BlightSense: spatiotemporal analyses of potato late blight outbreaks in Great Britain


Information from the late blight (Phytophthora infestans) outbreak sampling (Fight Against Blight) was used to investigate why blight strains predominate in certain areas and their rate of spread between locations. This type of information helps improve blight risk assessment, especially in relation to new strains that enter GB.

Late blight severity is highly dependent on the local weather, so the timing, number and distribution of blight outbreaks reported through the FAB programme obviously varies from year to year. Nevertheless, there were still statistically significant patterns in space and time for early outbreaks, overall incidence, and the distribution of various pathogen genotypes. Together, these analyses provide valuable information on the variable risk posed by late blight across the potato production areas of GB.

Machine learning techniques were used to analyse weather, soil, geology, and topography data to identify the principle factors associated with late blight occurrence. The key findings were:

Early outbreaks

  • Spatial analyses: Averaged over all years, the statistically significant hot spots (clusters of high incidence) of early outbreaks of disease were generally found in the south of England and Wales, particularly near the coast, whereas a large cold spot (clusters of low incidence) extends across the production regions of Scotland. This supports the role of climate in the earliest occurrence of outbreaks.
  • Spacetime analyses: When analysed as a time series, the spatial hot spots of early disease tended to be sporadic: i.e. locations that are on-again off-again hot spots. There were also some sporadic and consecutive hot spots (a run of recent hot spots) of early outbreaks in the Angus / Aberdeenshire regions. This may relate to a warming climate resulting in an increased frequency of early outbreaks in Scotland in recent years.
  • Driving factors: As expected, the date of first outbreak was later in the north of GB, but only by a matter of weeks. Machine learning was used to develop a model that was 91.2% accurate in predicting low and high levels of incidence of early outbreaks. It identified temperature and precipitation as the most important predictors of early outbreak incidence.
  • Visual aids for decision-making: Colour-coded maps were produced to show the overall risk of early outbreaks by postcode district, and the week of the year these were most likely to occur.

Spatial spread of disease across the whole season

  • Risk of spread of disease among neighbouring postcode districts was highest in the potato growing regions of Tayside, Fife, Lothian, and East Anglia.
  • The velocity of spatial spread was calculated from early foci of genotypes 36_A2 and 37_A2 and ranged from 3-17 km per week.
  • Visual aids for decision-making: A colour-coded map was produced to show the risk of spread of disease among postcode districts.

 Pathogen genotypes

  • Spatial analyses: The mean spatial patterns of the genotypes 13_A2, 6_A1, 8_A1, 37_A2, and 36_A2 were analysed and differences in the central tendency, dispersion, and directional trends of these distributions were observed. The pattern of hot and cold spots varied markedly for each genotype, and some opposing patterns suggested competition and displacement.
  • Driving factors: A model was developed to predict the dominant genotype in each postcode district. It identified precipitation and humidity as the most important predictors, suggesting moisture plays an important role in competition among genotypes. The emergence and rapid spread of genotypes 13_A2 and 6_A1 in 2006-2007 was clear but there were few clear patterns to their distribution in the subsequent years. One obvious feature was the local spread of 6_A1 in eastern Scotland in 2011 followed by its dominance in the subsequent years. This indicates the significance of local sources of primary inoculum carried over from the previous season. However, the drivers of the large variation in other genotype distributions between years were not clear and require further investigation.
Project code:
01 September 2017 - 31 August 2019
Project leader:
Siobhan Dancey and Pete Skelsey


11120032_Summary Report_2017-2020 11120032_Full Report_2017-2020

About this project

  1. Determine whether the pattern of the earliest outbreaks of disease within each growing season is random, regular, or aggregated, and the spatial scales at which these patterns occur.
  2. Track the spread of late blight between postal districts over the course of each growing season to determine whether the pattern of outbreaks is random, regular, or aggregated, and the spatial scales at which these patterns occur.
  3. Identify geographic areas where incidence is consistently low or high over multiple growing seasons. Identify the principle factors associated with persistently low and high incidence of disease. Determine the relative risk of late blight occurrence within each postal district.
  4. Track the change in the spatial distribution of pathogen genotypes over the duration of the study period. Identify the principle drivers of this change.