This is a valuable and really interesting dataset; multi-basin, multi-FFG mixing-model estimates of allochthony across a broad climatic gradient are rare, and the non-perennial coverage is especially welcome. I want to raise one concern about the central conclusion that air temperature is the dominant driver, because the Table 2 model comparison may be carrying more interpretive weight than it actually supports.

By the paper's own LOO framework (Vehtari et al. 2017), the top model is not distinguishable from its nearest competitors. Temperature (M7) leads aridity (M9) by ΔLOOic = 1.4 with se_dLOOic = 2.1; C:N (M10) by 2.1 ± 1.6; precipitation (M8) by 2.8 ± 2.4. In each case the standard error of the difference equals or exceeds the difference itself, which suggests these models have indistinguishable expected predictive performance. Temperature carries the most model weight (0.389), but that leaves over 60% distributed across the other models, none of which are clearly beaten by the temperature model. Notably, the basin-only null (M1) is only 5.6 ± 5.6 LOOic behind temperature in the full data (within one SE), so the evidence that any single continuous climate predictor beats the simple nulls can be interpreted as weak.

The sharpest version of this may be temperature versus C:N of allochthonous sources. These predictors are collinear (r = −0.65), both track the allo -> auto shift (Fig. 3A,C), and are statistically tied (2.1 ± 1.6 in Table 2; 0.7 ± 1.4 in Table S4). A bottom-up resource-quality mechanism fits the ranking as well as the consumer-metabolic one, and the two make different, testable predictions.

Two more points regarding the mechanism:

1. The predictor is 800-m PRISM mean annual air temperature, and the metabolic chain invoked (air -> water temperature -> metabolic demand -> GPP -> algal availability; L377–389) is cited rather than measured (no water temperature, stream metabolism, GPP, or feeding measured). While STIC loggers were deployed to measure intermittency, could you use the recorded in-stream temperature measurements (when reaches were flowing) or at least compare them against air temperatures to confirm the air x water temperature relationship the interpretation relies on?

2. An independent line of evidence cuts against temperature as the operative driver. On the south-central Texas precipitation gradient (Kinard 2024, PhD dissertation; Strickland et al. 2026, Ecology), mean annual air temperature is nearly constant across sites while annual precipitation spans roughly 60–135 cm. Despite this, that system reproduces the same endpoint: rising autochthonous assimilation, a shift to fish herbivory, shorter food chains, and wider isotopic niche in the drier (not warmer) streams, with the mechanism attributed to nutrient concentration, flow stability, reduced scour, and light. A near-isothermal moisture gradient yielding the same outcome suggests temperature is not necessary to generate it. (Disclosure: I am a coauthor on Strickland et al.)

None of this is to say climate is unimportant! The thermal/resource-quality axis clearly tracks allochthony better than stream size or FFG alone. The narrower point is that the data support "allochthony declines along a correlated temperature–litter-quality gradient" more than "air temperature is the dominant driver," and the ~15 °C threshold in particular rests on the crossover of two posterior ribbons (Fig. 3A) whose 95% intervals overlap across most of the sampled range. I'd suggest softening the causal language around temperature, or adding an analysis able to separate temperature from C:N.

But ultimately, all of the candidate models perform very similarly and making the case for one model (temperature) being the most important thing in determining allochthony in food webs is difficult based on the model comparisons reported here.

Referenced works if you're interested: Stickland et al: https://doi.org/10.1002/ecy.70501 Kinard: https://scholarworks.wm.edu/server/api/core/bitstreams/e6220cc9-157a-455d-9705-0026df4b8424/content