Google Bird: What Can Search Data Do for Bird Conservation?

big_bird_google_logoIf Google searches can so accurately predict the incidence and spread of the flu why not the ‘incidence and spread’ of birds?

eBird is only one of many mechanisms we have for amassing and disseminating big bird data (big data). Another, that potentially holds promise, is the Google Trends database. The google Trends site is the front end of a database containing nearly a decade’s worth of geospatial search history – that is, web browser search terms tagged with the time and geographic location of the search; essentially the when and where of the what that we search.

Still in relative infancy, Google Trends was maybe considered a bit of a gimmick until it delved into the world of epidemiology (without actually doing so) and announced its search trends database could be used to predict the spread of flu in the United States at an accuracy rivalling that of the US Centers for Disease Control (CDC).

The power of the flu-prediction model lies in the premise that when we, or those around us, contract the flu we go to the web for answers. Of course, we go to the web for answers about the flu other times of year, but most of us do so most frequently when the illness is looming or prevalent. And when big data is involved – thousands of searches over long time periods – the signal can greatly outweigh the noise and yield meaningful patterns. The bottom line is that in many cases our search behaviour directly reflects what we see and experience day to day, on the ground.

If this theory applies so strongly to contagious illness, why should it not also hold for other natural, geospatial phenomena that pique our curiosities and cause us to lay at the feet of the Google throne – can Google Trends yield any insight about the incidence (a.k.a. presence or abundance) and spread (a.k.a. movement) of birds? If so, what about other natural phenomena that incite us to mad fits of Google searching such as bird food (“how do i keep mosquitoes from carrying my kids away“, “what is this green caterpillar destroying my garden“)? If so, “wow!” and let’s get this working for bird research and conservation.

But first, what kind of data are we talking about here on the Google Trends site? In an armslength example, the first graph below shows search interest over time for the search term “gift ideas” (wait, this is a bird blog isn’t it?). In these graphs the search interest, the number of searches at each given time, is scaled so that 100 represents the highest search interest across the time period in question and 0 represents relatively few searches. Before looking at the graph we could make some predictions about when we might expect to see peaks based on when we think people would be most likely to Google “gift ideas”: how about Valentine’s day, Easter and Christmas. With some differences among regions, the trends data suggest that we are hopelessly-robotic consumers (at least the mass of the searchers are).

Screen shot 2013-09-11 at 8.33.24 PM

Let’s leave consumerism aside and delve into a more important and interesting topic – birds! If there was a strong correlation between the timing of bird phenomena and our bird-Googling behaviour, we might expect the Trends data to tell the story that we see unfold in front of us every year: migrant birds arrive in spring, build a nest, lay eggs, baby birds hatch and migrants migrate south for winter. Perhaps that would be asking too much of the database…but let’s see where we end up… (*sound of numbers crunching*)…

…Tada! The Trends data suggest that our searches about birds are, in fact, remarkably interlinked with the timing of bird phenomena. Relatively consistently over time we see peaks in “bird migration” (blue line) corresponding to March/April (spring migration), followed by annual peaks in “bird nests” (red line) in May/June, “bird egg” (yellow line) slightly later in May/June, “baby bird” (green line), and followed by another “bird migration” peak (blue line) in September/October (fall migration). The first graph below shows the consistency of these patterns (blue-red-yellow-green-blue) over long time periods (2004-preset) and the second graph shows more detail on the intra-annual patterns (2010-present).

Screen shot 2013-09-11 at 9.16.37 PM

Google Trends data for bird breeding-phenology-related terms for the United States from 2004 to present

Screen shot 2013-09-11 at 9.17.18 PM

Google Trends data for bird breeding-phenology-related terms for the United States from 2010 to present

As a Canadian reading this, several alarm bells are likely going off in your mind: i) Why is the peak for “bird nest” mainly in May? ii) why is there so much overlap between peaks of “bird nest”, and “bird egg” ” iii) Was there a mass hatching of baby chickadees in January 2012 or something, eh? The short answers, I suggest, are: i) we see broad peaks here since we are looking at search trends ranging from Texas to Maine and breeding phenology is highly dependent on latitude,  ii) the difference in timing between nest building and egg laying is only a few days in most temperate-breeding bird species and renesting during the season means that there is significant overlap in these two breeding stages during the season; and iii) the search terms used in this example are not exclusive to the world of birds and are subject to peaks from other types of searches (the News headlines box on the Trends site allows you to see timing of trending news stories – incidentally “the baby bird project” by Gavin Parsons peaked in Jan 2012).

So, what about other breeding-season phenomena – what about the relationship between baby birds and baby bird food? From theory and research we predict that to maximize chick growth and survival breeding activities should be timed such that baby birds hatch when baby bird food (for example, the highly nutritious caterpillar sandwich, or caterpillar “donair”, in Canada), is abundant. Recent studies have suggested that climate change might be disrupting this tight relationship in the timing of baby bird and peak insect emergence. Could these types of questions be assessed using Trends data? Below is the remarkable relationship between the search terms “baby bird” and “caterpillars” for the U.S., Canada, and the U.K since 2007:

Screen shot 2013-09-11 at 10.35.34 PM

Google Trends data for “baby bird” and “caterpillars” for the United States from 2007 to present

Screen shot 2013-09-11 at 10.35.53 PM

Google Trends data for “baby bird” and “caterpillars” for Canada from 2007 to present

Screen shot 2013-09-11 at 10.36.19 PM

Google Trends data for “baby bird” and “caterpillars” for the United Kingdom from 2007 to present

There are many interesting patterns evident in these graphs. First, there is a tight link between the peaks of these two search terms in all three geographic regions. Second, “baby bird” peaks often lag behind “caterpillar” peaks, but the relative timing differs by year. Third, the search term “caterpillars” seems to frequently have two peaks during each year.

The fact that the Trends results look so much like biological data, and look as though they are “reporting” biological phenomena, is exciting. On the surface, the data appear to be a powerful, promising tool for answering questions about bird lives at regional, or national, scales. Could Google Trends (gBird?) data be used to answer questions of arrival and distribution as clearly as direct eBird observation data? Could we monitor longterm activity of breeding birds and their food supplies?

A part of the answer lies in how well our search behaviour correlates with true timing of bird activities and for this we need to judge validity of trends data against field-collected bird data (stay tuned…). Why should we expect our search behaviour to so accurately reflect the timing of natural phenomena? I would argue that internet searching has begun to take on the role – for better or for worse – of our parents, teachers, colleagues and reference books. When presented with a question or mystery we might traditionally have taken our questions to these sources. When the woods began to reverberate with the songs of migratory warblers in spring, our curiosities might have led us to ask a teacher or natural history expert “what bird has a red crown?”, “how does bird migration occur?”, “what bird lines its nest with moss”. These days, it is clear that search engines are a whiteboard of some of our most instinctual thoughts and questions. But Google searchers also inherently bring biases: we are urbanized animals with attenuating attention spans, that respond to and think most about urgent things in our immediate environments; there is good reason to think that we would bring these biases to our searches as well.

Even if the Google Trends data prove not to be top-notch indicators of bird phenomena, exercizes like this can still help propel us towards novel hypotheses that we can rigorously test using big data. Like death and taxes, big data is a certainty of our future and the sooner we get it working for birds the better.

Screen shot 2013-09-12 at 12.48.54 AM

Google Trends data for “big data” for the United States, the United Kingdom and Canada from 2004 to present

This entry was posted in Bird Behaviour, Bird Canada, Bird Conservation Canada, Canadian Birds, Citizen Science, Conservation, Insects, Migration, Songbirds and tagged , , , , , , . Bookmark the permalink.

Comments are closed.