This study presents our best estimate of the spatial distribution of horses which can be geo-located within mainland GB from the NED (n = 842,653 horses) at both regional and postcode area resolutions. The number of horses represented in this study is likely to be an underestimate of the true size of the horse population. A further approximately 500,000 horses may reside in the UK, but it is not clear from the NED that they are still alive or where they are located. In addition, it is not currently possible to produce an estimate for the numbers of horses which are registered within mainland GB, but which may have been subsequently exported to EU or other foreign countries. The origin of horses which have re-registered with a UK passport is also untraceable under the current system. The likely maximum estimate of the horse population in mainland GB is therefore in the region of 1,350,000 horses.
Apart from Scotland, the spatial distribution of horses, at the regional level, in the NED owner dataset was not very different to that of the Stakeholder horse data. However, important differences were apparent at the level of the postcode area. Compared to the Stakeholder horse dataset, there was an ostensibly higher density of horses in London in the NED owner dataset. This supports the previously held view that owner location cannot be viewed as a direct substitute for horse location in certain parts of the country [7
]. It also suggests that models parameterized with the NED owner data may incorrectly overestimate the impact of disease introduction and spread in these areas relative to others. However, it is also important to note that the number of horses reported to reside in these urban locations is small (approximately 1% of the total equine population of mainland GB) and as such the impact of these inaccuracies may be minimal in terms of GB-wide disease modeling or spread. The differences between datasets are attributable in part to the inevitable bias in the Stakeholder horse location dataset and in part to inaccuracies in the NED and could have an impact on future studies which attempt to model disease spread at different geographical resolutions.
Reassuringly, differences between datasets were minimal at a regional level, but at higher data resolutions (postcode areas) meaningful and potentially concerning differences were detected. This would suggest that these data could be used, for example, to parameterise the relative likelihood of contact (e.g. via horse movements) between geographic regions, and the relative likelihood that disease would persist in those regions, if introduced. However, this observation puts important limits on how fine the resolution at which detailed model predictions and recommendations can be confidently made to aid disease control. The importance of high resolution horse demography data depends on the type of pathogen involved in an infectious disease outbreak and disease control policy implemented. On one hand, regional data may be sufficient (for example in vector-borne disease models), but perhaps when dealing with non-vector-borne diseases, postcode area or still finer resolution data may be more desirable. Whilst the precise effects of the differences between datasets at different resolutions are difficult to predict, the risk exists that decisions based on more generic analyses at a lower resolution are likely to be less effective at a local level. Over-interpretation of any dataset, particularly when used for predictive modeling, may lead to errors being made in terms of disease control methods applied to individual animals in the face of a disease outbreak. These decisions may not only result in regrettable consequences for animals and their owners, but could well also erode public confidence in and compliance with, scientific advice.
Obtaining independently collected data on horse location from sectors of the industry (Stakeholder horse data) was difficult due to the diverse and fragmented nature of the 'equestrian industry' within GB. Apart from the data collected in the Stakeholder horse dataset, there is no other centralised database which is maintained by the equestrian industry which could be used independently to cross-reference the data in the NED. Although other important sectors of the equestrian industry (apart from the affiliations mentioned in the Stakeholder horse dataset) collect demographic data, these are typically recorded as owner rather than horse location. As expected, horse location data are routinely and rigorously collected for many registered horses (such as Thoroughbreds in racing and breeding and other competitions such as eventing). However, these sectors of the industry represent a very small proportion of all horses within GB. As a result, the spatial distribution of the horses in the horse dataset is largely driven by the location of horses on agricultural land (41% of the Stakeholder horse data). Clearly, both the NED owner and the integrated stakeholder datasets have their limitations. It is likely that horses not currently included in the Stakeholder horse dataset, will follow a different spatial distribution than reported here. This is an obvious bias associated with such Stakeholder data provided by individual sectors of the equine industry. The majority of horses not included in the Stakeholder dataset are likely to be classified as leisure horses used for riding and other leisure purposes and it is possible they are kept within livery yards or stables and have a more clustered distribution within urban and semi-urban areas.
In real time, the location of many horses in GB is likely to be far more dynamic than can be captured by any database. Even if the horse registration location is known and it is considered plausible that this is where the horse is kept, it does not necessarily represent where a horse is located on any given day or even from month to month or year to year. For example, many mares will leave their residence to go to stud for breeding in the spring. However, in the event of a disease outbreak in the GB horse population, national or regional decisions may have to be made regardless of whether or not detailed, validated population information is available. It is therefore reassuring that the two relatively independent estimates of population distribution (the NED owner and Stakeholder horse datasets) vary so little in the vast majority of mainland GB, though results from disease models based on any (or combinations) of these datasets would need to be interpreted cautiously in light of the uncertainty which exists between real-time horse location compared to what is recorded as horse or owner residence. Neither the NED dataset nor the aggregated stakeholder data were designed to aid disease control. Although the creation and maintenance of a single database that collates the horse population in its entirety has considerable value, it is critical that such data be validated against independent sources.