使用直方图将变量 Y 采样为 X 的函数时，python 舍入误差

我try 使用函数直方图将一个变量(SST)作为另一个变量(TCWV)的函数进行采样，并将权重设置为示例变量，如下所示:

# average sst over bins
num, _   = np.histogram(tcwv, bins=bins)
sstsum, _ = np.histogram(tcwv, bins=bins,weights=sst)
out=np.zeros_like(sstsum)
out[:]=np.nan
sstav  = np.divide(sstsum,num,out=out, where=num>100)

重现性的完整代码如下所示.我的问题是，当我绘制原始数据的散点图，然后绘制计算出的平均值时，平均值就像这样位于数据"云"之外(请参见右侧的点):

我想不出为什么会发生这种情况，除非这可能是一个舍入误差？

这是我的全部代码:

import numpy as np
import matplotlib.pyplot as plt
from netCDF4 import Dataset

# if you have a recent netcdf libraries you can access it directly here 
url = ('http://clima-dods.ictp.it/Users/tompkins/CRM/data/WRF_1min_mem3_grid4.nc#mode=bytes')
ds=Dataset(url)

### otherwise need to download, and use this:
###ifile="WRF_1min_mem3_grid4.nc"
###ds=Dataset(idir+ifile)


# axis bins
bins=np.linspace(40,80,21)

iran1,iran2=40,60

# can put in dict and loop 
sst=ds.variables["sst"][iran1:iran2+1,:,:]
tcwv=ds.variables["tcwv"][iran1:iran2+1,:,:]

# don't need to flatten, just tried it to see if helps (it doesn't)
sst=sst.flatten()
tcwv=tcwv.flatten()

# average sst over bins
num, _   = np.histogram(tcwv, bins=bins)
sstsum, _ = np.histogram(tcwv, bins=bins,weights=sst)
out=np.zeros_like(sstsum)
out[:]=np.nan
sstav  = np.divide(sstsum,num,out=out,where=num>100)

# bins centroids
avbins=(np.array(bins[1:])+np.array(bins[:-1]))/2

#plot
subsam=2
fig,(ax)=plt.subplots()
plt.scatter(tcwv.flatten()[::subsam],sst.flatten()[::subsam],s=0.05,marker=".")
plt.scatter(avbins,sstav,s=3,color="red")
plt.ylim(299,303)
plt.savefig("scatter.png")