Separated legend between two variables in ggplot

I have a dataframe which look like this :

From it, I want to make a plot which look like this :

I used ggplot and LibreOffice Draw to make the above plot. I wish to use ggplot only.

Here is a reproducible example :

library (dplyr)
n = 5

df = data.frame (    
    time = c(1 : n),
    prediction_model1 = rnorm (n) %>% sort,
    prediction_model2 = rnorm (n) %>% sort,
    observed_values = rnorm (n) %>% sort    
)

df_plot = df %>% 
    pivot_longer (cols = c (prediction_model1, prediction_model2), names_prefix = "prediction_", values_to = "happyness")

ggplot (df_plot) +

    geom_point (aes (x = time, y = happyness, color = name), size = 2) +
    geom_line (aes (x = time, y = happyness, color = name)) +

    geom_point (aes (x = time, y = observed_values), shape = 5, size = 4) +
    geom_line (aes (x = time, y = observed_values)) +

    labs (color = "") +

    theme (legend.position = "bottom")

>Solution :

You could map the shape and linetype aesthetics to a character value:

ggplot(df_plot, aes(x = time, y = happyness)) +
  geom_point(aes(color = name), size = 2) +
  geom_line (aes(color = name)) +
  geom_point(aes(y = observed_values, shape = 'Observed values'), size = 3) +
  geom_line (aes(y = observed_values, linetype = 'Observed values')) +
  scale_shape_manual(name = NULL, values = 5) +
  scale_linetype_manual(name = NULL, values = 1) +
  labs(color = NULL) +
  guides(linetype = guide_legend(keyheight = unit(20, 'mm'))) +
  theme_classic(base_size = 16) +
  theme(legend.position = "bottom",
        axis.line = element_blank(),
        panel.background = element_rect(fill = '#f3faff', color = 'gray60'))