Bastrop (TX, USA) forest fire cross-sensor change detection images

Abstract. 

This dataset is composed of a set of four images acquired by different sensors over the Bastrop County, Texas (USA). On September 4, 2011, the region has been struck by ‘‘the most destructive wildland-urban interface wildfire in Texas history’’ which caused 2 casualties, more than 1300 destroyed buildings and almost burned entirely the Bastrop county state park.

This dataset is composed of a pair of pre- and post-event images from the same sensor, the Landsat 5 TM (L5T1 and L5T2), which are completed by a post-event of another sensor, the Advanced Land Imager (ALI) from the Earth Observing (EO-1) mission, acquired very shortly after the L5T2 (denoted ALIT2). These three scenes are very similar to each other and no apparent changes between L5T2 and ALIT2 are visible, since images were acquired within 1 day. Differences between these pre- and post-event pairs are only due to burned forest since they were acquired at a 16 days interval during summer. We also dispose of a fourth image of the same area acquired one year and 9 months after the forest fire by the Landsat 8 Operational Land Imager (OLI, L8T2 hereafter). The differences between L5T1 and L8T2 are significant, due both to sun/sensor angles and the long temporal interval between acquisitions. A whole new series of building has been constructed in the burn scar and many cultivated crops are at a different stage of growth.

Table : Dataset description

|           | Pre-event    | Post-event 1 | Post-event 2 | Post-event 3  |
|---------  |--------------|--------------|--------------|---------------|
| Filename  | t1_L5        | t2_L5        |  t2_ALI      | t2_L8         |
| Sensor    | Landast 5 TM | Landsat 5TM  |  EO-1 ALI    | Landsat 8 OLI |
| Channels  | 7            | 7            |  9a          | 7b            |
| GSD       | 30, 120      | 30, 120      |  30          | 30            |
| Sp. Range | [0.45–2.35,  | [0.45–2.35,  |  [0.4–2.4]   | [0.43–2.25]   |
| [\mu m]   | 10.40–12.50] | 10.40–12.50] |              |               |

We prepared the ground truth for pairs of change detection problems by photo-interpretation and relying on the maps provided on the emergency response website [1]. In the ground truth involving the L5T1-L8T2 problem we also included changes related to vegetation density and vegetation/bare soil transitions, since these changes are of the same spectral class as those related to the burned scar.

This data has been collected from the NASA LP DAAC Program [2], we are free to redistribute the data. For this reason we provide the Bastrop data and the ground truth we defined and used in these experiments (subset of original crops and ground truth).

Data Files

The archive contains the following files:

data
 ├── t1_L5.tif
 ├── t2_L5.tif
 ├── t2_ALI.tif
 ├── t2_L8.tif
 ├── ROI_1.tif
 ├── ROI_2.tif
 ├── Cross-sensor-Bastrop-data.mat

Where:

• `t1_L5.tif` is the pre-event image acquired by Landsat 5 TM
• `t2_L5.tif` is the post-event image acquired by Landsat 5 TM - `t2_ALI.tif` is the post-event image acquired by EO-1 ALI
• `t2_L8.tif` is the post-event image acquired by Landsat 8 OLI
• `ROI_1.tif` is the ground truth for the change detection problem between `t1_L5.tif` and `t2_L5.tif`
• `ROI_2.tif` is the ground truth for the change detection problem between `t1_L5.tif` and `t2_L8.tif`
• `Cross-sensor-Bastrop-data.mat` is a Matlab file containing the data in a more convenient format for Matlab users

Citation

If you are using this dataset, please cite the following paper:

@article{volpi2015jisprs,
  author = {Michele Volpi and Gustau Camps-Valls and Devis Tuia},
  title = {Spectral alignment of multi-temporal cross-sensor images with automated kernel canonical 
  correlation analysis},
  journal = {ISPRS Journal of Photogrammetry and Remote Sensing},
  volume = {107},
  pages = {50-63},
  year = {2015},
  doi = {https://doi.org/10.1016/j.isprsjprs.2015.02.005},
  url = {https://www.sciencedirect.com/science/article/pii/S0924271615000404},
  issn = {0924-2716},
}

Volpi, M., Camps-Valls, G., Tuia, D. (2015). Spectral alignment of multi-temporal cross-sensor images with automated kernel canonical correlation analysis, ISPRS Journal of Photogrammetry and Remote Sensing, Volume 107, 2015, Pages 50-63. https://doi.org/10.1016/j.isprsjprs.2015.02.005