---
image: https://hackmd.io/_uploads/ryeP8Grgbp.png
---
# 用R做年度每日氣溫圖
###### tags: `用R做`,`學習筆記`

之前用R做過[氣候圖](/k8S0DAi5Sle1-w-_VGT8pQ)，這次換個視覺化的方法，繪製全年的每日氣溫分布圖。從圖中可以快速了解一年當中每天的高、低溫與平均溫度。

一開始，當然要從氣象署的[觀測資料查詢系統](https://e-service.cwb.gov.tw/HistoryDataQuery/)抓出每天的氣溫資料。以下用已撤銷的香山測站為例，參考[阿好伯的做法](https://hackmd.io/@LHB-0222/Wind_Rose)，抓出設站開始到撤銷之前完整年的資料：

library(dplyr)
library(jsonlite)
library(rvest)
library(magrittr)
library(lubridate)
library(ggplot2)
library(lubridate)
#載入資料，站名記得換
url_start <- "https://e-service.cwb.gov.tw/HistoryDataQuery/DayDataController.do?command=viewMain&station=C0D570&stname=%25E9%25A6%2599%25E5%25B1%25B1&datepicker="
#先寫自訂函式
climatecrawler <- function(x,y){
    start_date <- as.Date(x)
    end_date <- as.Date(y)
    date_seq <- seq(start_date, end_date, by = "days")
    date_seq_str <- as.vector(format(date_seq, "%Y-%m-%d"))
    data1 <- data.frame()
    data2 <- data.frame()
    for(single_date in date_seq_str){
      url <- paste0(url_start, single_date, "&altitude=15m") #高度計得換

      temp <- url %>%
        read_html() %>%
        html_nodes(xpath = '//*[(@id = "MyTable")]//td[(((count(preceding-sibling::*) + 1) = 4) and parent::*)]') %>%
        html_text(trim	= T) %>% 
        gsub("X", "-9999", .) %>% 
        as.numeric()
      
      data1 <- cbind(date = single_date, ta = mean(temp), tn = min(temp), tx = max(temp))
      data2 <- rbind(data2, data1)
    }
    return(data2)
  }
#可以開始抓資料了
HSClimate_20071201_20211130 <- climatecrawler("2007-12-01", "2021-11-30")
write.csv(HSClimate_20071201_20211130, "香山測站20071201_20211130觀測資料.csv")

資料抓完，就可以開始視覺化。這邊從讀取 .csv 開始，當然也可以連著上面直接用抓好的資料接著做。

HSdf <- read.csv("香山測站20071201_20211130觀測資料.csv", header = T, row.names = 1)
HSdf <- na.omit(HSdf)
HSdf$day <- yday(HSdf$date)

HSdf.o <- aggregate(HSdf[2:4], HSdf["day"], mean)
HSdf.o <- mutate(HSdf.o, day = as_date(day, origin = "1999-12-31"))

col_temp <- c("#0b144f","#0e2680","#223b97","#1c499a","#2859a5",
              "#1b6aa3","#1d9bc4","#1ca4bc","#64c6c7","#86cabb",
              "#91e0a7","#c7eebf","#ebf8da","#f6fdd1","#fdeca7",
              "#f8da77","#fcb34d","#fc8c44","#f85127","#f52f26",
              "#d10b26","#9c042a","#760324","#18000c")

ggplot(HSdf.o) + 
  geom_linerange(aes(day, ymax = tx, ymin = tn, colour = ta), linewidth = .8, alpha = .7) + 
  scale_y_continuous(breaks = seq(0, 40, 10), limits = c(0, 40), expand = expansion()) +
  scale_colour_gradientn(colours = col_temp, limits = c(10, 30), breaks = seq(10, 35, 5)) + 
  scale_x_date(date_breaks = "month", date_labels = "%b") +
  labs(title = "香山氣象站2007-2021全年日溫變化", colour = "日均溫") +
  theme_minimal(base_size = 12) +
  theme(plot.title = element_text(size = 14), axis.title.x = element_blank(), panel.grid.minor = element_blank())

主要就是把資料依據 1–366 日，把各年的高、低、均溫平均，再找一個閏年（這邊用 2000 年）塞進去，用該年度的日期作為 x 軸。y 軸當然就是溫度了，以每日低溫與高溫決定線段長度，並依照每日均溫上色（視覺化的程式碼與色票參考自[此](https://www.r-bloggers.com/2021/09/climate-circles/?fbclid=IwAR3zil87S00U2iIQQxaWn-z9-7ZUewWbirvWj59IdZtoLOeMitkZKomd4Xg)）。

做出來的效果就像下圖：

<br>

如果加上 `coord_polar()`，就可以把圖轉換一個圓：


一開始抓的資料，包含四年一度的 2 月 29 日，過少的資料會讓平均值和相鄰的日期不同，可以把這天去掉。如果把 2 月 29 日去除，那在 `as_date()` 指令中，就需要選擇一個非閏年，日數才對得上。

<span style="font-size:30px">🐕‍🦺</span><font color="dcdcdc">2023.10.08</font>