DIST-ALERT detects disturbances to any kind of vegetation cover, including forests, grasses, shrubs and even crops, occurring anywhere on Earth in near real-time.

1. DIST-ALERT: Mapping Near Real-Time
Vegetation Extent and Loss Based on
Harmonized Landsat and Sentinel-2 data
Matthew Hansen, Amy Pickens, Zhen Song,
Andre Lima, Andrew Poulson, Antoine Baggett
University of Maryland, College Park

2. DIST-ALERT
• Tracks disturbances globally
• Primary algorithm: Vegetationloss using time-series offractionvegetationcover
• Secondary algorithm: General spectral anomalies
• Employs Harmonized Landsat Sentinel-2(HLS)data
• 4 sensors: Landsat 8, Landsat 9, Sentinel-2A, Sentinel-2B
• Revisit rate of ~2-3 days
• 30 m
• Runs hourly as new data become available

3. Bare ground
Grass/shrubs/moss
Trees
0% 100%
Landsat 2010 percent cover
History - VegetationContinuous Fields
Hansen et al. 2002, 2003;
Carroll et al. 2010
Song et al. 2018

4. Algorithm Overview
• Collect drone data across manybiomes
• Calculate FVC from drone images and aggregatedto HLS-pixel scale
• Covert four bands of coincident HLS data to 3 Principal Components through PCA
• Build turn-key k-nearest neighbors (KNN) model and apply model to every HLS tile
NDVI
Drone Image
Drone Raw Data
Drone FVC
CoincidentHLS data
PCAs from HLS R,N,S1,S2
HLS-pixel level FVC
Trainingsamples
k-nearest neighbors
regression KNN Model

5. Spatial Distribution of Drone Sites
Drone data
 Bands: blue, green, red, red edge, nir
 Spatial resolution: ~ 8cm
 265 scenes collected

6. Drone image examples: various vegetation types
Diverse tree species and
selective logging, ROC
Fire impact, Republicof Congo (ROC) Growing and harvested cropland, US
R: red
G: green
B: blue

7. HLS tile of Sentinel 2
image of April 20, 2022

10. Battle of Upperville
Historic Park, VA
April 22,2022
drone image
NIR-RED-GREEN

11. Drone FVC(%)
100
75
50
25
0
Battle of Upperville
Historic Park, VA

12. Battle of Upperville
Historic Park, VA
Drone FVC(%)
100
75
50
25
0

13. HLS tile of
Sentinel 2 image
of April 20, 2022

14. Training samples and KNN model
HLS PCA1
HLS
PCA2
 Iterate to collect representative land cover/use
 Match drone data with coincident HLS data
 Collect 85K+ sample pixels from 265 drone images
 Fill the feature space by training samples
KNN model:
Drone FVC(%)
100
75
50
25
0
Trainingsamples

15. Training data and KNN model
HLS PCA1
HLS
PCA2
Predicted FVC(%)
100
75
50
25
0
KNN Model KNN model
 Iterate to collect representative land cover/use
 Match drone data with coincident HLS data
 Collect 85K+ sample pixels from 265 drone images
 Fill the feature space by training samples
 Build KNN model and predict the FVC at global scale
KNN model:

16. Battle of Upperville
Historic Park, VA
Predicted FVC(%)
100
75
50
25
0

17. 21 March 2023
Landsat8

18. 21 March 2023
Landsat8
Landsat9

19. 21 March 2023
Landsat8
Landsat9
Sentinel-2A

20. 21 March 2023
Landsat8
Landsat9
Sentinel-2A
Sentinel-2B

21. Fractional vegetationmapping
Vegetation cover percent
is mapped per HLS pixel,
defined as the amount of
skylight orthogonal to the
surface that is intercepted
by vegetation.
100%
0%
Washington, DC and Baltimore, MD, USA

22. Fractional vegetationmapping
Vegetation cover percent
is mapped per HLS pixel,
defined as the amount of
skylight orthogonal to the
surface that is intercepted
by vegetation.
100%
0%
Washington, DC and Baltimore, MD, USA

23. Fractional vegetationmapping
Vegetation cover percent
is mapped per HLS pixel,
defined as the amount of
skylight orthogonal to the
surface that is intercepted
by vegetation.
100%
0%
Washington, DC and Baltimore, MD, USA

24. Vegetation Change Monitoring
(DIST-ALERT Product)
• Near real-time vegetation fraction is
compared to a seasonal baseline
• The baseline is the minimum of the three
previous years of HLS-based vegetation
cover within a seasonal window of ±15 days
• Disturbance is monitored by tracking
vegetation fraction anomalies through time

25. Fire

26. DIST-ALERT
Forest fires in Quebec, Canada
100%
10%
Vegetation
loss
10 km
Potapovetal.The Global 2000-2020 Land
Coverand LandUse Change Dataset
DerivedFromthe LandsatArchive:First
Results.FrontiersinRemote Sensing
https://www.frontiersin.org/articles/10.33
89/frsen.2022.856903

27. DIST-ALERT

28. Logging

29. 100%
10%
Vegetation
loss
5 km
DIST-ALERT
Logging in Republic of Congo

30. Mining

31. 100%
10%
Vegetation
loss
Gold mining in Ghana

32. Urban
expansion

33. Residential growth Houston, Texas USA

34. Drought

35. 100%
10%
Vegetation
loss

36. Conversion
of natural
land

37. DIST-ALERT
Cerrado clearing for agriculture in Piaui, Brazil
100%
10%
Vegetation
loss

38. DIST-ALERT
Cerrado clearing for agriculture in Piaui, Brazil
100%
10%
Vegetation
loss

39. New soybean since 2020
Soybean established
before 2020
100%
10%
Vegetation
loss
DIST-ALERT
Song et al., 2021, Nature
Sustainability

40. Status and next steps
• Improved validated release of DIST-ALERT (V1) released March 14, 2024
• Operationally produced the provisional release of DIST-ALERT (V0) Feb
2023 to Feb 2024
• Land cover specific validation to provide accuracy for forests, cropland,
other short vegetation, and urban areas

41. https://lpdaac.usgs.gov/products/opera_l3_dist-alert-hls_provisional_v0v000/
https://lpdaac.usgs.gov/products/opera_l3_dist-alert-hls_v1v001/