Find index of the first and/or last value in a column that is not NaN

You can agg :

df.notna().agg({'x1':'idxmax','x2':'idxmax','x3':lambda x: x[::-1].idxmax()})
#df.notna().agg({'x1':'idxmax','x2':'idxmax','x3':lambda x: x[x].last_valid_index()})

---

x1    503
x2    502
x3    504

Another way would be to check if first row is nan and according to that apply the condition:

np.where(df.iloc[0].isna(),df.notna().idxmax(),df.notna()[::-1].idxmax())

---

[503, 502, 504]

first_valid_index() and last_valid_index() can be used.

    >>> df
             x1    x2    x3
    Depth
    500     NaN   NaN  25.0
    501     NaN   NaN  27.0
    502     NaN  33.0  24.0
    503     4.0  32.0  18.0
    504    12.0  45.0   5.0
    505     8.0  38.0   NaN
    >>> df["x1"].first_valid_index()
    503
    >>> df["x2"].first_valid_index()
    502
    >>> df["x3"].first_valid_index()
    500
    >>> df["x3"].last_valid_index()
    504

IIUC

df.stack().groupby(level=1).head(1)
Out[619]: 
Depth    
500    x3    25.0
502    x2    33.0
503    x1     4.0
dtype: float64