Data mining for surface change discovery on the lunar surface from orbital images

Lead Research Organisation: University College London
Department Name: Mullard Space Science Laboratory


Transient events on the lunar surface, known as impact flashes (Boulet et al., 2012), have been observed by amateur and professional astronomers, for at least the last 450 years (Buratti et al., 2000). These are believed to be primarily due to meteoritic impacts, although there is some multispectral image evidence of volcanic emissions. However, there is only very limited evidence of the origin of these flashes in orbital images (loc.cit. and Buratti & Johnson, 2003) which have been published to date. One of the reasons for this is the massive task of working through a very large number of images acquired of the Moon from sources as diverse as Apollo astronaut photography, Lunar orbiter digitised film, CLEMENTINE multispectral frame images and the more recent pushbroom images from LROC.

Unfortunately, these images are not all uniformly co-registered to any base dataset and many do not even have any georeferencing information. Under the STFC change detection using data mining project (MSSL Consolidated grant ST/K000977/1), techniques have been developed for automated co-registration, DTM retrieval and subsequent orthorectification to produce time series of sub-pixel co-registered ORIs. However, these have only been applied to orbital images of Mars.

In this project, it is planned to modify and apply these automated registration techniques to the Moon employing all the relevant image data from the 1960s to the present-day very high resolution (<20cm) image data of the Moon from Japanese, Indian, Chinese and US sensors.

Data mining techniques will be employed to search for changes in lunar images after applying image normalisation to take into account differences in the viewing and illumination conditions and their interaction with the surface. These data mining results will also be validated using data from the MoonZoo project, where available, to assess which is the best source. Any changes detected will then be studied using the vast panoply of spectral and directional measurements available from orbit today in addition to the generation of the highest possible DTMs. Automated crater detection software, developed by the incoming student during his MSc Space Science dissertation, will be used to map large regions where new craters have been detected to determine what effect such random impacts have on the age dating using crater chronology methods developed by Neukum (1983,1984)

Cited references
Bouley et al. (2012) Power and duration of impact flashes on the Moon: Implication for the cause of radiation. Icarus, vol. 218 pp. 115-124

Buratti et al. (2000) Lunar Transient Phenomena: What Do the Clementine Images Reveal? Icarus, vol. 146 pp. 98-117

Buratti and Johnson. (2013) Identification of the lunar flash of 1953 with a fresh crater on the moon's surface. Icarus, vol. 161 pp. 192-197

Neukum, G.(1984) Meteorite bombardment and dating of planetary surfaces. Thesis - Feb. 1983. NASA Technical Memorandum TM-77558 (English translation of German thesis published originally in February 1983), pp. 1-158.


10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
ST/R505171/1 01/10/2017 30/09/2021
1912521 Studentship ST/R505171/1 06/11/2017 05/08/2021 Alistair Francis
Description Visiting Researcher position at ESA's Phi Lab 
Organisation European Space Agency
Department Centre for Earth Observation
Country Italy 
Sector Charity/Non Profit 
PI Contribution Worked on the ESRIN site within the Phi Lab team on computer vision models.
Collaborator Contribution Their computing resources, technical support, domain expertise and networking capability.
Impact Forthcoming labelled dataset of Sentinel-2 imagery Forthcoming publication on deep learning model
Start Year 2019
Title CloudFCN 
Description Novel deep learning cloud masking algorithm written in Python. Designed for use on Landsat 8 dataset, with scripts for experiments as conducted in the related paper: 
Type Of Technology Software 
Year Produced 2019 
Open Source License? Yes  
Impact User-friendly deep learning model for cloud masking in Landsat 8