consider using HWLWno in `hatyan.calc_HWLW12345to12()`

[veenstrajelmer]

The computation of extremes (1/2) from extremes including aggers (1/2/3/4/5) is quite complex and slow. There is an alternative method suggested that is simpler, faster and includes all necessary extremes.

The new method has some issues though:

• improve hwlw numbering: https://github.com/Deltares/hatyan/issues/327
• the new method depends on hatyan.calc_HWLWnumbering() which is not always robust (e.g. for long 1870-2024 timeseries for HOEKVHLD, and even not when providing station argument).
• not always robust, like for https://github.com/Deltares-research/kenmerkendewaarden/issues/101 >> https://github.com/Deltares/hatyan/issues/329
• if the timeseries starts with HWLWcode 5 (second LW), the new method includes this as 2, while the preceeding 3 could also be lower. In the current method this LW is not included in the 1/2 timeseries, which is more accurate.

Example code:

import os
import pandas as pd
import matplotlib.pyplot as plt
plt.close('all')
import hatyan
import kenmerkendewaarden as kw # pip install git+https://github.com/Deltares-research/kenmerkendewaarden

# set logging level to INFO to get log messages
import logging
logging.basicConfig() # calling basicConfig is essential to set logging level for sub-modules
logging.getLogger("kenmerkendewaarden").setLevel(level="INFO")

dir_base = r'p:\11210325-005-kenmerkende-waarden\work'
dir_meas = os.path.join(dir_base,'measurements_wl_20101201_20220201')
# dir_meas = os.path.join(dir_base,'measurements_wl_18700101_20240101')

current_station = 'HOEKVHLD'

print(f'loading data for {current_station}')
data_pd_meas_all = kw.read_measurements(dir_output=dir_meas, station=current_station, extremes=False)
data_pd_HWLW_all = kw.read_measurements(dir_output=dir_meas, station=current_station, extremes=True)

# original 12345 to 12 conversion method in hatyan
data_pd_HWLW_all_12 = hatyan.calc_HWLW12345to12(data_pd_HWLW_all) #convert 12345 to 12 by taking minimum of 345 as 2 (laagste laagwater)
data_pd_HWLW_all_12 = hatyan.calc_HWLWnumbering(data_pd_HWLW_all_12)

# alternative 12345 to 12 conversion (faster, better, but calc_HWLWnumbering is not robust)
data_pd_HWLW_all_nos = hatyan.calc_HWLWnumbering(data_pd_HWLW_all) # raises "Exception: tidal wave numbering: LW numbers not always increasing" for HOEKVHLD 1870-2024-timeseries
data_pd_HWLW_all_35 = data_pd_HWLW_all_nos[data_pd_HWLW_all_nos["HWLWcode"].isin([3,5])]
data_pd_HWLW_all_35_min = data_pd_HWLW_all_35.loc[data_pd_HWLW_all_35.groupby('HWLWno').values.idxmin()]
data_pd_HWLW_all_35_min["HWLWcode"] = 2
data_pd_HWLW_all_12_new = data_pd_HWLW_all_nos[data_pd_HWLW_all_nos["HWLWcode"].isin([1,2])]
data_pd_HWLW_all_12_new = pd.concat([data_pd_HWLW_all_35_min, data_pd_HWLW_all_12_new], axis=0).sort_index()

# plot clearly shows missing lowwaters in original attempt
fig, (ax1,ax2) = hatyan.plot_timeseries(ts=data_pd_meas_all, ts_ext=data_pd_HWLW_all_12_new, ts_ext_validation=data_pd_HWLW_all_12)
ax1.set_xlim(pd.Timestamp("2012-12-18"), pd.Timestamp("2012-12-23"))

Gives:

[veenstrajelmer]

At this stage it is to tricky to let this function depend on the extreme numbering. In many datasets there are incorrect duplicates and this messes up the HWLWnumbering method, leading it to fail. The hatyan.calc_HWLW12345to12() is indeed quite slow, but it is robust at least.