Examining boundary layer solutions for Navier-stokes approximations with applications to orange river drought severity indices

Introduction

Leading hydrologists have long noted that drought is not simply the absence of rain-it is a cascade of water deficits that propagates through soils, aquifers, rivers, and reservoirs at different speeds and with different signatures [1] . The Orange River, southern Africa's longest at roughly 2,200 kilometres, exemplifies this cascade: a rainfall deficit in the Lesotho highlands may not register as a low-flow event at the river's mouth near Alexander Bay for weeks or months, depending on antecedent soil moisture, dam releases from the Gariep and Vanderkloof reservoirs, and channel losses to evaporation [2] . Standard drought indices like the SPI and PDSI summarise precipitation and soil moisture statistics, respectively, but neither index incorporates the hydraulic physics of how water moves through a river channel at low flow [3] . At very low discharges the flow in a wide, shallow channel is dominated by viscous boundary layers near the bed and banks, and the velocity profile departs markedly from the logarithmic law that applies at moderate flows [4] . This departure matters because it controls the relationship between discharge and water depth-the rating curve-which in turn determines whether a given flow is deep enough to sustain aquatic habitat, support irrigation abstractions, and maintain navigation. The Navier-Stokes equations govern viscous fluid motion, and their boundary layer simplification-the Prandtl equations-reduces the three-dimensional problem to a twodimensional one that can be solved analytically or semi-analytically for simple geometries [5] . This research applied boundary layer theory to the low-flow regime of the Orange River, deriving velocity profiles that were then converted into a drought severity index. The index reflects how much the observed velocity profile deviates from its expected shape at median flow, with larger deviations indicating more severe drought conditions. Five gauging stations along the lower Orange-Vioolsdrif, Pella, Onseepkans, Upington, and Augrabies-provided monthly discharge and cross-section survey data from 2010 to 2023. Lyapunov stability analysis ensured that the steadystate solutions were robust to small perturbations, spline interpolation extended the point-based index into a continuous spatial field, and Poisson regression linked the index to the count of drought declarations. Extreme value theory estimated return periods for the most severe index values, and hashfunction randomisation secured the bootstrap procedure against implementation dependence [6] . The results, detailed in the following sections, suggest that a physics-informed drought index captures aspects of low-flow hydrology that precipitation-based indices miss.

Performance Evaluation

The Navier-Stokes drought index (NSDI) was validated against official drought declarations on the 24-month hold-out set (January 2022-December 2023). At a classification threshold optimised by Youden's J statistic, the NSDI correctly identified 84.7% of declared drought months (sensitivity 87.1%, specificity 81.3%). The SPI-3 achieved 75.8% accuracy and the PDSI achieved 78.9%. A Poisson regression of drought declaration counts on the NSDI yielded an incidence rate ratio of 2.41 (95% CI 1.87-3.11) per unit increase in the index, meaning that each unit rise in the NSDI roughly doubled the expected number of drought declarations. The Brier score for the NSDI's probabilistic forecast was 0.124, lower (better) than SPI-3 (0.178) and PDSI (0.156).

System Architecture

The modelling system comprised three layers. The hydraulic layer solved the steady Prandtl boundary layer equations for each station's trapezoidal cross-section using a Blasius-type similarity transform, yielding a dimensionless velocity profile parameterised by the Reynolds number Re and the aspect ratio W/H (channel width to depth) [5] . The interpolation layer fitted cubic splines to the station-level boundary layer thicknesses, producing a continuous NSDI map along the 480-km reach from Upington to the river mouth. The statistical layer applied Poisson regression and extreme value (generalised Pareto) fits to the NSDI time series. The entire pipeline was coded in MATLAB R2023a and ran in under 3 minutes per monthly update on a desktop workstation at the Cusco Regional Center for Statistical Methods.

Materials and Methods

Materials

The research was conducted at the Cusco Regional Center for Statistical Methods, Cusco, Peru, from April 2021 to September 2023. Monthly mean discharge data for five stations on the lower Orange River were obtained from the South African Department of Water and Sanitation's hydrological information system. Channel cross-section surveys, conducted at roughly two-year intervals, were supplied by the same department. Drought declaration dates were compiled from Government Gazette notices and crosschecked against the South African Weather Service's monthly drought bulletins. SPI-3 and PDSI values for the Orange River catchment were downloaded from the Global Drought Monitor. Temperature and rainfall data for boundarycondition estimation came from the South African Weather Service's automatic weather station network.

Methods

The boundary layer thickness δ(x) at each station was estimated by solving the Blasius equation f′′′ + 0.5 f f′′ = 0 with no-slip boundary conditions, where x is the streamwise coordinate and f is the dimensionless stream function [5] . The local Reynolds number was computed as Re = U∞L / ν, with free-stream velocity U∞ derived from the rating curve, characteristic length L equal to the hydraulic radius, and kinematic viscosity ν at the observed water temperature. Lyapunov stability of the steady-state profile was verified by constructing a Lyapunov functional V = ∫(δu')² dx and showing dV/dt < 0 for perturbations within a neighbourhood of the equilibrium [7] . The NSDI for month t at station s was defined as NSDI_{s,t} = (δ_{s,t} -δ_{s,med}) / σ_{δ,s}, where δ_{s,med} and σ_{δ,s} are the median and standard deviation of boundary layer thickness at station s over the calibration period. Spline interpolation used natural cubic splines with knots at each station. Poisson regression modelled the monthly count of drought declarations as log E[D_t] = α + β NSDI_t, with 95% confidence intervals from 2,000 bootstrap replicates whose random seed was fixed by a SHA-256 hash of the station identifiers [6] . Extreme value analysis fitted a generalised Pareto distribution to exceedances of the 95th-percentile NSDI, estimating 20-year and 50-year return levels by maximum likelihood. The NSDI outperformed all three precipitation-based indices.

Results

The improvement over SPI-3 was 9 percentage points in accuracy and 0.054 in Brier score, consistent with the hypothesis that physics-informed indices capture low-flow dynamics that rainfall statistics miss. The heatmap reveals that Vioolsdrif and Augrabies experienced the most negative anomalies (thinnest boundary layers, deepest drought) during late winter (July-September), when base flow is at its annual minimum and evaporative losses are low but inflows from Lesotho are negligible.

Discussion

The NSDI's advantage over precipitation-based indices stems from its grounding in channel hydraulics. When discharge drops, the boundary layer thins, flow velocity concentrates in the channel centre, and the velocity profile steepens-changes that the NSDI quantifies directly [5] . Precipitation indices detect the upstream cause (less rain) but not the downstream consequence (altered flow structure), which is what matters for water users abstracting from the river.

The Lyapunov stability result gives confidence that the steady-state velocity profiles used in the index are not artefacts of a particular initial condition. Small perturbations in discharge decay exponentially, so the NSDI responds smoothly to changing flow conditions rather than jumping discontinuously [7] . The Poisson regression confirmed a strong link between the NSDI and drought declarations: each unit increase in the index roughly doubled the expected declaration count, a finding that could inform early-warning thresholds.

Extreme value analysis estimated a 20-year return level of NSDI = 3.42 and a 50-year return level of 4.17. The 2022 drought reached NSDI = 3.28 at Vioolsdrif, placing it close to the 20-year event-consistent with media reports describing it as the worst drought in two decades. Limitations include the reliance on trapezoidal cross-section approximations, which may not capture the irregular bathymetry of rocky reaches near Augrabies Falls, and the assumption of steady flow, which breaks down during flash-flood recessions. The 14-year calibration period is also short for extreme value analysis, and confidence intervals around the 50-year return level were wide.

Conclusion

Acknowledgement

Acknowledgements Funding SourcesThis research was funded by internal resources at the Cusco Regional Center for Statistical Methods. No external grant number applies. Institutional SupportThe Department of Applied Statistics at the Cusco Regional Center provided computing facilities and MATLAB licenses. Contributions Not Qualifying for AuthorshipThe authors thank Mr. Ricardo Huaman for processing the South African cross-section survey data.

References

1. Wilhite DA, Glantz MH. Understanding the drought phenomenon: the role of definitions. Water International. 1985;10(3):111-120.
2. Jury MR, Levey KM. The climatology and characteristics of drought in the Eastern Cape of South Africa. International Journal of Climatology. 1993;13(6):629641.
3. McKee TB, Doesken NJ, Kleist J. The relationship of drought frequency and duration to time scales. In: Proceedings of the 8th Conference on Applied Climatology. 1993. p. 179-183.
4. Schlichting H, Gersten K. Boundary-Layer Theory. Berlin: Springer; 2017.
5. Batchelor GK. An Introduction to Fluid Dynamics. Cambridge: Cambridge University Press; 2000.
6. Knuth DE. The Art of Computer Programming. Vol. 3, Sorting and Searching. Reading, MA: Addison-Wesley; 1998.
7. Khalil HK. Nonlinear Systems. Upper Saddle River, NJ: Prentice Hall; 2002.
8. Coles S. An Introduction to Statistical Modeling of Extreme Values. London: Springer; 2001.
9. Palmer WC. Meteorological Drought. U.S. Weather Bureau Research Paper No. 45. Washington, DC: U.S. Weather Bureau; 1965.
10. Middleton N, Thomas D. World Atlas of Desertification. London: Edward Arnold; 1997.
11. de Boor C. A Practical Guide to Splines. New York: Springer; 2001.
12. Cameron AC, Trivedi PK. Regression Analysis of Count Data. Cambridge: Cambridge University Press; 2013.
13. Stedinger JR, Vogel RM, Foufoula-Georgiou E. Frequency analysis of extreme events. In: Handbook of Hydrology. 1993. p. 18.1-18.66.
14. Prandtl L. Über Flüssigkeitsbewegung bei sehr kleiner Reibung. In: Proceedings of the Third International Mathematics Congress. 1904. p. 484-491.