保存pcolormesh图时Matplotlib渲染器出现float类型转换错误

背景与代码变更 #

我将处理NetCDF数据集的绘图脚本中，把contourf替换为pcolormesh，用于生成多时间步的地图和特定位置的变量时间剖面图。

原代码：

im = ax.contourf(X,Y,np.transpose(data),*args,**kwargs)

其中：

• X：float64类型的时间数据（单位：秒）
• Y：float64类型的高度数据
• data：NetCDF数据集的切片，类型为float32

替换后的代码：

im = ax.pcolormesh(X,Y,np.transpose(data),shading='gouraud',*args,**kwargs)

kwargs仅包含色条及色条范围配置，无特殊设置。

问题现象 #

本地Windows的Spyder环境中测试用例运行完全正常，但在Linux服务器命令行执行时，保存图像阶段抛出如下错误：

Traceback (most recent call last):
  File "(redacted)/main.py", line 37, in <module>
    dataset.make_plots()
  File "(redacted)/plotDataset.py", line 75, in make_plots
    self._crossecPlot(data, variable, crossec)
  File "(redacted)/plotDataset.py", line 196, in _crossecPlot
    plt.savefig(self.settings['output_path']+'/'+filename+'.png', bbox_inches='tight')
  File "(redacted)/miniconda3/lib/python3.10/site-packages/matplotlib/pyplot.py", line 1023, in savefig
    res = fig.savefig(*args, **kwargs)
  File "(redacted)/miniconda3/lib/python3.10/site-packages/matplotlib/figure.py", line 3343, in savefig
    self.canvas.print_figure(fname, **kwargs)
  File "(redacted)/miniconda3/lib/python3.10/site-packages/matplotlib/backends/backend_qtagg.py", line 75, in print_figure
    super().print_figure(*args, **kwargs)
  File "(redacted)/miniconda3/lib/python3.10/site-packages/matplotlib/backend_bases.py", line 2366, in print_figure
    result = print_method(
  File "(redacted)/miniconda3/lib/python3.10/site-packages/matplotlib/backend_bases.py", line 2232, in <lambda>
    print_method = functools.wraps(meth)(lambda *args, **kwargs: meth(
  File "(redacted)/miniconda3/lib/python3.10/site-packages/matplotlib/backends/backend_agg.py", line 509, in print_png
    self._print_pil(filename_or_obj, "png", pil_kwargs, metadata)
  File "(redacted)/miniconda3/lib/python3.10/site-packages/matplotlib/backends/backend_agg.py", line 457, in _print_pil
    FigureCanvasAgg.draw(self)
  File "(redacted)/miniconda3/lib/python3.10/site-packages/matplotlib/backends/backend_agg.py", line 400, in draw
    self.figure.draw(self.renderer)
  File "(redacted)/miniconda3/lib/python3.10/site-packages/matplotlib/artist.py", line 95, in draw_wrapper
    result = draw(artist, renderer, *args, **kwargs)
  File "(redacted)/miniconda3/lib/python3.10/site-packages/matplotlib/artist.py", line 72, in draw_wrapper
    return draw(artist, renderer)
  File "(redacted)/miniconda3/lib/python3.10/site-packages/matplotlib/figure.py", line 3140, in draw
    mimage._draw_list_compositing_images(
  File "(redacted)/miniconda3/lib/python3.10/site-packages/matplotlib/image.py", line 131, in _draw_list_compositing_images
    a.draw(renderer)
  File "(redacted)/miniconda3/lib/python3.10/site-packages/matplotlib/artist.py", line 72, in draw_wrapper
    return draw(artist, renderer)
  File "(redacted)/miniconda3/lib/python3.10/site-packages/matplotlib/axes/_base.py", line 3064, in draw
    mimage._draw_list_compositing_images(
  File "(redacted)/miniconda3/lib/python3.10/site-packages/matplotlib/image.py", line 131, in _draw_list_compositing_images
    a.draw(renderer)
  File "(redacted)/miniconda3/lib/python3.10/site-packages/matplotlib/artist.py", line 72, in draw_wrapper
    return draw(artist, renderer)
  File "(redacted)/miniconda3/lib/python3.10/site-packages/matplotlib/collections.py", line 2095, in draw
    renderer.draw_gouraud_triangles(
TypeError: Cannot cast array data from dtype('float128') to dtype('float64') according to the rule 'safe'

已尝试的解决方法 #

• 将shading改为默认的nearest，报错位置变为renderer.draw_quad_mesh，问题依旧
• 更换Anaconda/Miniconda环境，未解决问题

异常细节 #

同一数据集的另一个绘图场景中，使用pcolormesh搭配Cartopy时无报错：

im = ax.pcolormesh(lons,lats,values,transform=ccrs.PlateCarree(),*args, **kwargs)

目前不清楚float128类型的来源，也无法理解为何不能转换为float64（绘图场景下精度损失完全可接受）。

解决方案 #

1. 显式转换所有输入数组为float64 #

强制将X、Y、data转换为matplotlib后端支持的float64类型，避免内部生成float128数组：

# 在调用pcolormesh前添加类型转换
X = X.astype(np.float64)
Y = Y.astype(np.float64)
transposed_data = np.transpose(data).astype(np.float64)
im = ax.pcolormesh(X, Y, transposed_data, shading='gouraud', *args, **kwargs)

2. 检查并调整numpy浮点类型配置（可选） #

Linux环境下的numpy可能默认启用了float128支持，可在脚本开头强制设置默认浮点类型为float64：

import numpy as np
np.set_default_dtype(np.float64)

注意：此设置会影响脚本中所有numpy数组的默认类型，若有其他高精度计算需求，优先使用显式类型转换。

3. 原因分析 #

Linux环境中numpy的浮点类型支持与Windows存在差异，pcolormesh在处理无Cartopy transform的网格数据时，内部计算可能将数组类型提升为float128，而matplotlib的Agg后端仅支持float64，且转换时使用了safe规则（不允许精度损失），导致报错。而Cartopy的transform内部会自动将输入数组转换为float64，因此未触发问题。

内容的提问来源于stack exchange，提问作者Florian Sauerland