Skip to content

Valencia-Schaake Temporal Disaggregation (Valencia and Schaake 1973)
Type Parametric
Resolution Annual to Monthly
Sites Univariate / Multisite
Class ValenciaSchaakeDisaggregator

Overview

The Valencia-Schaake method is the foundational parametric temporal disaggregation approach in stochastic hydrology. It disaggregates an aggregate flow volume (e.g., annual total) into sub-period values (e.g., 12 monthly flows) using a linear regression model that preserves the conditional mean and covariance structure of the sub-periods given the aggregate. The method models sub-period flows as a multivariate normal distribution conditioned on the known aggregate, then samples from this conditional distribution.

This is the classical baseline against which all subsequent disaggregation methods are compared.

Algorithm

Preprocessing

  1. Validate input: observed flows at the finer resolution (e.g., daily or monthly) with DatetimeIndex.
  2. Aggregate observed flows to the coarser resolution (e.g., monthly to annual by summation).
  3. Organize into matrices: for each year y, form the vector of sub-period flows: 
    X_y = [Q_{y,1}, Q_{y,2}, ..., Q_{y,m}]^T
    where m is the number of sub-periods (e.g., 12 months).
  4. Optional transformation: apply log or Box-Cox transform to improve normality.

Fitting

  1. Compute sub-period statistics:
  2. Mean vector: mu_X = E[X] (m x 1)
  3. Covariance matrix: S_XX = Cov(X, X) (m x m)
  4. Compute aggregate statistics:
  5. Mean: mu_Y = E[Y] where Y = sum(X)
  6. Variance: sigma_Y^2 = Var(Y)
  7. Compute cross-covariance between sub-periods and aggregate: 
    S_XY = Cov(X, Y) = S_XX * 1_m
    where 1_m is an m-vector of ones.
  8. Compute regression parameters:
  9. Regression coefficients: A = S_XY / sigma_Y^2 (m x 1)
  10. Conditional covariance: S_e = S_XX - A * sigma_Y^2 * A^T (m x m)
  11. Cholesky decomposition of S_e: 
    S_e = C * C^T
    If S_e is not positive semi-definite, apply spectral repair (set negative eigenvalues to small positive value).
  12. Store mu_X, mu_Y, sigma_Y^2, A, C, and transformation parameters.

Disaggregation

For each synthetic aggregate value Y_syn:

  1. Compute conditional mean: 
    mu_X|Y = mu_X + A * (Y_syn - mu_Y)
  2. Sample residuals: 
    Z ~ N(0, I_m)
X_syn = mu_X|Y + C * Z
  3. Proportional adjustment to enforce consistency (sub-periods sum to aggregate): 
    X_adj = X_syn * (Y_syn / sum(X_syn))
    Note: This multiplicative adjustment is simple but can distort the conditional covariance. See Grygier-Stedinger (1988) for a more rigorous conservation correction.
  4. Inverse transform if log/Box-Cox was applied.
  5. Enforce non-negativity: clip to zero.

Parameters

Parameter Type Default Description
Q_obs pd.Series or pd.DataFrame - Observed streamflow at sub-period resolution with DatetimeIndex
n_subperiods int 12 Number of sub-periods per aggregate period
transform str 'log' Transformation before fitting: 'log', 'boxcox', or 'none'
conservation_method str 'proportional' Method to enforce sum consistency: 'proportional' or 'none'
name Optional[str] None Optional name identifier for this disaggregator instance
debug bool False Enable debug logging

Properties Preserved

  • Conditional mean of sub-periods given aggregate (by construction)
  • Conditional covariance structure (via S_e)
  • Monthly means and standard deviations (approximately)
  • Cross-correlations between sub-periods (via full covariance S_XX)
  • Aggregate total (exactly, via proportional adjustment)

Not preserved: - Exact conditional covariance after proportional adjustment - Non-Gaussian features of sub-period distributions - Inter-annual temporal correlations between sub-periods

Limitations

  • Assumes multivariate normality of sub-period flows (often violated for daily/monthly data)
  • Proportional adjustment distorts the conditional covariance
  • Does not model serial correlation between consecutive years' sub-periods
  • For high m (many sub-periods), covariance estimation requires long records
  • Better suited for annual-to-monthly than monthly-to-daily (daily data requires many more sub-periods)

References

Primary: Valencia, R.D., and Schaake, J.C. (1973). Disaggregation processes in stochastic hydrology. Water Resources Research, 9(3), 580-585. https://doi.org/10.1029/WR009i003p00580

See also: - Stedinger, J.R., and Vogel, R.M. (1984). Disaggregation procedures for generating serially correlated flow vectors. Water Resources Research, 20(1), 47-56. https://doi.org/10.1029/WR020i001p00047 - Salas, J.D., Delleur, J.W., Yevjevich, V., and Lane, W.L. (1980). Applied Modeling of Hydrologic Time Series. Water Resources Publications.

Implementation: src/synhydro/methods/disaggregation/temporal/valencia_schaake.py Tests: tests/test_valencia_schaake_disaggregator.py