pith. sign in

arxiv: 2606.10224 · v1 · pith:YMJZNC26new · submitted 2026-06-08 · 📊 stat.ME · stat.AP

Spatial Prediction of Local Soil Erosion Distribution in the Wasserstein Space

Jiaming Qiu , Xiongtao Dai , Zhengyuan Zhu , Shuiqing Yin This is my paper

Pith reviewed 2026-06-27 15:10 UTC · model grok-4.3

classification 📊 stat.ME stat.AP
keywords soil erosionspatial predictionWasserstein spaceKrigingdistribution regressionbasis expansionlocal regressionexceedance probability
0
0 comments X

The pith

Treating local soil erosion distributions as objects in Wasserstein space and mapping them to square-integrable trajectories enables Kriging-based spatial prediction at arbitrary locations.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a spatial prediction method that represents local erosion distributions as points in the Wasserstein space. These distributions are converted into square-integrable trajectories through basis expansion, which creates a multivariate random field that encodes spatial dependence. Local regression and Kriging are then performed directly on this field to generate predicted distributions at unsampled sites. The approach yields improved accuracy for derived quantities such as distribution means and exceedance probabilities. Simulation results show gains over both misspecified parametric models and existing Fréchet regression methods, and the framework is applied to extend watershed measurements across Shaanxi province using land-use and elevation covariates.

Core claim

Local erosion distributions are treated as objects in the Wasserstein space, mapped into square-integrable trajectories, and expanded in a basis to produce a multivariate random field; local regression and Kriging on this field then produce distribution predictions at new locations that improve accuracy for functionals such as means and exceedance probabilities.

What carries the argument

Wasserstein-space representation of distributions, converted to square-integrable trajectories via basis expansion to form a multivariate random field on which local regression and Kriging operate.

If this is right

  • Predicted distributions at new locations yield more accurate estimates of mean erosion and of probabilities that erosion exceeds any chosen threshold.
  • The same mapped random-field representation supports direct comparison with existing Fréchet regression methods and shows lower error in simulations.
  • Covariates such as land use and elevation can be incorporated through the local regression step to refine predictions across an entire province.
  • The method extends fieldwork measurements from a small number of watersheds to distribution-valued predictions over large unsampled domains.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same trajectory representation could be used for other environmental variables whose local distributions are observed at scattered sites, such as rainfall intensity or pollutant concentrations.
  • Choice of basis functions other than the one used here might further reduce the number of trajectories needed while still preserving spatial structure for Kriging.
  • If the Wasserstein embedding is replaced by an alternative metric on distributions, the downstream Kriging step would require corresponding adjustments to the covariance model.

Load-bearing premise

The basis expansion of the mapped distributions preserves enough spatial dependence structure that Kriging produces accurate out-of-sample predictions of the full distributions.

What would settle it

In a hold-out validation on surveyed watersheds, the Wasserstein distance between predicted and observed erosion distributions at unsampled sites exceeds the distance achieved by a correctly specified parametric spatial model.

Figures

Figures reproduced from arXiv: 2606.10224 by Jiaming Qiu, Shuiqing Yin, Xiongtao Dai, Zhengyuan Zhu.

Figure 1
Figure 1. Figure 1: Left panel: the province boundary (polygon) and surveyed sampling units (dots). Right [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: A visualization for the proposed model, where the quantities are [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: True and the predicted densities using a local Fréchet regression and the proposed [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Flow chart for pre-processing, modeling, and prediction of erosion. [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Province-wide prediction map for local erosion distributions. [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Uncertainty in predicting local erosion mean per land type, quantified by the width of [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: County-level erosion severity. Each polygon represents one county in the Shaanxi [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Mode of variation in terms of quantile functions of the working variable (effectively [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Comparing predicted densities of the log-normal model (solid curves) and the proposed [PITH_FULL_IMAGE:figures/full_fig_p016_9.png] view at source ↗
read the original abstract

Obtaining precise erosion measurements requires costly fieldwork, making it infeasible to directly survey large domains such as a province or river basin. To extend fieldwork results across such extensive domains, we propose a novel spatial prediction method that treats local erosion distributions as objects in the Wasserstein space. These distributions are mapped into square-integrable trajectories and represented via basis expansion, forming a multivariate random field that captures spatial dependence. By applying local regression and Kriging in this representation, our approach flexibly models and predicts erosion distributions at arbitrary locations. This framework improves prediction for functionals of the distribution, such as the mean and exceedance probabilities. Simulation studies demonstrate that the proposed method outperforms a misspecified parametric alternative and existing Fr\'echet regression approaches. We illustrate the approach with a detailed erosion analysis in Shaanxi province, China, where local measurements from surveyed watersheds are extended to predict erosion distributions across the entire province using covariates such as land use and elevation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a spatial prediction method for local soil erosion distributions by embedding them as objects in the Wasserstein space. These distributions are mapped into square-integrable trajectories via basis expansion to form a multivariate random field, after which local regression and Kriging are applied in the coefficient space to predict full distributions (and their functionals such as means and exceedance probabilities) at unsampled locations. Simulation studies are claimed to show outperformance relative to a misspecified parametric model and existing Fréchet regression methods; the approach is illustrated on erosion data from surveyed watersheds in Shaanxi province, China, using covariates including land use and elevation.

Significance. If the linear L2 embedding via basis expansion transmits the original spatial dependence structure intact, the framework would extend functional-data and distributional-data techniques to spatial prediction problems in environmental statistics, offering a route to predict nonlinear functionals without parametric assumptions on the erosion distributions themselves.

major comments (2)
  1. [Abstract] Abstract: the central claim of simulation outperformance and improved prediction of functionals rests on an unstated mapping from Wasserstein distributions to L2 trajectories followed by basis expansion and Kriging; no equations, truncation rules, or error metrics are supplied, so the claim cannot be evaluated from the provided text.
  2. [Method] § (method description): the weakest assumption—that the basis expansion of the mapped quantile functions (or equivalent) preserves the spatial covariance structure sufficiently for second-order Kriging to recover accurate out-of-sample distributions—is not tested; standard Kriging on coefficients does not automatically respect Wasserstein geometry or non-Euclidean dependence, and no diagnostic or sensitivity check is described.
minor comments (2)
  1. [Abstract] Abstract and throughout: the phrase "square-integrable trajectories" should be replaced by an explicit statement of the representation (e.g., quantile functions on [0,1] or cumulative distribution functions) together with the chosen basis (Fourier, B-spline, etc.) and truncation level.
  2. [Simulations] Simulation section: the parametric alternative is described only as "misspecified"; the precise data-generating process, sample sizes, and quantitative error measures (e.g., Wasserstein distance, integrated squared error on exceedance probabilities) must be stated so that the reported superiority can be reproduced.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for these constructive comments. We address each major point below and indicate planned revisions to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim of simulation outperformance and improved prediction of functionals rests on an unstated mapping from Wasserstein distributions to L2 trajectories followed by basis expansion and Kriging; no equations, truncation rules, or error metrics are supplied, so the claim cannot be evaluated from the provided text.

    Authors: The abstract is a high-level summary and therefore omits technical specifics. The full manuscript details the quantile-function mapping to L2 trajectories (Section 2.1), the basis expansion with truncation at 95% cumulative variance (Section 2.2), the multivariate Kriging step (Section 3), and the simulation error metrics (Wasserstein distance and functional errors in Section 4). To improve self-containment we will revise the abstract to briefly reference the embedding, basis expansion, and performance metrics used. revision: yes

  2. Referee: [Method] § (method description): the weakest assumption—that the basis expansion of the mapped quantile functions (or equivalent) preserves the spatial covariance structure sufficiently for second-order Kriging to recover accurate out-of-sample distributions—is not tested; standard Kriging on coefficients does not automatically respect Wasserstein geometry or non-Euclidean dependence, and no diagnostic or sensitivity check is described.

    Authors: We agree this assumption merits explicit verification. The simulation results demonstrate superior out-of-sample recovery of distributions and functionals relative to alternatives, providing indirect support, yet we did not report direct diagnostics (e.g., covariance-matrix comparisons pre- and post-embedding) or sensitivity to basis truncation. We will add a dedicated subsection with such checks, including Frobenius-norm differences in empirical covariances and cross-validation over basis dimension. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The derivation maps erosion distributions into Wasserstein space, applies basis expansion to obtain square-integrable trajectories, treats the coefficients as a multivariate random field, and then performs established local regression and Kriging. None of these steps reduce the claimed out-of-sample predictions (including functionals such as exceedance probabilities) to quantities defined by the same data or by self-citation chains; the central claim rests on the empirical performance of standard geostatistical tools applied to the transformed representation rather than on any tautological redefinition or fitted-input-as-prediction construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Based on abstract only; central claim rests on the domain assumption that Wasserstein geometry plus basis expansion yields a representation in which spatial dependence can be modeled by standard Kriging without loss of key distribution features.

axioms (1)
  • domain assumption Erosion distributions can be embedded into square-integrable trajectories via basis expansion while retaining sufficient structure for spatial dependence modeling
    Invoked when the abstract states distributions are mapped into trajectories forming a multivariate random field.

pith-pipeline@v0.9.1-grok · 5694 in / 1246 out tokens · 36974 ms · 2026-06-27T15:10:29.057944+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

34 extracted references · 22 canonical work pages

  1. [1]

    and IRPINO, A

    BALZANELLA, A. and IRPINO, A. (2020). Spatial Prediction and Spatial Dependence Monitoring on Georeferenced Data Streams.Statistical Methods & Applications29101–128. https://doi.org/10.1007/ s10260-019-00462-0

    2020

  2. [2]

    and MÜLLER, H.-G

    BHATTACHARJEE, S. and MÜLLER, H.-G. (2023). Single Index Fréchet Regression.The Annals of Statistics51 1770–1798. https://doi.org/10.1214/23-AOS2307

    work page doi:10.1214/23-aos2307 2023

  3. [3]

    and LÓPEZ, A

    BIGOT, J., GOUET, R., KLEIN, T. and LÓPEZ, A. (2017). Geodesic PCA in the Wasserstein Space by Con- vex PCA.Annales de l’Institut Henri Poincaré, Probabilités et Statistiques531–26. https://doi.org/10.1214/ 15-AIHP706 MR3606732

    2017

  4. [4]

    CRESSIE, N. A. C. (1993).Statistics for Spatial Data.Wiley Series in Probability and Statistics. John Wiley &

    1993

  5. [5]

    https://doi.org/10.1002/9781119115151

    Sons, Inc., Hoboken, NJ, USA. https://doi.org/10.1002/9781119115151

    work page doi:10.1002/9781119115151

  6. [6]

    DUNSON, D. B. and PARK, J.-H. (2008). Kernel Stick-Breaking Processes.Biometrika95307–323. https: //doi.org/10.1093/biomet/asn012

    work page doi:10.1093/biomet/asn012 2008

  7. [7]

    FARR, T. G. and KOBRICK, M. (2000). Shuttle Radar Topography Mission Produces a Wealth of Data.Eos, Transactions American Geophysical Union81583–585. https://doi.org/10.1029/EO081i048p00583

    work page doi:10.1029/eo081i048p00583 2000

  8. [8]

    GELFAND, A. E., ed. (2010).Handbook of Spatial Statistics.Chapman & Hall/CRC Handbooks of Modern Statistical Methods. CRC Press, Boca Raton

    2010

  9. [9]

    E., KOTTAS, A

    GELFAND, A. E., KOTTAS, A. and MACEACHERN, S. N. (2005). Bayesian Nonparametric Spatial Modeling With Dirichlet Process Mixing.Journal of the American Statistical Association1001021–1035. https://doi. org/10.1198/016214504000002078

    work page doi:10.1198/016214504000002078 2005

  10. [10]

    Electronic Journal of Statistics , volume =

    GHOSAL, A., MEIRING, W. and PETERSEN, A. (2023). Fréchet Single Index Models for Object Response Re- gression.Electronic Journal of Statistics171074–1112. https://doi.org/10.1214/23-EJS2120

    work page doi:10.1214/23-ejs2120 2023

  11. [11]

    HE, X. (1997). Quantile Curves without Crossing.The American Statistician51186–192. https://doi.org/10. 1080/00031305.1997.10473959

    Pith/arXiv arXiv 1997

  12. [12]

    and FILZMOSER, P

    HRON, K., MENAFOGLIO, A., TEMPL, M., HR ˚UZOVÁ, K. and FILZMOSER, P. (2016). Simplicial Principal Component Analysis for Density Functions in Bayes Spaces.Computational Statistics & Data Analysis94 330–350. https://doi.org/10.1016/j.csda.2015.07.007 19

    work page doi:10.1016/j.csda.2015.07.007 2016

  13. [13]

    and PENG, L., eds

    KOENKER, R., CHERNOZHUKOV, V., HE, X. and PENG, L., eds. (2017).Handbook of Quantile Regression. Chapman and Hall/CRC, New York. https://doi.org/10.1201/9781315120256

    work page doi:10.1201/9781315120256 2017

  14. [14]

    and REIMHERR, M

    KOKOSZKA, P. and REIMHERR, M. (2017).Introduction to Functional Data Analysis.Texts in Statistical Science Series. CRC Press, Taylor & Francis Group, Boca Raton

    2017

  15. [15]

    Y., ZHANG, K

    LIU, B. Y., ZHANG, K. L. and XIE, Y. (2002). An Empirical Soil Loss Equation. InProcess of Soil Erosion and Its Environment EffectII21–25. Tsinghua University Press, Beijing

    2002

  16. [16]

    and LIU, X

    LIU, B., GUO, S., LI, Z., XIE, Y., ZHANG, K. and LIU, X. (2013). Sample Survey on Water Erosion in China. Soil and Water Conservation in China1026–34

    2013

  17. [17]

    and CHI, W

    SHI, X., JIANG, N., YU, D., PAN, X. and CHI, W. (2014). Spatiotemporal Characteristics, Patterns, and Causes of Land-Use Changes in China since the Late 1980s.Journal of Geographical Sciences24195–210. https://doi.org/10.1007/s11442-014-1082-6

    work page doi:10.1007/s11442-014-1082-6 2014

  18. [18]

    and GIRALDO, R., eds

    MATEU, J. and GIRALDO, R., eds. (2021).Geostatistical Functional Data Analysis.Wiley Series in Probability and Statistics. Wiley, Hoboken, NJ

    2021

  19. [19]

    MENAFOGLIO, A. (2021). Spatial Statistics for Distributional Data in Bayes Spaces: From Object-Oriented Krig- ing to the Analysis of Warping Functions. InAdvances in Compositional Data Analysis: Festschrift in Honour of Vera Pawlowsky-Glahn(P. Filzmoser, K. Hron, J. A. Martín-Fernández and J. Palarea-Albaladejo, eds.) 207–224. Springer International Publi...

    work page doi:10.1007/978-3-030-71175-7_11 2021

  20. [20]

    and SECCHI, P

    MENAFOGLIO, A., GUADAGNINI, A. and SECCHI, P. (2014). A Kriging Approach Based on Aitchison Ge- ometry for the Characterization of Particle-Size Curves in Heterogeneous Aquifers.Stochastic Environmental Research and Risk Assessment281835–1851. https://doi.org/10.1007/s00477-014-0849-8

    work page doi:10.1007/s00477-014-0849-8 2014

  21. [21]

    PANARETOS, V. M. and ZEMEL, Y. (2020).An Invitation to Statistics in Wasserstein Space.Springer- Briefs in Probability and Mathematical Statistics. Springer International Publishing. https://doi.org/10.1007/ 978-3-030-38438-8

    2020

  22. [22]

    and MÜLLER, H.-G

    PETERSEN, A. and MÜLLER, H.-G. (2016). Functional Data Analysis for Density Functions by Transformation to a Hilbert Space.The Annals of Statistics44183–218. https://doi.org/10.1214/15-AOS1363 MR3449766

    work page doi:10.1214/15-aos1363 2016

  23. [23]

    Fr \'e chet regression for random objects with Euclidean predictors

    PETERSEN, A. and MÜLLER, H.-G. (2019). Fréchet Regression for Random Objects with Euclidean Predictors. Annals of Statistics47691–719. https://doi.org/10.1214/17-AOS1624 MR3909947

    work page doi:10.1214/17-aos1624 2019

  24. [24]

    Modeling probability density functions as data objects

    PETERSEN, A., ZHANG, C. and KOKOSZKA, P. (2022). Modeling Probability Density Functions as Data Objects. Econometrics and Statistics21159–178. https://doi.org/10.1016/j.ecosta.2021.04.004

    work page doi:10.1016/j.ecosta.2021.04.004 2022

  25. [25]

    and SECCHI, P

    PIGOLI, D., MENAFOGLIO, A. and SECCHI, P. (2016). Kriging Prediction for Manifold-Valued Random Fields. Journal of Multivariate Analysis145117–131. https://doi.org/10.1016/j.jmva.2015.12.006

    work page doi:10.1016/j.jmva.2015.12.006 2016

  26. [26]

    Spatial Prediction of Local Soil Erosion Distri- bution in the Wasserstein Space

    QIU, J., DAI, X., ZHU, Z. and YIN, S. (2026). Supplement to “Spatial Prediction of Local Soil Erosion Distri- bution in the Wasserstein Space”. https://doi.org/10.1214/[providedbytypesetter]

    2026

  27. [27]

    and Fuentes, Montserrat and Dunson, David B

    REICH, B. J., FUENTES, M. and DUNSON, D. B. (2011). Bayesian Spatial Quantile Regression.Journal of the American Statistical Association1066–20. https://doi.org/10.1198/jasa.2010.ap09237

    work page doi:10.1198/jasa.2010.ap09237 2011

  28. [28]

    STEIN, M. L. (1999).Interpolation of Spatial Data: Some Theory for Kriging.Springer Series in Statistics

    1999

  29. [29]

    https://doi.org/10.1007/978-1-4612-1494-6

    Springer, New York. https://doi.org/10.1007/978-1-4612-1494-6

    work page doi:10.1007/978-1-4612-1494-6

  30. [30]

    TUCKER, D. C. and WU, Y. (2025). Partially-Global Fréchet Regression.Statistica Sinica35713–736. VAN DENBOOGAART, K. G., EGOZCUE, J. J. and PAWLOWSKY-GLAHN, V. (2014). Bayes Hilbert Spaces. Australian & New Zealand Journal of Statistics56171–194. https://doi.org/10.1111/anzs.12074

    work page doi:10.1111/anzs.12074 2025

  31. [31]

    XU, S. G. and REICH, B. J. (2023). Bayesian Nonparametric Quantile Process Regression and Estimation of Marginal Quantile Effects.Biometrics79151–164. https://doi.org/10.1111/biom.13576

    work page doi:10.1111/biom.13576 2023

  32. [32]

    and WANG, W.-T

    YIN, S.-Q., WANG, Z., ZHU, Z., ZOU, X.-K. and WANG, W.-T. (2018). Using Kriging with a Heterogeneous Measurement Error to Improve the Accuracy of Extreme Precipitation Return Level Estimation.Journal of Hydrology562518–529. https://doi.org/10.1016/j.jhydrol.2018.04.064

    work page doi:10.1016/j.jhydrol.2018.04.064 2018

  33. [33]

    and LI, Y

    ZHANG, H. and LI, Y. (2021). Unified Principal Component Analysis for Sparse and Dense Functional Data un- der Spatial Dependency.Journal of Business & Economic Statistics1–15. https://doi.org/10.1080/07350015. 2021.1938085

    work page doi:10.1080/07350015 2021

  34. [34]

    ZHEN, L. (2013). The National Census for Soil Erosion and Dynamic Analysis in China.International Soil and Water Conservation Research112–18. https://doi.org/10.1016/S2095-6339(15)30035-6

    work page doi:10.1016/s2095-6339(15)30035-6 2013