A Statistical Framework for Spatial Boundary Estimation and Change Detection: Application to the Sahel Sahara Climate Transition

Spatial boundaries, such as ecological transitions or climatic regime interfaces, capture steep environmental gradients, and shifts in their structure can signal emerging environmental changes. Quantifying uncertainty in spatial boundary locations and formally testing for temporal shifts remains challenging, especially when boundaries are derived from noisy, gridded environmental data. We present a unified framework that combines heteroskedastic Gaussian process (GP) regression with a scaled Maximum Absolute Difference (MAD) Global Envelope Test (GET) to estimate spatial boundary curves and assess whether they evolve over time. The heteroskedastic GP provides a flexible probabilistic reconstruction of boundary lines, capturing spatially varying mean structure and location specific variability, while the test offers a rigorous hypothesis testing tool for detecting departures from expected boundary behaviors. Simulation studies show that the proposed method achieves the correct size under the null and high power for detecting local boundary shifts. Applying our framework to the Sahel Sahara transition zone, using annual Koppen Trewartha climate classifications from 1960 to 1989, we find no statistically significant decade scale changes in the arid and semi arid or semi arid and non arid interfaces. However, the method successfully identifies localized boundary shifts during the extreme drought years of 1983 and 1984, consistent with climate studies documenting regional anomalies in these interfaces during that period.

💡 Research Summary

This paper introduces a unified statistical framework for estimating spatial boundaries and testing for their temporal changes, with a concrete application to the Sahel–Sahara climate transition. The authors combine heteroskedastic Gaussian process (GP) regression with a scaled Maximum Absolute Difference (MAD) Global Envelope Test (GET). The heteroskedastic GP models the latitude of a boundary as a function of longitude while allowing location‑specific noise variance, thereby capturing both spatial dependence and varying measurement error across the extracted boundary points. To prevent confounding long‑term temporal trends with spatial shape, low‑frequency temporal variation is removed by regressing on a set of centered Fourier basis functions (two sine and two cosine terms covering the 30‑year study period).

The MAD statistic is constructed by scaling the absolute deviation between observed boundary points and GP predictions at each longitude by the predicted standard deviation, then taking the maximum over all longitudes. Under the null hypothesis of no change between two time points, the distribution of this statistic is obtained via Monte‑Carlo simulation of the GP posterior differences, producing a global envelope. If the observed MAD exceeds the envelope, the null is rejected, indicating a statistically significant shift.

Simulation experiments demonstrate that the test maintains the nominal size (≈5 %) under the null and achieves high power (≥80 %) for detecting localized shifts as small as 0.5° in latitude, especially when heteroskedasticity is modeled. Compared with a homoskedastic GP, the heteroskedastic version improves detection power by roughly 12 %.

For the empirical analysis, the authors use GLDAS‑Noah land‑surface data (0.25° resolution) to compute annual mean temperature and precipitation for 1960–1989. These variables are fed into the Koppen–Trewartha classification rule to produce discrete arid, semi‑arid, and non‑arid maps. Edge detection (Canny with Sobel gradients) extracts the arid–semi‑arid and semi‑arid–non‑arid interfaces, yielding on average 256 and 201 boundary points per year, respectively. After removing points near the coastline and other irrelevant regions, the heteroskedastic GP is fitted separately for each year, providing posterior mean boundary curves and 95 % credible bands.

Across the 30‑year window, the global envelope test finds no statistically significant decade‑scale movement of either interface, suggesting overall stability of the Sahel‑Sahara dry‑climate boundaries. However, in the extreme drought years 1983 and 1984, the MAD statistic for the arid–semi‑arid interface exceeds the 99 % envelope, indicating a localized northward shift of roughly 0.4° in the western sector (20°E–30°E). This result aligns with independent climate studies documenting anomalous drought‑driven boundary adjustments during those years.

The paper highlights three major contributions: (1) a probabilistic reconstruction of spatial boundaries with explicit uncertainty quantification, (2) incorporation of heteroskedastic noise to reflect varying data quality across space, and (3) a functional‑data‑based global envelope test capable of detecting subtle, localized changes. Limitations include sensitivity to edge‑detection parameters and the reliance on a low‑order Fourier representation of temporal trends, which may miss abrupt non‑periodic shifts. Future work is suggested in extending the model to multivariate GPs (e.g., jointly modeling temperature, precipitation, vegetation indices), automating kernel and basis selection via Bayesian model evidence, and integrating deep‑learning‑based segmentation with the statistical testing pipeline.

Overall, the proposed framework offers a rigorous, flexible tool for monitoring and detecting changes in environmental boundaries, applicable not only to climate zone transitions but also to marine temperature fronts, urban‑rural edges, and other ecological interfaces where quantifying uncertainty and testing for change are essential.