Efficient way to plot data on an irregular grid

Question

I work with satellite data organized on an irregular two-dimensional grid whose dimensions are scanline (along track dimension) and ground pixel (across track dimension). Latitu

Answer 1

I think data.table might be helpful here. The merging is almost instantly.

"5839200 rows merged in 1.24507117271423 seconds"

library(data.table)
pixel_cornersDT <- as.data.table(pixel_corners)
so2dfDT <- as.data.table(so2df)

setkey(pixel_cornersDT, id)
setkey(so2dfDT, id)

so2dfDT <- merge(pixel_cornersDT, so2dfDT, by=c("id"), all = TRUE)

Having the data in a data.table, the subsetting in the plotting function is also gonna be considerably faster.

---

• Questions 1 / 2 / 4:

I dont think the process would be faster with raster or sf but you can experiment with the functions rasterFromXYZ() or st_make_grid(). But most of the time will be spent on the conversion to raster/sf objects, as you would have to transform the whole dataset.

I would suggest doing all the data processing with data.table including the cropping, and from there you could switch to raster/sf objects for plotting purposes.

---

• Question 3:

The google plot shows correctly, but you have specified a black/white map, and its overlayed with the "raster", so you won't see a lot. You could change the basemap to a satellite-background.

base_map <- get_map(location = c(lon = (latlon[4]+latlon[3])/2, lat = (latlon[1]+latlon[2])/2), 
                    zoom = 7, maptype="satellite")

---

• Question 5:

You could use the rescale function from the scales package. I included two options below. The first (uncommented) takes the quantiles as breaks and the other breaks are individually-defined. I wouldn't use the log transformation (trans- argument) as you would create NA values, since you have negative values aswell.

ggplot(df_sub) + 
  geom_polygon(aes(y=lat_bounds, x=lon_bounds, fill=so2tc, group=id), alpha=0.8) +
  borders('world', xlim=range(df_sub$lon), ylim=range(df_sub$lat),
          colour='gray20', size=.2) +
  theme_light() + 
  theme(panel.ontop=TRUE, panel.background=element_blank()) +
  # scale_fill_distiller(palette='Spectral', type="seq", trans = "log2") +
  scale_fill_distiller(palette = "Spectral",
                       # values = scales::rescale(quantile(df_sub$so2tc), c(0,1))) +
                       values = scales::rescale(c(-3,0,1,5), c(0,1))) +
  coord_quickmap(xlim=c(latlon[3], latlon[4]), ylim=c(latlon[1], latlon[2])) +
  labs(title=title, subtitle=subtitle, 
       x="Longitude", y="Latitude", 
       fill=expression(DU)) 

The whole process takes now about 8 seconds for me, including the plotting without background-map, although the map rendering will also take additional 1-2 seconds.