Analysis of Historical Togolese Temperatures

This repository serves as a home for an analysis of historical monthly minimum and maximum temperatures for 10 cities in the west African nation of Togo. The data were obtained from Togolese meteorologists, and the cities include: Lomé, Sokodé, Kara, Atakpamé, Mango, Dapaong, Niamtougou, Sotouboua, Tabligbo, and Kouma Konda. The project project is organized as follows:

• R: collection of R scripts responsible for cleaning and imputing the raw data, applying Singular Spectrum Analysis (SSA) to each city individually, and development of global models.
• Figures: all of the figures generated by the R code. Mostly diagnostic plots from the SSA phase where seasonal components were manually selected.
• Presentations: a collection of presentations given at different conferences by myself and/or Professor Andy Long

Map of Togo

The main component of the analysis is the identification of important seasonal trends in both sets of temperature data. This is done manually using SSA, and the resulting periods are then incorporated into linear mixed models which also include a random intercept for location and geographical parameters (elevation, longitude, and latitude).

For this project I use the Rssa package, though I encourage you to visit my website for a walkthrough of how one could conduct such an analysis from scratch.

Global Models

Here is quick summary of the findings for the global models:

Global Minimum Model Fit

As far as minimum monthly average temperatures go, we see an overall average increasing trend of approximately .032 degrees Celsius per year. Some cities, such as Mango, have much greater variability in temperatures, but the trend persists. In other cities like Niamtougou, observed temperatures do not appear to be rising as dramatically, but we still expect an increase due to the overall global trend.

Global Maximum Model Fit

In the average monthly maximum temperatures, the increasing trend of 0.019 degrees Celsius per year is not quite as alarming as the trend in the minima, but worrisome nonetheless. Once again, there is heterogeneity across cities, but the global trend persist throughout most and the singular spectrum analysis appears to have adequately captured seasonal trends.

3D Visualization of the impact of Geography

Here I’ve included an interactive 3D visualization of the impact of latitude and longitude on both models. While I cannot share the data, feel free to download the models from the Output folder and run the following code to create your own versions of these to play around with.

max_model <- readr::read_rds(here::here('Output','max_model.rds'))
min_model <- readr::read_rds(here::here('Output','min_model.rds'))

# Minimum temperature model
min_geo_coef <- coef(summary(min_model))[c('Longitude','Latitude', 
                                           'I(Latitude^2)','I(Longitude^2)',
                                           'I(Latitude * Longitude)'),
                                         'Estimate']
min_geo <- function(long, lat){
  cbind(long, lat, lat^2, long^2, long*lat) %*% min_geo_coef
}
rgl::open3d()

## null 
##    6

rgl::plot3d(min_geo, 
            col = colorRampPalette(c("blue", "white", "red")), 
            xlab = "longitude", ylab = "latitude", zlab = "GeoMinTrend", 
            xlim = c(0,2), ylim = c(6,12),
            aspect = c(1, 1, 0.5))

# Maximum temperature model
max_geo_coef <- coef(summary(max_model))[c('Longitude','Latitude', 
                                           'I(Latitude^2)','I(Longitude^2)',
                                           'I(Latitude * Longitude)'),
                                         'Estimate']

max_geo <- function(long, lat){
  cbind(long, lat, lat^2, long^2, long*lat) %*% max_geo_coef
}

rgl::open3d()

## null 
##   10

rgl::plot3d(max_geo, 
            col = colorRampPalette(c("blue", "white", "red")), 
            xlab = "Longitude", ylab = "Latitude", zlab = "GeoMaxTrend", 
            xlim = c(0,2), ylim = c(6,12),
            aspect = c(1, 1, 0.5))