Jan 272017

I’ve spent the last few days working on the data pipeline I first mentioned in The Cloud Lottery that is Scottish satellite imagery. This pipeline consists of a series of R scripts that will place an order for Landsat 8 surface reflectance data and vegetation indices from the United States Geological Service, download the products as they become available then strip out the layers I’m interested in analysing, removing any areas of cloud or high atmospheric aerosols as they go. The hope is I’ll be able to leave this running and build a continuously updated dataset covering areas of interest that I can then query for data analysis purposes.

The Open Street Map of Edinburgh with Enhanced Vegetation Index (EVI) overlay. A particularly clear day on 25th October of 2016

When working on a job like this it is easy to get buried in the detail and lose sight of the end goal. As it is Friday afternoon and next week I need to turn my attention to other projects for a while I thought I’d visualise one of the clearer images and check I’m still on course. The images are fascinating so I wanted to share them.

You may have seen a recent article in the Guardian “How green is your city? UK’s top 10 mapped and ranked”. Edinburgh came out top with 49% “green”. These maps were produced by ESRI based on Normalised Difference Vegetation Index and are very simplified in comparison with what I hope to do. I’d recommend taking a look at that article as I don’t think I can reproduce the images here for copyright reasons. (I also think Edinburgh cheats a bit in the ranking by including a chunk of the Pentland Hills within its boundary which one wouldn’t do in a serious analysis!).

Close up of EVI for the botanics and surrounding area. Note how bright green the playing fields are compared to the botanics. Also note the mosaic of gardens and buildings.

Landsat 8 data has a resolution of 30m. Each pixel is about 100 foot across. I’m hoping that at this resolution we’ll be able to produce a proxy metric for how green and area “feels” to people living and working there especially when using Google Street View analysis to paint a fuller picture. It is not going to be simple. As you can see large blocks of green represent the biodiversity deserts that are playing fields. I’d also consider these aesthetic deserts. The mosaics of different greens seen in the botanics, the allotments, Warriston Cemetery and St Marks Park tell a different story as do the built up areas of the new town. I’m excited to see where we can take this analysis.

Jan 112017

We instinctively know that a walk in the garden or somewhere else filled with natural beauty is good for us but it is difficult to justify expensive or restrictive planning decisions on the basis of instinct alone. This is why scientists have been trying for years to quantify just how exposure to green space improves our mental and physical health. They have managed to show that what we see out of the window or walk past on the street really does matters for our stress levels and it is particularly good if what we see is plants. But how do we extend these results beyond small studies based on a few volunteers?

As I’ve posted before there is an emerging field (Biophilomatics) which is using big data resources to look at these problems. This latest study from RBGE takes a new approach to the subject by plugging together some of the latest Google technologies in a novel way.

Computers are getting better at assessing the contents of our photos. If you use photo sharing sites you may have noticed that they use this technology to sort your photos by subject. Google has now given developers access to the algorithms behind their image classification systems through the Google Vision API. By combining this image recognition technology with Google Street View images from around Edinburgh we have demonstrated the possibility of assessing our cities streets for their perceived naturalness and potential restorative value using automatically sampled images.

Map of Edinburgh showing sample area and points. Greener dots are more natural. Redder halos are more deprived according to the Scottish Index of Multiple Deprivation. Where Street View panoramas are offset they are joined to sample points by lines.

It is very early days yet and before we can draw strong conclusions we need to develop it further but this could be an addition to the tool box for building more liveable cities.

The work was done as part of the Edinburgh Living Landscape (ELL) project which involves a number of partner organisations with interest and skill in our shared environment. ELL promotes improving, expanding and connecting up the green space in the city and encourages innovation in greening up built structures recognising that the local natural environment is a health asset. There are key questions about how best to achieve this for all sectors of the population and the methodology described in this paper contributes to building the evidence base for decision making.

Below is a slide show of the randomly sampled Street View images. Is this the Edinburgh you know?

Note: There are privacy and other issues around street view images and Google therefore remove any areas they are requested to. In the slide show above images are loaded directly from Google and so any that have been removed or replaced since the experiment was originally carried out in early 2016 will appear as unavailable.

Nov 292016
Three Landsat 8 Scenes covering central Scotland

Three Landsat 8 Scenes covering central Scotland

I’ve been looking at producing a good quality Normalized Difference Vegetation Index (NDVI) dataset for central Scotland so that I can investigate correlation between green space, biodiversity and well-being indicators. To do this I need access to satellite imagery.

Fortunately every sixteen days the Landsat 8 satellite passes over every part of the earth photographing at a resolution of 30 metres in eleven spectral bands. The U.S. Geological Survey generously make this data available for free and Amazon kindly distribute it as one of their public datasets. NDVI can be calculated by simply comparing the red and near infra-red bands of these images. It sounds very simple until we get into the detail.

Landsat 8 Panchromatic sample captured on 3rd June 2016 (note incomplete new Forth crossing)

Landsat 8 Panchromatic sample captured on 3rd June 2016 (note incomplete new Forth crossing)

NDVI is a measure of light absorbed by chlorophyll but much of the vegetation in Edinburgh and Glasgow that I am interested in is only actively growing for half the year. It is therefore important to compare images from different times of year. For consistency I’d like to restrict the data to the Landsat 8 satellite which entered into service in 2013 which further restricts the number of potential images. The Google Earth snapshot above shows the extent of the three Landsat scenes that cover the area of interest. Because we are so far North they overlap and so most of our area of interest is covered by at least two of the scenes and therefore imaged on average every eight days.  From mid 2013 until today this adds up to 187 images. More are being added all the time. This all seems good but we then run into the next problem.

If you are in central Scotland and look straight up the chances are you will see clouds. This isn’t just a stereotype of Scottish weather (or character?) but is supported by the data. The graph below shows the percentage of cloud cover in each of the 187 available images. The average cloud cover is 55%. For a crude analysis I may be able to select individual images from different years to represent winter and summer but if I want to have a more granular understanding of seasonal change, some indication of change through the years and some redundancy to correct for errors I’ll need to do something more sophisticated.

cloud_cover_2013-16To build my dataset I will therefore have to write scripts that examine all images created for these scenes, remove clouded areas and calculate NDVI for the remaining parts. These parts will then form a jigsaw puzzle that can be assembled in different ways to give seasonal or temporal change for different sub areas. The system should update automatically or semi-automatically as new scenes are released onto the Amazon hosting platform. Well that is plan anyway.

This all makes me very appreciative of the Google Earth data which, apparently effortlessly, does this process at higher resolution for visual satellite imagery, merging it with aerial photography as you zoom in.

Update: I just discovered the USGS service for surface reflectance corrected NDVI and EVI products – this may save some work and cut Amazon out of the loop.

Nov 012016


At the end of last month I spent a Thursday evening at IKEA Edinburgh, not for the usual reason of eating chips in the café with my daughter but to contribute to an event called Untangling Resilience to Depression as part of Midlothian Science Festival.

Recently we have been working with Stella Chan, a clinical psychologist from the Edinburgh University, as part of the Edinburgh Living Landscape project. Our discussions are around the effect connection with nature has on people’s well-being and the role that botanic gardens have in mediating those relationships.

The evening of talks at IKEA was part of the public outreach for the STRADL project lead by Prof Andrew McIntosh in which Stella is involved. STRADL is a multimillion pound, five year, Welcome Trust funded project that tries to understand the different ways in which people are resilient to depressive illnesses. There are many factors effecting depression ranging from genes through support networks out to the wider environment. The talks were arranged along this spectrum with me speaking last on green space and its affect on well-being. Speakers included Prof McIntosh, Prof Andrew Gumley from University of Glasgow, Prof Matthew Smith from University of Strathclyde and myself. Stella and Prof Stephen Lawrie acted as compères.

Stella asked me to speak because I had been describing some of my work in this area. I have been developing a mobile phone app called the Ten Breaths Map which aims to measure people’s engagement with natural spaces and been working on a paper that uses automated image categorisation to predict how restorative an image of a place is. This somewhat overlaps with Stella’s Project Soothe. (We’ve also been talking with Sarah Payne at Heriot Watt University about running experiments on the restorative value of the gardens.)

Stating that a walk in beautiful, natural surroundings might be good for mental and physical well-being seems so obvious as to not be worth investigating but in the context of rapid urbanisation where over half the world’s population and 80% plus of those in the developed countries  live in towns and cities the question of why a walk in the garden is better than a walk on a treadmill becomes more urgent. Coming from the other direction the expansion of urbanisation and associated agricultural intensification means there is less room for the biodiversity and those same urban green spaces become important as nature reserves. There is also evidence that exposure to green space makes people more pro-environmental and therefore more likely to support the lifestyle changes necessary to protect the planet in the face of threats like global warming. Research that leads to policy that helps us get our urban green spaces right is likely to have a big impact on future well-being of both humans and nature.


In fact there is good empirical evidence dating back to the 1960s that exposure to nature is beneficial for dealing with psycho-physiological effects of stress.  Just viewing images of nature has been shown to have a restorative effect. The two main theoretical models for why this happens are Stress Recovery Theory (SRT) and Attention Restoration Theory (ART). These theories are complementary. SRT is more concerned with physiological and negative affect whilst ART is concerned with attentional fatigue. Major questions that remain to be answered are: What is it about natural environments that produce these benefits and is this compatible with those same spaces acting as biodiversity refugia that nudge people to be more pro-environmental? Is it enough to just expose people to green space or is education, background and cultural meaning important? Is it equally beneficial for all or are there genetic, personality, gender or age factors?

I like to walk in the garden and intrinsically feel that my training in mindfulness techniques helps me connect with nature but what excites me professionally is that, in this age of big data, some of these questions are going to become much more computable. This is best illustrated with an example. A recent paper by James et al (2016) looked at how exposure to greenness around where you live may be linked to your chances of dying early. They combined two sources. The first dataset came from the Nurses’ Health Study, a project that has been tracking the health of nurses in the United States for the last forty years. These women fill in regular questionnaires about their health and lifestyle. The second dataset was Normalized Difference Vegetation Index (NDVI) calculated from satellite imagery. The study tracked 108,630 women over eight years during which 8,604 of them died. It looked at how much greenness there was within 250m and 1,250m of their homes and, because the study design was longitudinal and the health data is so comprehensive, it could confidently say: “Higher levels of green vegetation were associated with decreased mortality.”

The James et al study was only possible because two large, disparate datasets could be combined in ways their original authors probably never envisaged yet the results are good, highly relevant and potentially impactful. In a far, far less major way my recent studies have been using automated sampling and mapping techniques to try and infer human-nature connections. I’m reading similar studies and there is scope to do much more. My problem is how to describe what this is in less than the eight hundred words I’ve taken here. In conversation with a long suffering colleague we half jokingly came up with Biophilomatics. But many a good neologism is created in jest and this one is worth documenting.

There is a field called Bioinformatics  from bio- (Greek), information (Latin) and -matic (Greek) which has come to be associated with the computationally complex tasks associated with DNA and protein structure. The smaller scale bits of biology.

The field we generally work at the Botanics is more Biodiversity Informatics than bioinformatics. This is information at the taxonomic and systematic levels, the names of organisms, how they are related, where they occur and in what combinations. Whole organism stuff.

Biophilia is a term coined by E.O. Wilson in his 1984 book of the same name to mean ”love of life or living systems” or “the connections that human beings subconsciously seek with the rest of life”. It is the notion that we need connections with nature to thrive. It has spawned several movements notably Biophilic Design, bring a love of nature into architecture, and Biophilic Cities, integrating nature into urban planning.

The new field of Biophilomatics is a specialisation of Biodiversity Informatics in that it includes much of the same data used in describing the natural world but then combines it with data about human well-being. It is cross domain as it requires collaboration between the biodiversity and health worlds.

  • Formally: Measurement of the effects on human well-being and pro-environmental behaviour of the quality and quantity of connection with nature.
  • Informally: Describing the human love affair with nature.
  • Literally:  The willingness to perform (-matics) life (bio-) loving (philo).

Whether or not biophilomatics takes off as a new term the process of thinking it up has helped clarify, at least in my mind, what I’m working on. Thanks are due to Midlothian Science Festival and IKEA for hosting an evening that helped me through this process.