CHANGE DETECTION

FOR DIGITIZING LAVA

FLOW USING LANDSAT

8 IMAGERY

Robert Kozub

Advanced Remote Sensing

Spring 2022

Introduction/Overview

Thematic Change: Supervised

Multi-Date Change Detection

Sources

Imagery

Conclusion

Thematic Change: Unsupervised

TABLE OF

CONTENTS

La Palma is the northwest island of the Canary Islands, Spain. The total population as of 2021 was

86,267 (source: statista ). On September 19th, Volcán Cumbre Vieja began to erupt, resulting in

over 1000 buildings and roads destroyed and 1 known death. In addition to the destruction

caused by the eruption, as many as 6,400 people have been forced to evacuate the area (per

science.org).

In a previous effort, the lava flow from the eruption of Volcán Cumbre Vieja was manually

digitized by evaluating Landsat 8 data for the purpose of estimating the number of buildings and

roadway segments that were either destroyed or damaged by lava (see the previous project here).

However, the manual digitization was not as accurate compared to efforts performed by others

studying the area. The purpose of this continuation is to see if using change detection methods

would increase the digitation accuracy of where the lava has flowed for a more accurate analysis

of the number of buildings and roadway segments encompassed by lava.

This report will detail the process of experimenting with different change detection methods

such as Multi-Date Visual Change Detection Using Layer Stacking, and using unsupervised and

supervised Thematic Change to perform post-classification change detection.

INTRODUCTION

Western

Sahara

La Palma

IMAGERY

Three Landsat 8 images were downloaded via Earth Explorer from May 2021, September

2021 (shortly after initial eruption), and February 2022 (after eruption stopped). Landsat 9

imagery was available for 2022 dates, but there was too much cloud cover to conduct any

land cover analysis.

Landsat 8 OLI/TIRS C2 L1

May 2021

No. 01 —

Landsat 8 OLI/TIRS C2 L1

September 2021

No. 02 —

Landsat 8 OLI/TIRS C2 L1

February 2022

No. 03 —

Stack Layers in ArcGIS Pro

Using the Imagery tab, bands 1-7 for each image are combined into singular composite images

(Imagery -> Process -> Composite). This function is similar to layer stacking in ERDAS imagine,

but takes significantly less time.

Multi-Date Visual Change Detection Using Layer Stack

To test this method, the May 2021 and September 2021 Landsat 8 images were used to see if lava

flow was apparent. For this process, the NIR bands (band 5) was used to show if multi-date

change detection would work for purposes of showing the lava flow. First, the NIR bands were

composited and then individually evaluated. The blue band for each image was turned off so that

only the red and green remained. The difference was then computed using the Arithmitic

function in ArcGIS Pro (Imagery Tab -> Raster Functions -> Math -> Arithmitic). The result is seen

below in Figure 1.

Unfortunately, this method did not work well and no discernable changes were present as can be

seen in the area of interest (black polygon).

MULTI-DATE

CHANGE DETECTION

Figure 1: Multi-Date Visual Change

Unsupervised Classification for Land Cover

The first unsupervised classification performed was for the May 2021 and September 2021

Landsat 8 images using the Iso Cluster Unsupervised Classification tool (Tools -> Spatial Analyst

Tools -> Multivarite -> Iso Cluster Unsupervised Classification) in ArcGIS Pro. This tool combines

the Iso Cluster and Maximum Likelihood Classification tools. Since there are some areas of the

imagery where different land cover types are not clear, this tool helps categorize the different

pixels automatically.

Five classes were used for each image to help differential the different land types and hopefully

separate out where there is active lava and volcanic ash from the current and other historic

eruptions. However, as seen in Figure 2, the classes were aggregated together and therefor did

not produce the desired results.

The September 2021 image (Figure 3) did show where there were active lava flows (red area seen

within the black polygon), but also classified other land types that were not active lava. The

Thematic Change image in Figure 4 also shows the active lava flow, but again combines other

land class types that are not lava. Additionally, volcanic ash could not be classified using the

unsupervised classification, which is important to show where the lava has already spread and

cooled.

THEMATIC CHANGE:

UNSUPERVISED

Figure 2: May 2021 Unsupervised Figure 3: Sept 2021 Unsupervised Figure 4: Thematic Change

(May - Sept 2021)

Supervised Classification for Land Cover

The Unsupervised Classification did not produce desired results in differentiating between

different landcover types, especially for volcanic ash and old lava flow areas. The Supervised

Classification method was then used to add more land cover types and produce training samples

that will help the classification process. Six classes were used: Water (1), Vegetation (2),

Rock/Dirt/Barren (3), Volcanic Ash/Old Lava (4), Urban (5), and Clouds(6) and 10-20 training

samples were collected for each class. This analysis was done using the May 2021 and February

2022 Landsat 8 images. Volcán Cumbre Vieja concluded erupting in early January 2022 and

therefor active lava flow was removed from the land cover classes. Clouds and rock/dirt/barren

soil were added to the classes so that it becomes more discernable from the output what is

volcanic ash and to differentiate cloud coverage so that it was not classified as another land cover

type.

The results from both the supervised classifications of the May 2021 and February 2022 images

showed good results in relaying where volcanic ash/old lava had been (yellow), but some of the

areas where there was volcanic ash/old lava in May 2021 did not appear in February 2022, but

did show the increase where Volcán Cumbre Vieja erupted. The thematic change also showed

good results and you can see where there was old lava from historic eruptions (maroon/brown)

and where the recent eruption occurred (bright red/orange). However, urban areas did not show

up very well and blends in with vegetation. Overall, this method seemed to be the best method to

show the recent lava flow/volcanic ash/old lava.

THEMATIC CHANGE:

SUPERVISED

Figure 5: May 2021 Supervised

Figure 6: February 2022 Supervised Figure 7: Thematic Change

(May 2021 - Feb 2022)

Figure 8: Supervised Thematic Change - May 2021 to February 2022

Water

Vegetation

Dirt/Rock

Volcanic Ash

Urban

Clouds

300%

200%

100%

-100%

Class

Original Pixel

New Pixel

Percent Change

Water

36,126,024.00

28,456,880.00

-21%

Vegetation

151,279.00

326,140.00

116%

Dirt/Rock

423,018.00

675,233.00

60%

Old Lava/Volcanic Ash

47,585.00

101,016.00

112%

Urban

3,531,058.00

6,801,467.00

93%

Clouds

1,261,201.00

4,732,605.00

275%

Increase in the amount of

Volcanic Ash/Old Lava Areas

112%

Percent Change

Percent change in land cover types from May 2021 to February 2022.

Increase in Urban Areas

93%

Lastly, the percent change in land cover types from May 2021 to February 2022 are

calculated by dividing the difference in pixels by the original number of pixels from the

initial classifications.

Key statistics - The was a drastic increase in the amount of volcanic ash that came from

the volcanic eruption that started in September 2021 and ended in early January 2022.

Key findings - The drastic decrease in water was most likely due to change in cloud

coverage and lava that made its way to the ocean. The vegetation, dirt/rock, and cloud

cover increases are due to misclassification and increased cloud coverage from one

image to the other.

-21%

116%

60%

112%

93%

275%

Figure 9: Percent Change in Land Cover Types

Table 1: Percent Change in Land Cover Types

Multi-Date Change Detection

This method was least effective in digitizing lava flow and

volcanic ash areas. Since the areas where volcanic ash/old

lava are located are sometimes hard to differentiate

between that of vegetation and barren land, many land

cover types blend together and makes it difficult to tell

where lava has been.

Thematic Change: Unsupervised

The Unsupervised Thematic Change Detection method

worked well to highlight the area where there was active

lava flow for the September 2021 image, but failed to

separate other land cover types and essentially blended

vegetation, urban areas, and volcanic ash/old lava areas

together across the island. When performing this method

on the February 2022 image and performing the thematic

change from May 2021, there was no change shown on the

map and therefor discarded from the analysis. There may

be ways to improve upon this method to help mitigate

those issues and this method does save time compared to

supervised.

Thematic Change: Supervised

The Supervised Thematic Change Detection method

worked the best of the three in relaying where volcanic

ash/old lava had been while also doing a decent job of

showing the other land cover types such as vegetation,

rock/barren land, and urban areas. Some land cover types

blend together, but that can be improved by performing

the analysis with more training samples.

CONCLUSIONS

SOURCES

EarthExplorer . (2021). Retrieved 6 November 2021, from

Survey, U. S. G. S.- U. S. G. (n.d.). Earthexplorer. EarthExplorer. Retrieved May 1,

2022, from https://earthexplorer.usgs.gov/

Copernicus Emergency Management Service Mapping.

Science | AAAS. (2021). Retrieved 6 November 2021, from

Statista The Statistics Portal. (2021). Retrieved 6

Land Cover Classification System. Google Books. Retrieved May 1, 2022, from

https://books.google.com/books?

id=xUyVNK98gTkC&lpg=PR4&ots=qs00BoHGs1&dq=land+cover+classification&lr&pg=

PR4#v=onepage&q=land%20cover%20classification&f=false

https://earthexplorer.usgs.gov/

(2021). Retrieved 6 November 2021, from

https://emergency.copernicus.eu/mapping/

https://www.science.org/

December 2021, from https://www.statista.com/