Skip to contents

CRU TS vs ERA5: India temperature comparison (2000-2018)

Source: vignettes/cru-ts-comparison.Rmd
cru-ts-comparison.Rmd

How well do ERA5 and CRU TS agree on monthly temperature over India? They are two independent records covering 2000-2018:

  • ERA5 - ECMWF reanalysis at 0.25°, pulled with era5ify_bbox
  • CRU TS v4.07 - station-interpolated observations from UEA at 0.5°, pulled with cru_ts_bbox

CRU TS variable dictionary with ERA5 equivalents:

list_cru_ts_variables() |>
  select(variable, name, unit, era5_equivalent)

Download 

# ERA5 monthly 2m_temperature for India
era5_tmp <- era5ify_bbox(
  request_id = "india_era5_tmp_monthly",
  variables = "2m_temperature",
  start_date = "2000-01-01", end_date = "2018-12-31",
  north = 37, south = 6, east = 98, west = 68,
  frequency = "monthly", resolution = 0.5
)

# CRU TS monthly mean temperature for India
cru_tmp <- cru_ts_bbox(
  variable = "tmp",
  start_year = 2000, end_year = 2018,
  north = 37, south = 6, east = 98, west = 68
)
data_path <- system.file("extdata", "india_era5_cru_comparison.rds",
  package = "varunayan"
)
if (!nzchar(data_path)) {
  knitr::opts_chunk$set(eval = FALSE)
} else {
  saved <- readRDS(data_path)
  era5_tmp <- saved$era5_tmp
  cru_tmp <- saved$cru_tmp
}

Monthly time series

Spatial mean temperature across all India grid cells, month by month:

era5_monthly <- era5_tmp |>
  group_by(year, month) |>
  summarise(temp = mean(value, na.rm = TRUE), .groups = "drop") |>
  mutate(source = "ERA5", date = as.Date(sprintf("%d-%02d-01", year, month)))

cru_monthly <- cru_tmp |>
  group_by(year, month) |>
  summarise(temp = mean(value, na.rm = TRUE), .groups = "drop") |>
  mutate(source = "CRU TS", date = as.Date(sprintf("%d-%02d-01", year, month)))

temp_all <- bind_rows(era5_monthly, cru_monthly)

ggplot(temp_all, aes(x = date, y = temp, color = source)) +
  geom_line(alpha = 0.5, linewidth = 0.5) +
  geom_smooth(method = "loess", span = 0.07, se = FALSE, linewidth = 1.1) +
  scale_x_date(date_breaks = "2 years", date_labels = "%Y") +
  scale_color_manual(values = c("ERA5" = "#E74C3C", "CRU TS" = "#2E86AB")) +
  labs(
    x = NULL, y = "Mean temperature (\u00b0C)", color = NULL,
    title = "India: Monthly Mean Temperature — ERA5 vs CRU TS (2000-2018)",
    subtitle = "Spatial mean across all grid cells"
  ) +
  theme_minimal(base_size = 11) +
  theme(legend.position = "top", panel.grid.minor = element_blank())

Comparison 

temp_wide <- inner_join(
  era5_monthly |> select(year, month, ERA5 = temp),
  cru_monthly |> select(year, month, `CRU TS` = temp),
  by = c("year", "month")
)

cor_val <- cor(temp_wide$ERA5, temp_wide$`CRU TS`, use = "complete.obs")
bias <- mean(temp_wide$ERA5 - temp_wide$`CRU TS`, na.rm = TRUE)

ggplot(temp_wide, aes(x = `CRU TS`, y = ERA5)) +
  geom_abline(slope = 1, intercept = 0, linetype = "dashed", color = "grey40") +
  geom_point(alpha = 0.5, size = 1.8, color = "#34495E") +
  geom_smooth(
    method = "lm", se = TRUE,
    color = "#E74C3C", fill = "#E74C3C", alpha = 0.15
  ) +
  coord_fixed() +
  labs(
    x = "CRU TS (\u00b0C)", y = "ERA5 (\u00b0C)",
    title = "ERA5 vs CRU TS: Spatial Mean Monthly Temperature",
    subtitle = sprintf("r = %.3f   bias (ERA5 \u2212 CRU TS) = %+.2f\u00b0C", cor_val, bias)
  ) +
  theme_minimal(base_size = 11)

Seasonal cycle

Averaged over 2000-2018, the seasonal cycles nearly overlap:

seasonal <- temp_all |>
  group_by(source, month) |>
  summarise(temp = mean(temp, na.rm = TRUE), .groups = "drop") |>
  mutate(month_label = factor(month.abb[month], levels = month.abb))

ggplot(seasonal, aes(x = month_label, y = temp, color = source, group = source)) +
  geom_line(linewidth = 1.2) +
  geom_point(size = 2.5) +
  scale_color_manual(values = c("ERA5" = "#E74C3C", "CRU TS" = "#2E86AB")) +
  labs(
    x = NULL, y = "Mean temperature (\u00b0C)", color = NULL,
    title = "India: Seasonal Temperature Cycle (2000-2018 mean)"
  ) +
  theme_minimal(base_size = 11) +
  theme(legend.position = "top", panel.grid.minor = element_blank())

Spatial bias map

To compare grid cell by grid cell, we snap ERA5 (0.25°) onto the CRU TS 0.5° grid and look at the difference:

# Snap ERA5 (0.25° grid at whole degrees) to CRU TS grid (0.5° grid offset by 0.25°).
# CRU TS grid centers: floor(x/0.5)*0.5 + 0.25 → 6.25, 6.75, 7.25, ...
era5_spatial <- era5_tmp |>
  mutate(
    lat_bin = floor(latitude / 0.5) * 0.5 + 0.25,
    lon_bin = floor(longitude / 0.5) * 0.5 + 0.25
  ) |>
  group_by(lat_bin, lon_bin) |>
  summarise(era5_mean = mean(value, na.rm = TRUE), .groups = "drop")

cru_spatial <- cru_tmp |>
  group_by(latitude, longitude) |>
  summarise(cru_mean = mean(value, na.rm = TRUE), .groups = "drop")

spatial_combined <- inner_join(
  era5_spatial, cru_spatial,
  by = c("lat_bin" = "latitude", "lon_bin" = "longitude")
) |>
  mutate(bias = era5_mean - cru_mean)

p_era5 <- ggplot(spatial_combined, aes(x = lon_bin, y = lat_bin, fill = era5_mean)) +
  geom_tile() +
  scale_fill_distiller(
    palette = "RdYlBu", direction = -1,
    name = "\u00b0C", limits = c(10, 36)
  ) +
  coord_fixed() +
  labs(title = "ERA5 Annual Mean") +
  theme_void(base_size = 10) +
  theme(plot.title = element_text(hjust = 0.5, face = "bold"))

p_cru <- ggplot(spatial_combined, aes(x = lon_bin, y = lat_bin, fill = cru_mean)) +
  geom_tile() +
  scale_fill_distiller(
    palette = "RdYlBu", direction = -1,
    name = "\u00b0C", limits = c(10, 36)
  ) +
  coord_fixed() +
  labs(title = "CRU TS Annual Mean") +
  theme_void(base_size = 10) +
  theme(plot.title = element_text(hjust = 0.5, face = "bold"))

bias_lim <- max(abs(spatial_combined$bias), na.rm = TRUE)
p_bias <- ggplot(spatial_combined, aes(x = lon_bin, y = lat_bin, fill = bias)) +
  geom_tile() +
  scale_fill_gradient2(
    low = "#2166AC", mid = "white", high = "#B2182B",
    midpoint = 0, limits = c(-bias_lim, bias_lim),
    name = "\u00b0C"
  ) +
  coord_fixed() +
  labs(title = "ERA5 \u2212 CRU TS") +
  theme_void(base_size = 10) +
  theme(plot.title = element_text(hjust = 0.5, face = "bold"))

p_era5 + p_cru + p_bias + plot_layout(ncol = 3, guides = "collect") &
  theme(legend.position = "bottom")

Annual trend 

annual <- temp_all |>
  group_by(source, year) |>
  summarise(temp = mean(temp, na.rm = TRUE), .groups = "drop")

ggplot(annual, aes(x = year, y = temp, color = source)) +
  geom_line(linewidth = 0.8) +
  geom_point(size = 1.5) +
  geom_smooth(method = "lm", se = FALSE, linetype = "dashed", linewidth = 0.8) +
  scale_color_manual(values = c("ERA5" = "#E74C3C", "CRU TS" = "#2E86AB")) +
  labs(
    x = NULL, y = "Annual mean temperature (\u00b0C)", color = NULL,
    title = "India: Annual Temperature Trend — ERA5 vs CRU TS",
    subtitle = "Dashed = linear trend"
  ) +
  theme_minimal(base_size = 11) +
  theme(legend.position = "top", panel.grid.minor = element_blank())

When to use which

CRU TS is the simpler choice when you need long historical baselines without CDS credentials, or monthly land-only temperature is enough. ERA5 wins when you need sub-monthly frequency, ocean coverage, or variables beyond temperature.