Distributed hydrological models subdivide a catchment into spatial computational units — grid cells, sub-basins, or hydrological response units — and this discretization is part of the model structure itself: it controls how processes, inputs, and parameters are aggregated, and therefore both the bias and the uncertainty of simulations. Since Wood, Sivapalan and Beven's representative elementary area concept — the drainage-area scale at which catchment storm responses become statistically similar — and Blöschl and Sivapalan's scale-issues review, a large case-study literature has documented that simulated water balance components, flood peaks, and predictive uncertainty all vary with grid size, watershed subdivision level, and unit-construction method — yet every published threshold is model-, catchment-, and objective-specific, and practitioners still choose discretization by convention or computational convenience. The open problem is to (i) characterize how simulated hydrological states, fluxes, and their calibration and predictive uncertainty depend on spatial discretization, and (ii) establish a practical, transferable criterion (or methodology) for selecting the discretization — grid cell size, subdivision level, and hydrological response unit construction — adapted to modelling objectives, dominant processes, input-data resolution, and computational budget, with demonstrated validity across model structures and catchments rather than case-by-case.

Background

A distributed hydrological model simulates catchment water fluxes (rainfall-runoff, evapotranspiration, soil moisture, river discharge) on a spatial subdivision of the basin, in contrast to a lumped model that treats the basin as a single unit. The subdivision takes several forms: regular grid cells; nested sub-basins ordered by the river network; or hydrological response units (HRUs) — patches of unique land-use/soil/slope combinations obtained by intersecting geospatial layers, typically derived from a digital elevation model (DEM), the standard gridded terrain dataset used to delineate slopes and drainage directions.

The discretization choice is not a neutral numerical setting. Finer resolution represents spatial heterogeneity of terrain, soil, land use, and rainfall better, but increases the number of calibrated parameters (aggravating equifinality, the situation where many different parameter sets fit observations equally well), amplifies sub-grid input errors where point observations must stand in for cell averages, and raises computational cost. Coarser discretization introduces scale-induced biases and makes parameter values and their posterior uncertainty depend on the chosen subdivision level. Published work poses this problem concretely: Götzinger and Bárdossy's generic error-model study separates discharge uncertainty into a meteorological-input component (handled by stochastic simulation) and a process-representation-and-parameterization component (handled by sensitivity analysis), the two sides of the resolution trade-off described above; Han and co-workers' Bayesian analysis of the SLURP (semi-distributed land use-based runoff processes) model on the Xiangxi River identified a subdivision threshold (about 45 aggregated simulation areas, ASAs — the sub-basin computation units of SLURP) beyond which posterior uncertainty stabilizes, but warn the threshold may shift with input-data resolution and was tested with only one model and one uncertainty framework; and Liu and co-workers' conceptual data model for catchment discretization organizes the spatio-temporal datasets underlying the discretization, with hydrological response units as the core between object-oriented databases and physically-based distributed hydrological models.

Classical framing exists but did not yield an operational rule. Wood, Sivapalan and Beven (1990) introduced the representative elementary area (REA) — a scale beyond which catchment storm responses statistically converge — as a candidate basis for choosing computational-unit size; Blöschl and Sivapalan (1995) systematized the wider scale problem (scaling, aggregation, sub-grid parameterization). The REA idea was never developed into a transferable discretization-selection criterion valid across models and objectives. Later, the hyperresolution debate sharpened the stakes: Wood et al. (2011) argued for kilometre-scale global land-surface modelling, while Beven and Cloke (2012) and Beven et al. (2015) replied that increasing resolution alone cannot remove sub-grid heterogeneity, input, and parameter uncertainty — more resolution yields information only when supported by commensurate data, otherwise only "hyperresolution ignorance". Methodology proposals such as Dehotin and Braud's nested discretization (sub-catchments, hydro-landscape units, modelling units) make the trade-off explicit and controllable, but the resulting choice remains case-dependent on data, dominant processes, and modelling objectives. Case studies across model families — SWAT (the Soil and Water Assessment Tool), SLURP, MIKE SHE (an integrated surface–subsurface distributed modelling system), TOPKAPI (a topography-based kinematic-wave model) and others — keep producing model- and basin-specific optimum-threshold or stability-interval results with no accepted generalization, and as late as 2020 the Raven hydrological modelling framework (Craig et al., 2020) is built around the premise that discretization schemes are interchangeable modelling choices to be tested and compared — grids, sub-basins and hydrological response units, lumped, or triangulated irregular network (TIN) meshes — while supplying no rule for choosing among them, leaving the decision to the modeller.

Problem Statement

For distributed (and semi-distributed) hydrological models, determine how simulated hydrological states, fluxes, and their calibration and predictive uncertainty depend on the spatial discretization — grid cell size, watershed subdivision level, and hydrological response unit (HRU) construction — and, on that basis, establish a practical, transferable criterion or methodology for selecting the discretization, adapted to the modelling objectives (e.g., water balance versus flood peaks versus water quality), the dominant hydrological processes, the resolution of available input data, and the computational budget. The modelling objectives, dominant hydrological processes, input-data resolution, and computational budget delimit the scope over which the criterion is claimed rather than four separate empirical demonstrations: for each axis the criterion must state whether and how the discretization choice adapts, and may declare an axis out of scope only where it does not bear on the choice. Validation of the criterion may rest on new simulations, a documented cross-study reanalysis, or an analytical derivation, and must span more than one model structure and more than one catchment; it must also resolve specific open issues left by the sources, including whether subdivision-level thresholds generalize across models and catchments, and at minimum the untested stability and generality of the threshold of about 45 aggregated simulation areas (ASAs, the sub-basin computation units of the SLURP model) found by Han et al. (2014) must be resolved. Answers that merely re-derive a case-specific threshold for one model-catchment pair, without transferability evidence, do not meet the objective.

References

  1. Götzinger, J.; Bárdossy, A. (2008). Generic error model for calibration and uncertainty estimation of hydrological models. Water Resources Research 44(12), W00B01. doi:10.1029/2007WR006691, https://doi.org/10.1029/2007WR006691
  2. Han, Jing-Cheng; Huang, Guo-He; Zhang, Hua; Li, Zhong; Li, Yong-Ping (2014). Bayesian uncertainty analysis in hydrological modeling associated with watershed subdivision level: a case study of SLURP model applied to the Xiangxi River watershed, China. Stochastic Environmental Research and Risk Assessment 28(4), 973–989. doi:10.1007/s00477-013-0792-0, https://doi.org/10.1007/s00477-013-0792-0
  3. Han, Jing-Cheng; Huang, Guo-He; Zhang, Hua; Li, Zhong; Li, Yong-Ping (2013). Effects of watershed subdivision level on semi-distributed hydrological simulations: case study of the SLURP model applied to the Xiangxi River watershed, China. Hydrological Sciences Journal 59(1), 108–125. doi:10.1080/02626667.2013.854368, https://doi.org/10.1080/02626667.2013.854368
  4. Liu, Yuanming; Zhang, Wanchang; Zhang, Zhijie (2015). A conceptual data model coupling with physically-based distributed hydrological models based on catchment discretization schemas. Journal of Hydrology 530, 206–215. doi:10.1016/j.jhydrol.2015.09.049, https://doi.org/10.1016/j.jhydrol.2015.09.049
  5. Wood, Eric F.; Sivapalan, Murugesu; Beven, Keith (1990). Similarity and scale in catchment storm response. Reviews of Geophysics 28(1), 1–18. doi:10.1029/RG028i001p00001, https://doi.org/10.1029/RG028i001p00001
  6. Blöschl, G.; Sivapalan, M. (1995). Scale issues in hydrological modelling: A review. Hydrological Processes 9(3–4), 251–290. doi:10.1002/hyp.3360090305, https://doi.org/10.1002/hyp.3360090305
  7. Dehotin, J.; Braud, I. (2008). Which spatial discretization for distributed hydrological models? Proposition of a methodology and illustration for medium to large-scale catchments. Hydrology and Earth System Sciences 12(3), 769–796. doi:10.5194/hess-12-769-2008, https://doi.org/10.5194/hess-12-769-2008
  8. Wood, Eric F.; Roundy, Joshua K.; Troy, Tara J.; van Beek, L. P. H.; Bierkens, Marc F. P.; Blyth, Eleanor; de Roo, Ad; Döll, Petra; Ek, Mike; Famiglietti, James; Gochis, David; van de Giesen, Nick; Houser, Paul; Jaffé, Peter R.; Kollet, Stefan; Lehner, Bernhard; Lettenmaier, Dennis P.; Peters-Lidard, Christa; Sivapalan, Murugesu; Sheffield, Justin; Wade, Andrew; Whitehead, Paul (2011). Hyperresolution global land surface modeling: Meeting a grand challenge for monitoring Earth's terrestrial water. Water Resources Research 47(5). doi:10.1029/2010WR010090, https://doi.org/10.1029/2010WR010090
  9. Beven, Keith J.; Cloke, Hannah L. (2012). Comment on "Hyperresolution global land surface modeling: Meeting a grand challenge for monitoring Earth's terrestrial water" by Eric F. Wood et al. Water Resources Research 48(1). doi:10.1029/2011WR010982, https://doi.org/10.1029/2011WR010982
  10. Beven, Keith; Cloke, Hannah; Pappenberger, Florian; Lamb, Rob; Hunter, Neil (2015). Hyperresolution information and hyperresolution ignorance in modelling the hydrology of the land surface. Science China Earth Sciences 58(1), 25–35. doi:10.1007/s11430-014-5003-4, https://doi.org/10.1007/s11430-014-5003-4
  11. Tripathi, M. P.; Raghuwanshi, N. S.; Rao, G. P. (2006). Effect of watershed subdivision on simulation of water balance components. Hydrological Processes 20(5), 1137–1156. doi:10.1002/hyp.5927, https://doi.org/10.1002/hyp.5927
  12. Carpenter, Theresa M.; Georgakakos, Konstantine P. (2006). Discretization scale dependencies of the ensemble flow range versus catchment area relationship in distributed hydrologic modeling. Journal of Hydrology 328(1–2), 242–257. doi:10.1016/j.jhydrol.2005.12.008, https://doi.org/10.1016/j.jhydrol.2005.12.008
  13. Gebeyehu, Bahru M.; Jabir, Asie K.; Tegegne, Getachew; Melesse, Assefa M. (2023). Subbasin Spatial Scale Effects on Hydrological Model Prediction Uncertainty of Extreme Stream Flows in the Omo Gibe River Basin, Ethiopia. Remote Sensing 15(3), 611. doi:10.3390/rs15030611, https://doi.org/10.3390/rs15030611
  14. Craig, James R.; Brown, Genevieve; Chlumsky, Robert; Jenkinson, R. Wayne; Jost, Georg; Lee, Konhee; Mai, Juliane; Serrer, Martin; Sgro, Nicholas; Shafii, Mahyar; Snowdon, Andrew P.; Tolson, Bryan A. (2020). Flexible watershed simulation with the Raven hydrological modelling framework. Environmental Modelling and Software 129, 104728. doi:10.1016/j.envsoft.2020.104728, https://doi.org/10.1016/j.envsoft.2020.104728