Measurement of ecosystem services is an important part of policy-making in land management. Ecosystem services such as pollination, pollution treatment, and food production can be measured relatively easily, and their contributions quantified. However, it is harder to measure much more subjective cultural ecosystem services – how are aesthetics or well-being benefits to be judged, for example, if no two people experience a landscape in the same way? This issue leads to cultural ecosystem services being overlooked when modelling future scenarios in land management.
Social media data – principally geotagged photos – are increasingly being used to show where recreational, tourism, and cultural activities are being done, and can therefore be used as proxies to infer the value of some cultural services being provided by an ecosystem.
Manaaki Whenua’s researchers Dr Dan Richards and Dr Sandra Lavorel extracted over 150,000 geotagged landscape photographs uploaded to the website Flickr in New Zealand to develop an indicator of landscape appreciation. Photographs specifically uploaded for public viewing were accessed using an official Flickr interface and the content of the photographs was analysed using the image recognition software Google Cloud Vision and a machine learning technique known as maximum entropy modelling that is also used to model species distributions.
“Many people go outside to appreciate landscapes, but it is difficult to measure why certain landscapes are more popular. There is huge untapped potential to use the public information in social media to help us better understand people’s use and appreciation of nature,” said Dan.
This unusual combination of techniques allowed the researchers to delineate what types of landscape were photographed and where, giving a likely indication of the value of the landscape to the photographer. Around 40% of the photographs were defined in this way as landscape appreciation photographs, and over 70% had keywords associated with landscape appreciation.
The researchers then extended the reach of their study, to see whether machine learning could allow social media data on landscape appreciation to inform future planning decisions. How might appreciation of an agricultural landscape change if, for instance, native forest was restored in that landscape, and how might enhancing landscape appreciation have trade-offs with other ecosystem services objectives – such as storing carbon?
The results of their analysis showed that it was rarely possible to optimise both aspects of ecosystem service value – landscape appreciation and carbon storage – at the same time. Areas of high landscape appreciation – which tended to be coastal and closer to highly populated areas – did not necessarily gain further landscape appreciation value when increases in forest restoration were modelled. For distinctive agricultural landscapes such as the Canterbury Plains, scenarios involving native forest restoration could have a negative effect on landscape appreciation. At a national scale, the optimal method to increase carbon storage would be to focus native forest restoration in small areas, mainly in Canterbury. On the other hand, to bring the greatest benefit for landscape appreciation, it would be better to encourage native reforestation widely and at a lower density around the country.
The work showed that it is possible to quantify landscape appreciation as an indicator of cultural ecosystem services, and that this indicator can be meaningfully included in future landscape management scenarios alongside other, more familiar, ecosystem service indicators.
