Using Physiography as a Lens to Understand Stream Network Expansion and Contraction Across Spatiotemporal Scales

Comment #228 Jeff Prancevic @ 2025-07-31 14:13

This preprint analyzes a valuable new dataset, measuring the permanence of streamflow at dozens of sensor locations and exploring possible controls on local wetting and drying. However, some of its conclusions about stream network expansion and contraction appear unfounded.

One of the main findings presented in this preprint is that an existing model for predicting stream network expansion and contraction (from Prancevic & Kirchner, 2019) performs poorly in low-relief upland landscapes. The manuscript purports to compare the observed variability in the length of three small stream networks with their predicted variability, based on the model of Prancevic & Kirchner (2019). However, it appears that both the observed and predicted values are flawed, and apart from these, there is otherwise no data supporting the main finding of the preprint.

Observed length variability:
It appears that stream network lengths were never directly measured as part of this study. Instead, length variability is inferred from time-series measurements of 20+ wet/dry sensors placed throughout each of the three small stream networks, and wet/dry conditions are extrapolated to the surrounding reaches. It is not demonstrated that the sensor locations are representative of the full extent of the stream networks.

In fact, the time series data suggest that these sensors do not accurately capture the variability in network length for these sites. Based on Figure 4, the sensor networks in the Piedmont and Coastal Plain are maximally saturated for entire seasons and record no change in stream network length for several months during the wet season (the flat lines in Fig. 4). This seems implausible and may result from the fact that the sensors cannot detect network expansion and contraction in reaches upstream of the sensor locations. The authors come to the same conclusion in Section 3.4 (see Figure 8) but appear to use the flawed data from Figure 4 for calculating length variability, regardless. This error likely results in a significant underestimation of length variability.


Predicted length variability:
Predictions of the variability of nearly all U.S. stream networks were previously published by Prancevic et al. (2025) using the model from Prancevic & Kirchner (2019). That published dataset includes model predictions for the USGS HUC-12 basins that contain the three stream networks examined in this preprint, and is publicly available here (huc_topo.gpkg): https://www.hydroshare.org/resource/d4b4922f81c44ccabfd2753111fbfad2/

Prancevic et al. (2025) is cited in the preprint, but the published variability values for the basins containing these three stream networks are not mentioned. Instead, the authors attempted to replicate the model from Prancevic & Kirchner (2019). The values of the expansion exponent (see “Background” section below) calculated by the authors differ significantly from those published in Prancevic et al., 2025. Below, I compare the expansion coefficients published by Prancevic et al., 2025 (the first of the two numbers) with those in the preprint (the second of the two numbers):
Appalachian Plateau network (in HUC 060300020103): 0.12 vs. 0.083
Piedmont network (in HUC 031501060501): 0.16 vs. -0.12
Coastal Plain network (in HUC 031601070306): 0.26 vs. 0.49
Notably, the Piedmont network is not predicted to have a negative expansion exponent as reported in the preprint.

The expansion exponents published in Prancevic et al. (2025) are closer to the preprint's “observed” values than the predicted values reported in the preprint. For example, the time series of stream network length for the Appalachian Plateau is the only one that looks realistic (Fig. 4A), and this case, the “observed” expansion exponent is nearly identical to the value reported in Prancevic et al. (2025) (0.122 vs. 0.128, based on Table S2 and data from the link above).

While the HUC-12 basins examined in Prancevic et al. (2025) include the basins examined in the preprint, they average over a much larger area and include parts of the landscape beyond the study basins themselves. This could explain some of the discrepancy between model predictions but it is unlikely to result in the large differences seen here. It may be that the calculations performed for the preprint overlook important methodological nuances from the Prancevic & Kirchner (2019) model. The preprint contains insufficient details about the calculations to reconcile the differences.


Suggestions:
Issues with the predicted variability can be corrected simply by using the values that are already published for these networks in Prancevic et al. (2025). However, adapting measurements from the wet/dry sensor networks to accurately measure length seems more challenging. To my knowledge, it has not been demonstrated that this is a reliable way to measure the variability in stream network length. All of the networks used to calibrate the model in Prancevic & Kirchner (2019) were mapped by walking the entire length of the stream network during multiple wetness conditions.


Below, for additional context, I quote the findings from the preprint text. I also provide some additional background about the expansion exponent and the model that is being used to predict it.

The purported findings, quoted directly from the manuscript, are:
Key Point #1: “Contrary to existing perceptual models, stream network expansion/contraction patterns in low-relief watersheds are not driven by topography.”

Abstract (sentences 6 and 8): “Network expansion and contraction was driven by a combination of physiographic variables, and existing topography-based methods of predicting network expansion and contraction performed poorly.” And “This study demonstrates that low-relief stream networks do not conform to existing topography-based perceptual models of network expansion and contraction, and that consideration of other factors such as soils and vegetation are required to explain network expansion and contraction in these ubiquitous landscapes.”

Conclusions (sentences 4 and 6): “Our study found that the low-relief systems in the Southeastern US did not align with previous observations linking topography and network expansion and contraction patterns.” And “Previously derived relationships between topography and network connectivity overestimated the magnitude of expansion and contraction in our two lower-relief watersheds.”

The only topography-based model to predict expansion and contraction of stream networks is that presented in Prancevic & Kirchner (2019), despite the preprint's reference to plural “models” and “relationships”.


Background:
The model of Prancevic & Kirchner (2019) uses topographic patterns in headwater valleys to estimate the power-law exponent that relates total length of the flowing stream network to streamflow measured at the outlet of the basin. The expansion exponent (typically symbolized as β) is a powerful predictive tool because it quantifies the sensitivity of a given stream network’s length to changes in landscape wetness. This exponent (β) is the variable used in the preprint to assess the performance of the Prancevic & Kirchner 2019 model in the three study basins examined in this study (Fig. 3 and Table S2 of the preprint).

The model of Prancevic & Kirchner (2019) was calibrated using direct observations of the expansion and contraction of 17 stream networks. These networks were painstakingly mapped by walking twice the entire length of the stream networks at multiple wetness conditions by previous investigators and published in other articles (see references in Prancevic & Kirchner, 2019). None of the calibration studies used wet/dry sensor networks to approximate stream network length.