinel-1 2018/08/14 Sentinel-1 2018/08/26


Analyzed result

Sentinel-1 2018/08/14 Sentinel-1 2018/08/26

Flooded area


Map publishing for flooded area

Documents

ARD Preparation and Integration in Taiwan Data Cube