1 of 27

EO BALTIC PLATFORM FOR GOVERNMENTAL SERVICES (EO-BALP) Reference nr. ESA AO/1-11741/23/I-NB ��ESA Contract No. 4000142702/23/I-NB

Forest Monitoring Application

06.02.2025

2 of 27

Agenda

​

10:30 - 10:45  Introduction of the EO-BALP Project and the Forest Monitoring Application

10:45 - 11:10  Demo of the Forest Monitoring Application

11:10 - 11:20  Accomplished accuracies of the forest data

​

Q&A session

​

3 of 27

GeoHub Platform

More information about the EO-BALP Project: https://www.forestradar.com/projects/esa-eo-balp

4 of 27

Forest Monitoring Application, beta version

  • Forest industry from all Baltic countries invited.
  • For Latvian organisations it is possible to make a seperate webinar in Latvian, please contact ilze@baltsat.lv
  • One month left for further developments – please, provide your feedback as it is very valuable.

https://docs.google.com/forms/d/e/1FAIpQLSfv3Nl3ZC8iBytFy9DXu4OzGNmJUBxB3fjuVV6rQ7yA_P6R9A/formResponse

​

5 of 27

Scenario 1: �Automated clear-cut monitoring system in the Baltics

Register

Review the plot analysis

1) current status (0-30%; 30-70%; 70-100%)

2) forest area (%, ha)

3) damaged area (%, ha)

4) previous monitoring date

5) change since last monitoring time

​

Upload forest plots with attributes and edit them in the web app if needed

​

Evaluate the area using various spatial data (all Baltics)

​

Download the result in GeoJSON and CSV formats

​

6 of 27

Scenario 2: �Automated forest health risk monitoring �(clear-cuts, windfalls, fires, pests, illnesses, increased moisture, etc.)�

Specify forest plots for analysis or send as shapefile to info@forestradar.com

Evaluate the area using various spatial data (all Baltics)

​

Receive risk areas in the app after the end of the vegetation season

Specify the time period of the analysis (year)

Register

​

Download the result in GeoJSON and CSV formats

7 of 27

Forest Monitoring - Layers (orthophoto map)

8 of 27

Forest Monitoring - Layers (satellite map)

9 of 27

Forest Monitoring - Layers (forest health)

10 of 27

Evaluate and download result (GeoJSON and CSV formats)�

Cadastre map

Topographic map

CSV file to easy copy and paste coordinates in other map apps

11 of 27

Layers

12 of 27

Bookmarks

The application saves your settings and map areas before closing.

13 of 27

Demo of the Forest Monitoring Application

14 of 27

EO BALTIC PLATFORM FOR GOVERNMENTAL SERVICES (EO-BALP) Reference nr. ESA AO/1-11741/23/I-NB ��ESA Contract No. 4000142702/23/I-NB

Acomplished accuracies of the forest data

06.02.2025

15 of 27

Data layers

  1. Tree cover masks: for cutting detection
  2. Damaged tree crown masks: bark beetle damage detection

16 of 27

  1. Tree cover masks: data layers
  • Based on Sentinel-2 satellite images (prepared using one date image)
  • Using Forest Registry (for Latvia, Lithuania) or CHM (for Estonia) as ground truth

​

Green: tree cover detected

Transparent: no tree cover

17 of 27

  • Tree cover masks: data layers
  • These masks are employed to calculate monitoring statistics for the polygons of interest
  • Case if the results of one date is available

​

18 of 27

  • Tree cover masks: data layers
  • These masks are employed to calculate monitoring statistics for the polygons of interest
  • Case if the results of two dates are available

​

19 of 27

  • Tree cover masks: accuracy

Accuracy was evaluated for masks based on summer images of Estonia using CHM of 2023 at stand level

​

​

Tree cover class according to CHM

Number of forest stands

Detected as non-forest

Detected as forest

0-5%

2055

2005 (97.6%)

50

6-10%

2258

2160 (96%)

98

11-20%

2199

1780 (81%)

419

21-40%

2081

919

1162 (56%)

41-60%

2554

118

2436 (95%)

61-80%

9200

6

9194 (99.9%)

81-100%

43660

0

43660 (100%)

20 of 27

  • Tree cover masks: accuracy

🟩At tree canopy cover between 10 and 40% there will always be high detection uncertainty due to the reflectivity characteristics of the background ground cover

​

​

Tree cover class according to CHM

Number of forest stands

True Positive

True Negative

False Positive

False Negative

0-5%

2055

-

97.6%

2.4%

-

6-10%

2258

-

96.0%

4.3%

-

11-20%

2199

​

​

​

​

21-40%

2081

​

​

​

​

41-60%

2554

95%

-

-

4.6%

61-80%

9200

99.9%

-

-

0.06%

81-100%

43660

100%

-

-

0%

21 of 27

  • Tree cover masks: accuracy

​

🟩At tree canopy cover lower than 10% and higher than 40%, detection accuracy is above 95%

🟩The smallest tree canopy cover that can be detected at pixel level with sufficient confidence is 40% per 900m2

🟩It should be noted that due to mixed pixel, illumination and co-registration errors, the average offset between the S2 result boundaries and the CHM boundaries is approximately 9 meters

​

22 of 27

  • Tree cover masks: accuracy

🟩Results obtained in autumn and winter can have lower accuracy than those obtained in summer/spring (no snow)as well as for images with haze/fog

​

23 of 27

2. Damaged tree crown masks: data layers

  • Based on PlanetScope 3m/pixel satellite images
  • Two stage algorithm: 1) detection at pixel level, 2) analysing context and adding confidence class for each potential damage (high, medium, low, cut)

​

​

24 of 27

2. Damaged tree crown masks: data layers

  • Based on PlanetScope 3m/pixel satellite images
  • Two stage algorithm: 1) detection at pixel level, 2) analysing context and adding confidence class for each potential damage (high, medium, low, cut)

​

​

25 of 27

2. Damaged tree crown masks: accuracy

​

  • Validation by field inspection was performed by State Forest Service

🟩 62 forest stands checked: 52 conform, 10 false positives mostly due to cutting acitivities

🟩After introducing context evaluation number of false positives reduced from 16% to 5%

​

​

26 of 27

2. Damaged tree crown masks: accuracy

​

  • Validation by comparing with UAV ortophotomaps (10 test stands in Latgale)

​

​

27 of 27

​

​

🟩 Sattelite images and machine learning provides regular information over vast areas

​

​

​