Why are the British tweeting about climate change at night?

Points of post:

•  To demonstrate how the library twitteR can be combined with mapping capabilities of ggplot

• To reveal the locations of people tweeting about climate change at two points in time.

• To show that climate change tweeters are mostly from the developed world.

• To show that some (sad) tweeters are twittering about climate change in the middle of the night.

In a previous blog post, we used twitteR and wordcloud to summarize which other words were occurring in tweets (combined from around the world) together with the keywords, ‘climate-change’ and ‘Bangladesh’ at two arbitrarily selected time-points. 

The geo-locations of tweets are, however, also often available and are potentially very interesting and revealing. 

Fifteen-hundred (1500) is the maximum number of tweets that can be captured using the twitteR library with one call.  There probably are ways to get more data but we guess you probably have then to spend money. 

Here we used #climatechange because it coincided with the last days of COP20 in Lima, Peru which ran from 1^st to 12^th December 2014.  It is an extremely important global forum at which nations can meet and discuss their options for reducing carbon emissions, see http://unfccc.int/meetings/lima_dec_2014/meeting/8141.php)

The first ‘tweet map’ (below) we produced is based on approximately 1500 geo-located tweets that contained the hash-tag, #climatechange, and which were ‘tweeted’ at about 10am GMT on the 11^th December 2014. It shows that #climatechange tweets were coming from 4 main areas: North America, Europe, India and Australia. There didn’t appear either to be too many tweets coming out of Lima which surprised us. Maybe the delegates were too busy enjoying the South-American hospitality, and catching up with old mates to take much interest in Climate Change!

Geo-located tweets with #climatechange tweeted at around 10am GMT 11th December 2014

The second ‘tweet-map’ (below), also based on approximately 1500 geo-located #climatechange tweets, is for a snapshot that took place at 5 hours later at around 3am GMT on the final day of the conference (12^th December 2014).  The overall pattern between the maps remains the same but the relative frequency of #climatechange tweeters from Europe, as compared to North America, has increased. People in the United Kingdom were particularly keen, twittering like mad about climate change at 3am. Why? We don’t know.

Geo-located tweets with #climatechange tweeted at around 3am GMT 12th December 2014

Note that tweets are geo-located, either by exploiting the users’ location as defined in their profile, or by ascertaining the exact location directly  if allowed by the user. This can be effected, either from GPS-enabled software which many people have installed on their smart-phones, or by using an IP-address.  This means that not all tweets can be geo-located with any great precision. Some are only geo-located at the National and/or regional levels, as evident from the large circle in the middle of Australia. That’s to say these cautious tweeters only gave ‘Australia’ as their location.

As we have explained in an earlier blog post on worldclouds and twitteR, to pull the data from Twitter using its API, you will need to have a Twitter account and carry out a 'twitter authentication'.  The R code to perform a search on twitter for the selected 'term(s)' and mapping them out is detailed below.

# Load required libraries
 
library(RCurl)
library(maps)
library(stringr)
library(tm)
library(twitteR)
library(streamR)
library(grid)
library(ggplot2)
library(rgdal)
library(ggmap)
 
# Set working directory
 
setwd("D:/ClimData/")
 
#### Fonts on Windows ####
windowsFonts(ClearSans="TT Clear Sans")
 
# Load Credentials
 
load("D:/ClimData/Twitter/twitter authentification.Rdata")
registerTwitterOAuth(twitCred)
options(RCurlOptions = list(cainfo = system.file("CurlSSL", "cacert.pem", package = "RCurl")))
 
# Search term on twitter
 
searchTerm <- "#climatechange"
searchResults <- searchTwitter(searchTerm,n=1500,since='2014-12-11', until='2014-12-12')  
tweetFrame <- twListToDF(searchResults) 
 
userInfo <- lookupUsers(tweetFrame$screenName)  
userFrame <- twListToDF(userInfo)
 
locatedUsers <- !is.na(userFrame$location)
 
# Geocode locations using 'ggpmap' library
 
locations <- geocode(userFrame$location[locatedUsers])
 
locations_robin <- project(as.matrix(locations), "+proj=robin")
 
locations_robin_df <- as.data.frame(locations_robin)
 
# Import world boundaries
 
world <- readOGR(dsn="D:/Data/ne_10m_admin_0_countries", layer="ne_10m_admin_0_countries")
 
world_robin <- spTransform(world, CRS("+proj=robin"))
 
world_robin_df <- fortify(world_robin)
 
counts <- aggregate(locations_robin_df$V1,by=list(x=locations_robin_df$V1,y=locations_robin_df$V2),length)
names(counts)[3] <- "count"
 
# Theme options for Map
 
theme_opts <- list(theme(panel.grid.minor = element_blank(),
                         panel.grid.major = element_blank(),
                         panel.background = element_blank(),
                         panel.border = element_blank(),
                         plot.background = element_blank(),
                         axis.line = element_blank(),
                         axis.text.x = element_blank(),
                         axis.text.y = element_blank(),
                         axis.ticks = element_blank(),
                         axis.title.x = element_blank(),
                         axis.title.y = element_blank(),
                         legend.position = "bottom",
                         legend.key = element_blank(),
                         legend.title = element_text(colour="black", size=12, face="bold",family="Clear Sans"),
                         legend.text = element_text(colour="black", size=10, face="bold",family="Clear Sans"),
                         plot.title = element_text(size=15,face="bold",lineheight=0.5,family="Clear Sans")))
 
# Plot map and tweet counts 
 
tp <- ggplot(world_robin_df)+
      geom_polygon(aes(x = long, y = lat, group = group), fill = "grey20")+
      geom_path(aes(x = long, y = lat, group = group),colour = "grey40", lwd = 0.2)+
      geom_point(data= counts, aes(x=x,y=y,size=count),color="#32caf6", alpha=I(8/10))+
      scale_size_continuous(name="Number of tweets")+
      ggtitle("Twitter Map of #climatechange\n")+
      xlab("")+ ylab("")+
      coord_equal()+
      theme_bw() + 
      guides(size = guide_legend(title.position = "top",title.hjust =0.5))+
      theme_opts
 
tp
 
# Save to png
 
ggsave(tp,file="D:/Twitter_ClimateChange_Map.png",dpi=500,w=10,h=6,unit="in",type="cairo-png")

Created by Pretty R at inside-R.org

13 decades on...has Penang gotten warmer?

• Here we compare recent temperature data for Penang with a report written in 1885.
• We find that air temperatures are higher now in the afternoon and evening
• Morning air temperatures seem to be about the same now as they were 130 years ago.
• There are sources of bias which may affect the interpretation, but about which we can do nothing.

Last week we received the email below from Mike Gibby (mikegibby@gmail.com).

“Hi Jason, Doug

I found your website / blog on Temperature Change in Penang while searching for info on rain, and more especially on rain gauges in Penang, and wondered if you could give any advice or help?

I'm part of a team who are researching for a guide to hiking trails in Penang; most of these trails were developed by farmers for access, but there are some trails that only 'stitch together' the sites of rain gauges, so they must have been created for data collection in the old, pre-digital days. An example here would be what is described as 'the longest trail in Penang', that runs from Telok Bahang to Penang Hill, via Bukit Laksamana.

So, my question is, can you point me at any sources of information, quite probably of the colonial era, on the setting up, siting etc of rain gauges?

In case you don't know it, I attach an article on the climate in Penang in 1885.

I look forward to hearing from you- best wishes !

Mike Gibby”

We replied to Mike that we didn’t know anything about the rain gauges or the old-colonial data collection protocols or policies.

We were, however, very interested in the 1885 weather report that Mike attached. We’ve since discovered that the report is available from the Malaysian Branch of the Royal Asiatic Society.  The 1885 report was compiled by Mr Thomas Irvine Rowell (Principal Civil Medical Officer) in Singapore in early 1886, about the same time Karl Benz patented the first gasoline-driven automobile and covers Singapore, Penang, ‘Province Wellesley’ and Malacca. 

Given our interest in Penang’s current and past weather, we decided to compare recent air temperature measurements with those taken 130 years previously in 1885 (please see below).

In the figure below we’ve plotted annual patters for six different Temperature parameters (mean monthly, maximum monthly, minimum monthly, 9am, 3pm, & 9pm) for both 1885 (blue lines) and the period 2003-2013 (red lines).  Monthly mean temperatures are around 28°C. March and April Penang average temperatures appeared to be a little hotter in 1885 than we’ve seen more recently, but for the rest of the year air Temperatures were fairly similar.  Recent maximum air temperatures are higher than they were in 1885 and minimum temperatures lower. We think, however, that this is probably caused by confounding/bias, ie. in 1885 data were only collected 3 times daily (9am, 3pm, & 9pm) whereas nowadays we can make observations, continuously, at 5 minute intervals and publish them instantly around the world! 

Given such problems of bias, the most meaningful comparisons we can make between these datasets are those observations taken at 9am, 3pm, and 9pm.  At 9am, temperatures in 1885 were similar to current measurements.  In the afternoon (3pm, typically around the hottest time of day in Penang) and evening (9pm), however, Temperatures are higher now in all months by about 1°C. 

So, in conclusion, if we assume that the measuring equipment is accurate, current air temperatures are around 1°C warmer in Penang in the afternoons and evenings, while Mr Rowell would probably have experienced very similar morning temperatures to us now!

These differences are less pronounced than we’d expected.

[Note: We’ve not addressed the potentially serious issue of ‘spatial bias’ and the locations that weather observations were made by us are not the same as those used by Mr Rowell].

The R code we used to produce the above line plots is outlined below and the 2003-2013 temperatures were downloaded from the NOAA NCDC's Global Summary of the Day (GSOD).

# Load required libraries
 
library(zoo)
library(lubridate)
library(plyr)
library(ggplot2)
library(grid)
library(XLConnect)
library(reshape2)
library(extrafont)
 
# Set working directory
 
setwd("D:/ClimData")
 
# Read table of 1885 Temperatures
 
temp1885<-readWorksheet(loadWorkbook("1885_Temps.xlsx"),sheet=1)
 
temps1885 <- temp1885[c(1,2,10,11,12,13,14,15)]
 
colnames(temps1885) <- c("month","monthf","temp_9am","temp_3pm","temp_9pm","mon_mean","mon_max","mon_min")
 
temps1885$year <- "1885"
 
temps1885_df <- temps1885[c(1,2,9,3:8)]
 
 
# Read data from NCDC
 
dat<-read.table("2110827004508dat.txt",header=TRUE,fill=TRUE,na.strings=c("*","**","***","****","*****","******","0.00T*****"))
 
colnames(dat)<-tolower(colnames(dat))
 
# Convert hourly data from UTC to local time zone
 
Sys.setenv(TZ = "UTC")
dat$dates <- as.POSIXct(strptime(dat$yr..modahrmn,format="%Y%m%d%H%M"))  + 8 * 60 * 60
 
# Convert temperatures in Degree Fahrenheit to Degree Celcius
 
dat$tempc <- (dat$temp-32) * (5/9)
 
dat$tempc[dat$tempc<=10] <- NA
dat$tempc[dat$tempc>=40] <- NA 
 
# Extract times and dates and reformat data
 
dat$year <- as.numeric(as.POSIXlt(dat$dates)$year+1900)
dat$month <- as.numeric(as.POSIXlt(dat$dates)$mon+1)
dat$monthf <- factor(dat$month,levels=as.character(1:12),labels=c("Jan","Feb","Mar","Apr","May","Jun","Jul","Aug","Sep","Oct","Nov","Dec"),ordered=TRUE)
dat$jday <- yday(dat$dates)
dat$hour <- as.numeric(format(strptime(dat$dates, format = "%Y-%m-%d %H:%M"),format = "%H"))
dat$minute <- as.numeric(format(strptime(dat$dates, format = "%Y-%m-%d %H:%M"),format = "%M"))
 
dat_01 <- subset(dat, year >= 2003 & year <= 2013 )
 
dat_02 <- dat_01[c(34,35,36,37,38,39,40,41)]
 
dat_9am <- subset(dat_02,hour == 9 & minute == 0 )
dat_3pm <- subset(dat_02,hour == 15 & minute == 0 )
dat_9pm <- subset(dat_02,hour == 21 & minute == 0 )
 
dat_9am_melt<- ddply(dat_9am, .(month,monthf), summarize, temp_9am=mean(tempc,na.rm=T))
dat_3pm_melt<- ddply(dat_3pm, .(month,monthf), summarize, temp_3pm=mean(tempc,na.rm=T))
dat_9pm_melt<- ddply(dat_9pm, .(month,monthf), summarize, temp_9pm=mean(tempc,na.rm=T))
 
dat_hourly <- cbind(dat_9am_melt,dat_3pm_melt[3],dat_9pm_melt[3])
 
dat_melt<- ddply(dat_01, .(month,monthf), summarize, mon_mean=mean(tempc,na.rm=T),mon_max=max(tempc,na.rm=T),mon_min=min(tempc,na.rm=T))
 
dat_yrs <- cbind(dat_melt,dat_hourly[-c(1,2)]) 
 
dat_yrs$year <- "10 Year Average (2003-2013)"
 
tempcomb <- rbind(dat_yrs,temps1885_df)
 
tempcomb_df <- melt(tempcomb, id.vars=c("month","monthf","year"),variable.name="type",value.name="temps")
 
# Plot data
 
p <- ggplot(data=tempcomb_df, aes(x=monthf,temps, colour=year))+
     geom_point(data=subset(tempcomb_df,year!=1885),size=3, alpha=0.5,aes(group=year))+ 
     geom_line(data=subset(tempcomb_df,year!=1885),size=1, alpha=0.5,aes(group=year))+
     geom_point(data=subset(tempcomb_df,year==1885),size=3, alpha=0.5,aes(group=year))+ 
     geom_line(data=subset(tempcomb_df,year==1885),size=1, alpha=0.5,aes(group=year))+
     geom_text(aes(label=round(temps,1)),hjust=1, vjust=-1,size=2.5)+
     facet_wrap(~type)+
     xlab("")+ ylab ("Temperature (Degree C)\n") +
     labs(title="Comparison of year 1885 and 10 year monthly average (2003-2013) temperatures in Penang\n")+
     theme(panel.background = element_rect(fill= "transparent"),
     plot.title = element_text(lineheight=1.2, face="bold",size = 13, colour = "grey20",family="Clear Sans"),
     panel.border = element_rect(colour = "grey90",fill=F,size=0.5),
     panel.grid.major.x = element_blank(),panel.grid.minor = element_blank(),
     axis.line=element_blank(),axis.title.x=element_blank(),
     axis.text.x=element_text(face="bold",size = 10, colour = "grey20",family="Clear Sans"),
     axis.title.y=element_text(face="bold",size = 10, colour = "grey20",family="Clear Sans"),
     axis.text.y=element_text(face="bold",size = 8, colour = "grey20",family="Clear Sans"),
     legend.title = element_text(colour="black", size=12, face="bold",family="Clear Sans"),
     legend.text = element_text(colour="black", size=10, family="Clear Sans"),
     legend.position="bottom",
     axis.ticks=element_line(size=0.5,colour="grey20"),panel.grid.major.y = element_line(colour="grey90",linetype="longdash",size=1))+
     guides(color = guide_legend(title = "Year", title.position = "top"))
 
p
 
 
# Save plot to png
 
ggsave(p,file="PenangTemps_1885_10yrAverage.png",dpi=500,w=12,h=8,unit="in",type="cairo-png")

Created by Pretty R at inside-R.org

Twittering about climate change and Bangladesh

Points of post:

• To describe how twitteR can be used, together with Word Clouds (wordcloud), to reveal what people are twittering about.
• To show how rapidly interest in various subjects can change.

Twitter (https://twitter.com/) is a social media network allowing users to send short (140 characters) messages called ‘tweets’. Twitter started up in 2006 and has had phenomenal success with 340 million tweets sent each day in 2012. 

In an earlier post on our blog, we described how archives of Google internet searches can tell us what people are interested in, and how these interests change over time. It is also possible to assess what factors might be affecting this interest, e.g. a disease outbreak or climatic disaster. Similarly Twitter might be viewed as a massive, passive opinion poll. 

Again wonderful R has libraries to analyze the worlds’ tweets (twitteR) which can be useful to perform social media mining and to reveal what people are thinking (twittering) about; whether it’s Brad Pitts’ facial hair (http://www.dailymail.co.uk/tvshowbiz/article-2792839/brad-pitt-continues-wear-rakish-moustache-long-sideburns-grew-sea-new-york.html) or climate variability and change. 

Jason and I are very interested in how climate change will affect the people of Bangladesh, in particular, and we’ve already explored the temporal and spatial dependence of various weather variables such as temperature and sea-level in the country. 

Hence we experimented by inserting the two keywords, ‘climate-change’ and ‘Bangladesh’ as our search parameters into twitteR. Our aim was to reveal with what other words or phrases these keywords were associated. ‘Word Clouds’ are a useful way to summarize the results. In Word cloud graphical representations, font sizes are scaled according to the frequencies at which words occur; in this case how often they are ‘tweeted’. Below is a Word cloud for September 2014 which shows that ‘URBANISATION’ and ‘ROUND TABLE’ occurred most often in tweets.

By December 2014, however, when we tried again, the identical keywords were associated with ‘FLOATING GARDENS’, and the increasingly nebulous ‘ADAPTATION’ and ‘RESILIENCE’.

Clearly social media such as Twitter, combined with R’s software packages and plotting facilities, are fantastic tools for understanding the often capricious winds affecting the movement of global public interest and opinion.

To carry out your own social media mining using the twitteR package, you will first need to have an account on twitter and create an 'app'. Please see this and this link to get a better idea on how to get that done.

You will also find information in the above links on the 'twitter authentication' process which is also required to be able to perform the keyword searches using twitteR.

The code used to produce the above word clouds is outlined below.

# Load required libraries

library(RCurl)
library(stringr)
library(tm)
library(wordcloud)
library(RColorBrewer)
library(twitteR)
library(streamR)
library(grid)
library(ggplot2)
library(wesanderson)
 
# Load credentials
load("D:/ClimData/Twitter/twitter authentification.Rdata")
registerTwitterOAuth(twitCred)
options(RCurlOptions = list(cainfo = system.file("CurlSSL", "cacert.pem", package = "RCurl")))
 
tweets <- searchTwitter("Climate Change Bangladesh", n=1500, lang="en") 
 
tweets.text <- sapply(tweets, function(x) x$getText())
 
# Remove non alphanumeric characters
tweets.text <- gsub("[^a-zA-Z0-9 ]","",tweets.text)
 
# Convert all text to lower case
tweets.text <- tolower(tweets.text)
 
# Replace blank space (“rt”)
tweets.text <- gsub("rt", "", tweets.text)
 
# Replace @UserName
tweets.text <- gsub("@\\w+", "", tweets.text)
 
# Remove punctuation
tweets.text <- gsub("[[:punct:]]", "", tweets.text)
 
# Remove links
tweets.text <- gsub("http\\w+", "", tweets.text)
 
# Remove tabs
tweets.text <- gsub("[ |\t]{2,}", "", tweets.text)
 
# Remove blank spaces at the beginning
tweets.text <- gsub("^ ", "", tweets.text)
 
# Remove blank spaces at the end
tweets.text <- gsub(" $", "", tweets.text)
 
# remove word amp
tweet.text <- gsub("amp","",tweets.text)
 
# Create corpus
tweets.text.corpus <- Corpus(VectorSource(tweets.text))
 
# Clean up by removing stop words
tweets.text.corpus <- tm_map(tweets.text.corpus, function(x)removeWords(x,stopwords()))
 
# Create document term matrix applying some transformations
tdm = TermDocumentMatrix(tweets.text.corpus,
      control = list(removePunctuation = TRUE,
      stopwords = c("climate change","bangladesh","climate", "change"), stopwords("english"),
      removeNumbers = TRUE, tolower = TRUE))
 
# Define tdm as matrix
m = as.matrix(tdm)
 
# Get word counts in decreasing order
word_freqs = sort(rowSums(m), decreasing=TRUE) 
 
# Create a data frame with words and their frequencies
dm = data.frame(word=names(word_freqs), freq=word_freqs)
 
# Plot and save the image in png format
png("BGD_ClimateChange_Dec2014.png", width=5, height=5, units="in", res=500)
 
wordcloud(dm$word, dm$freq, random.order=FALSE, min.freq = 2,scale=c(4,0.5),max.words = 100,colors=wes.palette(5,"Darjeeling"))
 
dev.off()

Created by Pretty R at inside-R.org

Trending climate-related keywords with GTrendsR

Key points of post

• Google stores data on millions of internet searches throughout the world and these data are publicly available. 
• Google trends (http://www.google.com/trends/?hl=en-GB)’ is a fascinating system that summarizes information from all these searches revealing what subjects are ‘trending’ worldwide.
• Conveniently R has a library (GTrendsR) capable of extracting, analyzing and plotting these data. 
• Here we explore time series data (2004-2014) summarizing Google searches for the keywords, ‘climate-change’, ‘global-warming’, ‘ocean-acidification’ and ‘sea-level rise’. 
• All four series have trends and have either unimodal or bimodal seasonal cycles.

The other day Jason decided to play around (nerdy I know) with R’s relatively new GTrendR library and analyze the time-series statistics for the following four key words: ‘climate change’, ‘ocean acidification’, ‘global warming’, and ‘sea-level rise’. The results are displayed below for the last decade (details of how Google calculates these statistics are available here: https://support.google.com/trends/answer/4355164?hl=en). 

First let’s talk about what has happened to the ‘popularity’ of these searches over the long-term, ie. between 2004 and 2014. Searches for ‘climate change’ were fairly static between 2004 and 2008 and have since increased in popularity. ‘Global warming’ was similarly stable but then fell. In the past we think that ‘global warming’ was the more commonly used term to describe ‘anthropogenic heating’. As the wider consequences of increasing temperatures for the entire global weather and climate system became increasingly more recognized, however, it became clear that a more general term was needed and hence, ‘climate-change’.

Searches for ‘ocean-acidification (OA)’ – a product of CO₂ pollution – and also known as the ‘ugly twin’ of climate-change have not changed substantially over the last decade or so, apart from a massive spike in mid-2013. My pal Prof. Hall-Spencer in Plymouth suggested the spike might be due to interest around this article (http://www.nbcnews.com/video/ann-curry-reports/54882960#54882960), and/or the fact that ocean acidification started getting discussed a lot in Washington DC in mid 2012. Furthermore some charitable organizations advertised research funding calls for OA research at that time.

Interest in sea-level rise (SLR) started to increase in 2011 and we have no idea why. Suggestions welcome!

The time-series displayed here are all available at weekly resolution. What particularly interests us in our time-series work is seasonality. Both the ocean-acidification and sea-level rise series are strongly seasonal with a single peak each year (unimodal) occurring between June and July. 

What causes this seasonality? Is it related to educational cycles in the northern hemisphere dictating when students undertake science related projects? Are there more climate change conferences in the northern hemisphere summer? Or perhaps it’s because there’s less football on TV distracting people from more pressing concerns? 

The other two series (‘climate change and ‘global warming’) are also seasonal but are different in having two peaks (bimodal). This is difficult to see in the plot above so we plotted the same data on another graph for a shorter time-period (2010-2014 below) which shows clearly that searches for climate-change most popular in March and October each year with troughs in January and July ? Does anyone have any idea why ? 

Could it be a confounding between seasonal educational cycles in the northern hemispheres (NH) and southern hemispheres (SH)? The argument might go like this: (i) Numbers of internet searches are far greater in developed countries; (ii) ‘The majority of developed countries are in the northern hemisphere; (iii) ‘Climate-change’ and ‘global-warming’ are much more popularly searched than the other two terms; (iv) The overwhelming predominance of northern-hemisphere searches somehow ‘drowns-out’ any seasonal signal from the south; and (v) The bimodal seasonal signals we see for OA and SLR are a combination of separate unimodal cycles from the NH and SHs. 

Clearly we need to subset and examine these data by location. This will form the subject of future blog articles.

In the meantime please feel free to suggest any alternative explanations.

As usual the R-code describing the extraction and plotting of the data is outlined below.

# Load required libraries
 
library(gtrend)
library(dplyr)
library(ggplot2)
library(scales)
library(ggthemes)
 
# Define terms to use for trend searches
 
terms <- c("Climate Change","Ocean Acidification","Sea Level Rise","Global Warming")
 
out <- gtrend_scraper("youremail@gmail.com", "yourpassword", terms) ## replace with your own google username and password
 
out %>%  trend2long() %>% plot() 
 
# Get plot of trends
 
a <- out %>%
     trend2long() %>%
     ggplot(aes(x=start, y=trend, color=term)) +
     xlab("\nYear") + ylab("Trend\n")+
     geom_line() + theme_stata()+
     facet_wrap(~term) + ggtitle("Phrase Search Trends on Google\n")+
     guides(color=FALSE)
 
a
 
# Save file to png
 
ggsave(a,file="GoogleTrend_Climate.png",dpi=500,w=10,h=6,unit="in",type="cairo-png")
 
 
# Extract just trends of 'climate change' search for years 2010 to 2014 and plot
# to observe seasonality
 
 
dat <- out[[1]][["trend"]]
colnames(dat)[3] <- "trend"
 
dat2 <- dat[dat[["start"]] > as.Date("2010-01-01"),]
 
rects <- dat2  %>% 
mutate(year=format(as.Date(start), "%y")) %>%
group_by(year) %>%
summarize(xstart = as.Date(min(start)), xend = as.Date(max(end)))
 
c <- ggplot() +
     geom_rect(data = rects, aes(xmin = xstart, xmax = xend, ymin = -Inf, 
     ymax = Inf, fill = factor(year)), alpha = 0.4) + theme_bw()+
     ylab ("Trend\n") + xlab("\nDate")+
     ggtitle("Search Trends of 'Climate Change' on Google (2010-2014)\n")+
     geom_line(data=dat2, aes(x=start, y=trend), size=0.75,colour="grey20",type="dashed") + 
     geom_point(data=dat2, aes(x=start, y=trend), size=1.25,colour="grey20") + 
     scale_x_date(labels = date_format("%b-%Y"), 
     breaks = date_breaks("3 months"),
     expand = c(0,0), 
     limits = c(as.Date("2010-01-01"), as.Date("2014-12-31"))) +
     stat_smooth(data=dat2,aes(x=start, y=trend), col="blue",method = "loess",span=0.1,se=T,size=1,linetype='twodash',alpha=0.5,fill="grey60")+
     scale_fill_discrete(guide = FALSE)+
     theme(axis.text.x = element_text(angle = -30, hjust =0,size=10),
     plot.title = element_text(lineheight=1.2, face="bold",size = 14, colour = "grey20"),
     panel.border = element_rect(colour = "black",fill=F,size=0.5),
     panel.grid.major = element_line(colour = "grey",size=0.25,linetype='longdash'),
     panel.grid.minor = element_blank(),
     axis.title.y=element_text(size=10,colour="grey20"),
     axis.title.x=element_text(size=10,colour="grey20"),
     axis.text.y=element_text(size=10,colour="grey20"),
     panel.background = element_rect(fill = NA,colour = "black"))
 
c
 
# Save plot to file
 
ggsave(c,file="GoogleTrend_ClimateChange2010-14.png",dpi=500,w=10,h=6,unit="in",type="cairo-png")

Created by Pretty R at inside-R.org