python 根据cartopy插件出气象等值线高度场图片

python 根据cartopy插件出气象等值线高度场图片,第1张

最近工作需要根据nc文件中要素出一张500hap高度场+850hpa风+850hpa相对湿度图片,

网上查找资料几天终于做出来了,下面直接上代码和效果图

import numpy as np
import matplotlib.pyplot as plt
import matplotlib as mpl
import scipy.ndimage as ndimage
import xarray as xr
import cartopy.crs as ccrs
import cartopy.io.shapereader as shpreader
import cartopy.feature as cfeature
from itertools import chain
from cartopy.mpl.ticker import LongitudeFormatter, LatitudeFormatter

# 打开文件
path = 'D:/file/wrfprs_d01.2022040508.f15.nc'

ds = xr.open_dataset(path)
lat = ds['NLAT_GDS3_SFC']  # 维度
lon = ds['ELON_GDS3_SFC']  # 经度
hght = ds['HGT_GDS3_ISBL'].sel(lv_ISBL2=500)  # 高度
# avor = ds['ABS_V_GDS3_ISBL'].sel(lv_ISBL2=500)#涡度
avor = ds['TMP_GDS3_ISBL'].sel(lv_ISBL2=500)-273.15  # 温度

# ds = xr.open_dataset(path)
# 数据处理
## 500hpa 数据
# hgt_500 = ds['HGT_GDS3_ISBL']
# temp_500 = ds['TMP_GDS3_ISBL']
u_500 = ds['U_GRD_GDS3_ISBL']
v_500 = ds['V_GRD_GDS3_ISBL']

lons = ds['NLAT_GDS3_SFC'][0, :]
lats = ds['ELON_GDS3_SFC'][:, 0]
# hgt_500=hgt_500[0]*0.1 ## 单位换算为dagpm
# temp_500 = temp_500[0]-273.15
# u_500 = u_500[0]
# v_500 = v_500[0]
# hgt_500=hgt_500.sel(lv_ISBL2=500)*0.1 ## 单位换算为dagpm
# temp_500 = temp_500.sel(lv_ISBL2=500)-273.15
u_500 = u_500.sel(lv_ISBL2=500)
v_500 = v_500.sel(lv_ISBL2=500)
# 网格化 ,生成网格点坐标矩阵
lons, lats = np.meshgrid(lons, lats)
# 绘图
# 一页4图
# 设置投影方式,地图边界
proj = ccrs.PlateCarree()  # 定义投影转换,后面都会用到,不必重复输入‘ccrs.PlateCarree()’
# proj=ccrs.LambertConformal(central_longitude=120.0,central_latitude=45,standard_parallels=[40])
# leftlon, rightlon, lowerlat, upperlat = (113,120,36,43)#113,120,36,43  70,140,15,55
leftlon, rightlon, lowerlat, upperlat = (70,140,15,55)  # 113,120,36,43  70,140,15,55
img_extent = [leftlon, rightlon, lowerlat, upperlat]

# 建立画布
fig = plt.figure(figsize=(12, 8))

# 添加第一子图
ax1 = fig.add_axes([0.1, 0.1, 0.8, 0.8], projection=proj)
# contour_map(ax1,img_extent,10)


# Smooth and re-plot the vorticity field 设置色斑图
avor_levels = np.linspace(-30, 30, 10)
arr=[]
for i in range(len(avor_levels)):
   arr.append( int(avor_levels[i]))

avor_smooth = ndimage.gaussian_filter(avor, sigma=1.5, order=0)
avor_contour = ax1.contourf(lon, lat, avor_smooth, levels=arr, zorder=2, cmap=plt.cm.YlOrRd,
                            transform=ccrs.PlateCarree())
# Smooth and re-plot the height field 高度场
hght_levels = np.arange(1000, 10000, 20)
hght_smooth = ndimage.gaussian_filter(hght, sigma=3, order=0)
hght_contour = ax1.contour(lon, lat, hght_smooth, levels=hght_levels, linewidths=1, colors='k',
                           zorder=2, transform=ccrs.PlateCarree())
# Plot contour labels for the heights, leaving a break in the contours for the text (inline=True)
plt.clabel(hght_contour, hght_levels, inline=True, fmt='%1i', fontsize=15)

# Get the wind components 风场
urel = ds['U_GRD_GDS3_ISBL'].sel(lv_ISBL2=500)  # *1.944 #U风
vrel = ds['V_GRD_GDS3_ISBL'].sel(lv_ISBL2=500)  # *1.944 #v风
# 风场
ax1.barbs(lon[::10, ::10], lat[::10, ::10], urel[::10, ::10], vrel[::10, ::10],
          linewidth=0.4, flagcolor='k', linestyle='-', length=5,
          pivot='tip', barb_increments=dict(half=2, full=4, flag=20),
          sizes=dict(spacing=0.15, height=0.5, width=0.12),zorder=2,transform=ccrs.PlateCarree())

# Create a colorbar and shrink it down a bit.
cb = plt.colorbar(avor_contour, shrink=0.5)
cb.ax.tick_params(labelsize=15)

##定义地理坐标标签格式
xstep, ystep = 2, 1
# xstep, ystep = 1, 1
###set_extent需要配置相应的crs,否则出来的地图范围不准确
ax1.set_extent(img_extent, crs=proj)
#### 标注坐标轴
ax1.set_xticks(np.arange(leftlon, rightlon + xstep, xstep), crs=proj)
ax1.set_yticks(np.arange(lowerlat, upperlat + ystep, ystep), crs=proj)
# zero_direction_label用来设置经度的0度加不加E和W
lon_formatter = LongitudeFormatter(zero_direction_label=False)
lat_formatter = LatitudeFormatter()
ax1.xaxis.set_major_formatter(lon_formatter)
ax1.yaxis.set_major_formatter(lat_formatter)
ax1.set_title('(a)', loc='left')

# 地图设置
# 湖
# ax1.add_feature(cfeature.LAKES, alpha=0.5)
# 国界
# ax1.add_feature(cfeature.BORDERS, linestyle='-',lw=0.25)
##海岸线,50m精度
# ax1.add_feature(cfeature.COASTLINE.with_scale('50m'))

file = 'D:/map/map/Province.dbf'
# 本地shp文件
china = shpreader.Reader(file).geometries()
# 绘制中国国界省界九等等
ax1.add_geometries(china, proj, facecolor='none', edgecolor='black', zorder=2)
china = shpreader.Reader(file).geometries()
#plt.show()
# 存储
plt.savefig('D:/123.png',format='png',dpi=500)

效果图

 图片没有细调 大概出来就是这样的 可以根据经纬度进行展示到指定范围呢  需要中国shp文件可以私聊

欢迎分享,转载请注明来源:内存溢出

原文地址: http://outofmemory.cn/langs/796361.html

(0)
打赏 微信扫一扫 微信扫一扫 支付宝扫一扫 支付宝扫一扫
上一篇 2022-05-06
下一篇 2022-05-06

发表评论

登录后才能评论

评论列表(0条)

保存