Processor: BiasAdjustModelToStation

Registry key: bias_adjust_model_to_station | Priority: 60 | Category: Data Refinement

Apply quantile delta mapping (QDM) bias adjustment to gridded climate model data using HDP (Historical Data Platform) weather station observations as the training reference. Output is bias-adjusted at the requested station locations, with a station dimension (and lat/lon/elevation coordinates along it) on a single data variable, rather than one data variable per station.

Algorithm

flowchart TD
    Start([execute]) --> DictCheck{result is dict?}
    DictCheck -->|Yes| ExecDict[_execute_dict<br/>per-key recursion]
    DictCheck -->|No| LoadObs[_load_station_data<br/>HDP reference]
    LoadObs --> TypeCheck{Dataset / DataArray?}
    TypeCheck -->|Yes| ProcSingle[_process_single_dataset]
    TypeCheck -->|No| TypeErr[raise TypeError]

    ExecDict --> ProcSingle
    ProcSingle --> Preproc[_preprocess_hdp<br/>rename, unit check, attrs]
    Preproc --> BiasAdjust[_bias_adjust_model_data<br/>QDM via xclim]
    BiasAdjust --> UpdateCtx[update_context]
    UpdateCtx --> End([Output: Dataset with a 'station' dimension])

    click Start "https://github.com/cal-adapt/climakitae/blob/main/climakitae/new_core/processors/bias_adjust_model_to_station.py" "execute"
    click ExecDict "https://github.com/cal-adapt/climakitae/blob/main/climakitae/new_core/processors/bias_adjust_model_to_station.py" "_execute_dict"
    click LoadObs "https://github.com/cal-adapt/climakitae/blob/main/climakitae/new_core/processors/bias_adjust_model_to_station.py" "_load_station_data"
    click ProcSingle "https://github.com/cal-adapt/climakitae/blob/main/climakitae/new_core/processors/bias_adjust_model_to_station.py" "_process_single_dataset"
    click Preproc "https://github.com/cal-adapt/climakitae/blob/main/climakitae/new_core/processors/bias_adjust_model_to_station.py" "_preprocess_hdp"
    click BiasAdjust "https://github.com/cal-adapt/climakitae/blob/main/climakitae/new_core/processors/bias_adjust_model_to_station.py" "_bias_adjust_model_data"
    click UpdateCtx "https://github.com/cal-adapt/climakitae/blob/main/climakitae/new_core/processors/bias_adjust_model_to_station.py" "update_context"

Parameters

The processor takes a dict:

Key	Type	Default	Description
`stations`	`list[str]`	`[]`	HDP `station_id` values to bias-adjust against (e.g. `"ASOSAWOS_69007093217"`). Required for non-trivial use.
`historical_slice`	`tuple[int, int]`	`(1980, 2014)`	Years used as the training period. Must overlap each station's actual observational coverage, which varies per HDP station.
`window`	`int`	`90`	Window size (days) for seasonal grouping in QDM.
`nquantiles`	`int`	`20`	Number of quantiles for the QDM mapping.
`group`	`str`	`"time.dayofyear"`	Temporal grouping passed to `xclim` QDM.
`kind`	`str`	`"+"`	Adjustment kind: `"+"` (additive) or `"*"` (multiplicative).

Requirements

Activity ID: WRF (dynamical downscaling). LOCA2 already includes statistical bias adjustment at the watershed level and is not the intended input.
Variable: Designed for hourly t2 (temperature, matched to HDP's tas) or dew_point (dewpoint, matched to HDP's tdps, or tdps_derived where tdps isn't available). Not every HDP network provides every variable — requesting a station from a network known not to provide the needed variable is rejected up front.
Time coverage: Input must overlap the requested historical training period (default 1980–2014). HDP station coverage varies widely by station, from a few years to multiple decades.
Multiple networks: Requested stations may span multiple HDP networks in a single call.
Calendar: All inputs are converted to a noleap calendar internally for consistency.

Example

from climakitae.new_core.user_interface import ClimateData

data = (ClimateData()
    .catalog("cadcat")
    .activity_id("WRF")
    .institution_id("UCLA")
    .variable_id("t2")
    .table_id("1hr")
    .grid_label("d03")
    .processes({
        "time_slice": ("1980-01-01", "2050-12-31"),
        "bias_adjust_model_to_station": {"stations": ["KSAC"]},
    })
    .get())

# Result: single data variable with a 'station' dimension, time-sliced to requested range.

Code References

Method	Link to Code	Purpose
`__init__`	View on Github	Read configuration dict with defaults
`_preprocess_hdp`	View on Github	Rename / unit-check / attribute the raw HDP station slice
`_load_station_data`	View on Github	Resolve station IDs against the HDP catalog, load HDP subset, return reference Dataset
`_bias_adjust_model_data`	View on Github	Build QDM (`xclim`) train/adjust, vectorized across all stations at once
`_process_single_dataset`	View on Github	Load reference, run vectorized QDM across stations, return dataset with a `station` dimension
`_execute_dict`	View on Github	Recursive dict path
`execute`	View on Github	Dispatcher: dict / Dataset / DataArray
`update_context`	View on Github	Record stations and QDM parameters in `new_attrs`
`set_data_accessor`	View on Github	Receive `DataCatalog` reference (used by `_load_station_data`)

Earlier docs implied an early "activity_id == WRF" guard with line numbers in the 80–145 range. Those line numbers do not exist in the source — the file's first method (__init__) starts at line 148. The activity-id constraint is enforced by data availability rather than an explicit early-return check.