Both Kcd and Kdd exhibited a wide range of values (Fig. 2), effectively spanning the theoretical maximum of what could have been observed with our deployment setup. This is evidenced by some cotton strips being completely consumed prior to retrieval (5.4 % of all strips, Kcd and Kdd maximized), while others were largely intact (Kcd and Kdd≈ 0). This variation is surprising given the relatively small spatial domain sampled by this study, and emphasizes that environmental heterogeneity can surpass the effects of spatial extent (Mancuso et al., 2022). The environments studied here ranged from pristine locations in the mountainous headwaters of the YRB to lowland locations with heavy agricultural influences (Fig. 1; Laan et al., 2025). This emphasizes the value of environmentally diverse watersheds like the YRB as useful testbeds to study variation in decomposition rates within single hydrologically connected basins.

Comparing decay rates from the YRB to a global dataset from >500 streams and rivers (Tiegs et al., 2024) showed that YRB rates spanned nearly the entire global range and had substantial overlap with the bulk of the global distributions (Fig. 2a and b). Dominant peaks in the YRB rate distributions were shifted slightly towards faster rates, relative to primary peaks in the global dataset (Fig. 2a and b). This shift towards faster rates and the wide range in rates are likely due to several factors linked to rates being estimated in later summer with high temperatures, and established biological communities due to several months since high-flow disturbances (Collier et al., 2013b; Grimm and Fisher, 1989; Mancuso et al., 2023). The two decay rates were also closely correlated with each other, though the relationship weakened towards locations with faster decay rates (Fig. 2c). This suggests a weak influence of temperature in the YRB; a strong influence of temperature should lead to a weak relationship between Kcd and Kdd. These results contrast previously reported influences of temperature on particulate organic-matter decomposition (Benbi et al., 2014; Griffiths and Tiegs, 2016), and likely reflect dominance of other influential factors such as variation in nutrient concentrations (Rosemond et al., 2015). Temperature-driven decomposition is also expected to lead to a strong relationship between Kcd and summed temperature, but we observed a very weak relationship despite ∼4-fold variation in summed temperature (Fig. 2d). This range in temperature among sites would likely be smaller in other seasons, and we do not therefore expect a strong influence of temperature to emerge in the YRB by conducting the study in other seasons. These results emphasize the need to understand factors governing variation in decay rates across the YRB. This is especially true given that rate distributions within this one basin span nearly all globally observed decay rates.

Across the YRB there appears to be potential for some spatial organization for both Kcd and Kdd (Fig. 3). Visual inspection of the maps suggests that the spatial organization may be stronger for Kcd than for Kdd. To evaluate this possibility more rigorously, we regressed each decay rate against upstream drainage area (Fig. 4). In this case, drainage area is meant to reflect position within the YRB. We used drainage area in preference to stream order because it is a continuous variable directly tied to the spatial domain a given stream integrates, whereas stream order is categorical and primarily reflects stream network topology. Associated regressions were significant (p<0.05) with both decay rates increasing with drainage area (Fig. 4). The relationship with drainage area was stronger, in terms of R2, for Kcd. Both relationships were, however, relatively weak with R2 values of 0.22 and 0.14 for Kcd and Kdd, respectively (Fig. 4). Nonetheless, the existence of a significant relationship after controlling for temperature (i.e., for Kdd) indicates that spatially structured factors other than temperature influence decay rates. This is not surprising as studies using cotton strips have found several factors that influence decomposition, such as nutrient concentrations, turbidity, and many others (Collier et al., 2013b; Pingram et al., 2020; Tiegs et al., 2024). Before exploring a broad suite of potential explanatory variables, we tested our hypothesis that decomposition rates will be better explained by sediment-associated respiration (ERsed) than by respiration in the water column (ERwc) or by respiration of the integrated stream system (ERtot).

Contrary to our hypothesis, we found that both decay rates were most strongly connected to ERtot, less so with ERsed, and not at all with ERwc (Fig. 5). Univariate models using ERtot were better than multivariate models using ERsed, ERwc, and their interaction. This is evidenced by univariate models using ERtot having AIC values more than two units lower than multivariate models containing ERsed and ERwc (Table 1). Multivariate models were not used with ERtot because it contains ERsed and ERwc. Garayburu-Caruso et al. (2025) showed that ERsed accounted for the majority of ERtot across most of the YRB, with 88 % of locations showing ERsed contributions exceeding 50 % of ERtot. The strong correspondence between these two metrics might explain the similar R2 values observed. We note, however, that the difference in explained variance between ERtot (R2=0.29 for Kcd; R2=0.16 for Kdd) and ERsed (R2=0.22 for Kcd; R2=0.13 for Kdd) was modest, and the stronger association with ERtot should be interpreted cautiously.

To interpret these results, we note that cotton strips were deployed a few centimeters into the riverbed sediments, beneath a brick set flush with the streambed surface. We conceptualized this deployment as the shallow hyporheic zone, which was a reason we hypothesized that decay rates would be most strongly connected to ERsed. However, the results indicate that point-scale decomposition of particulate organic matter within the shallow hyporheic zone is linked to respiratory processes occurring in both the sediment and water column. Therefore, our deployment depth might have reflected sediment-water interface processes in addition to shallow hyporheic zone processes, leading to stronger associations of cotton strip decomposition with ERtot. We also note that ERsed estimates may not exclusively reflect biological respiration, as non-respiratory O2-consuming processes can contribute to these estimates (see Methods). We propose that if our deployment configuration was complemented with a simultaneous deployment that enabled growth of benthic algal biofilms on the cotton strips, the combined decomposition from both deployments could, in some systems, capture more of the processes that contribute to ERsed. This would provide a more complete view of sediment-associated biogeochemical function, potentially leading to a stronger correlation between decomposition rates and ERsed. While this remains to be tested, the underlying idea is that primary producers support a large portion of respiration associated with riverbed sediments, which is supported by recent analyses showing a strong link between ERsed and gross primary production across the YRB (Garayburu-Caruso et al., 2025).

Given that our hypothesis was rejected and that ERtot explained only about 29 % and 16 % of Kcd and Kdd, respectively (Table 1 and Fig. 5), we used a discovery-based approach to explore other system variables that may explain further variation in decomposition rates. LASSO-based modeling indicated that total dissolved nitrogen (TDN) and the median grain size of sediment texture (D50) were most important for explaining Kdd, while TDN and aridity were most important for explaining Kcd (Table 2). Other variables were retained in the LASSO models (Table S1), but we interpreted TDN, D50, and aridity as the most important because they consistently had the largest normalized coefficients. This interpretation is based on these variables having mean normalized coefficients above 0.5 – in terms of absolute value – meaning they were at least 50 % as important as the most important variable in the 100 LASSO model runs. Further, the LASSO coefficients for these variables had a coefficient of variation less than 0.5, meaning that across the 100 LASSO model runs the values of their normalized coefficients were relatively stable (Table S1). The LASSO modeling also confirmed a relatively weak influence of temperature, evidenced by relatively small and highly variable β coefficients for summed temperature in the Kcd model (Table S1); temperature was not used in the Kdd model.

Both decomposition rates increased with higher TDN concentrations, while Kdd decreased with larger D50 and Kcd decreased with higher aridity index values (Table 2). To more deeply interpret these relationships, we examined Pearson-based univariate correlations between these three explanatory variables and other variables included in the LASSO models. This is important because of strong collinearity among some explanatory variables (Fig. S1 in the Supplement). In this case, variables identified as being the most important may be acting as proxies for one or more additional variables. We found that TDN was most strongly correlated with percent agricultural land cover of the upstream drainage area and sulfate concentrations in the stream water (Fig. S1). Increases in both decomposition rates with TDN may, therefore, reflect agricultural inputs of nutrients that increase overall microbial activity of the stream ecosystems we studied. D50 was most strongly correlated with the aridity index, which was most strongly correlated with percent forest cover; the correlation between D50 and aridity is unlikely to reflect a causal connection, while aridity and forest cover most likely are causally linked. If the relationship between decomposition and D50 is causal, it could be mediated by the total surface area available for microbial attachment. This would, however, influence decomposition only to the extent that microbial biomass in adjacent sediments impacts microbial activity on the cotton strips. Coarser sediments have much less surface area, potentially limiting overall microbial activity.

Considering the directionality of the univariate relationships, in context of the LASSO outcomes, suggests slower decomposition – for both rates – in streams with relatively coarse sediments and set within relatively wet forests. This contrasts with Clapcott and Barmuta (2010) who found faster decomposition in coarser sediments. The discrepancy is likely because we excluded macroinvertebrates while they did not, and they interpreted the link to sediment texture as due to greater habitat availability for macroinvertebrates in coarser sediments. Locations with slower decomposition should, thus, primarily be in higher elevation, relatively pristine parts of the YRB, while faster decomposition occurs at lower elevations impacted by agricultural inputs. These results are consistent with previous work showing greater cotton strip decomposition in impaired streams (Young and Collier, 2009), those with little natural land cover (Collier et al., 2013a; Webb et al., 2019), and in streams with higher nutrient concentrations (Ferreira et al., 2015; Pingram et al., 2020; Tiegs et al., 2013). In addition to differences in nutrient concentrations between higher and lower elevation sites, we expect less light penetration to streams in higher elevation sites because of more forest cover and smaller streams. Though not measured here, less light could suppress autotrophic production which may limit heterotrophic respiration (Bernhardt et al., 2022; Mulholland et al., 2001; Young and Huryn, 1999). Lower autotrophic production could therefore slow decomposition relative to high-light conditions by reducing the supply of labile carbon from phototrophs that can prime organic-matter degradation (Danger et al., 2013; Howard-Parker et al., 2020).

Our results collectively indicate that to study shallow hyporheic zone decomposition processes, it is not sufficient to conceptualize organic-matter decomposition by only considering sediment or water column processes; one must examine the integrated system. The implication of our analyses is that point-scale organic-matter decomposition potential was more closely associated with integrated system respiration than with individual respiration components. Our findings suggest that mechanistic models of stream ecosystem respiration may benefit from accounting for sediment processes, water-column processes, and land-cover influences from beyond the stream. Focusing exclusively on the hyporheic zone is insufficient, even in systems for which the hyporheic zone accounts for most reactions (Boano et al., 2014; Burrows et al., 2017; McClain et al., 2003). This is further emphasized by previous work showing that most respiration occurs in the water column of large rivers (Gardner and Doyle, 2018; Roley et al., 2023). Garayburu-Caruso et al. (2025) also show that fractional contributions of ERsed to ERtot is often high, but that there is significant variation in those fractional influences across the YRB. This variability is due to ERwc being fast enough, in some locations, to account for more than 80 % of ERtot (Garayburu-Caruso et al., 2025). Similarly, Laan et al. (2025) found substantial overlap between the distribution of ERwc rates from the YRB and ERtot from across the contiguous United States. The overall picture is that decomposition is the result of integrated processes occurring across the sediment-water continuum and influenced by external factors tied to land cover and land use. We infer that these integrated processes are influenced by biophysical variation across the YRB (Laan et al., 2025), leading to decomposition rates within this single basin that resemble global rate distributions and nearly span the global range of observed rates (Tiegs et al., 2024). Other basins that contain only one ecoregion or have homogeneous land cover may be expected to have a narrower range of decomposition rates (Webb et al., 2019). We note that our decomposition estimates represent point-scale conditions at wadeable locations and should be interpreted as such rather than as reach-integrated measures of decomposition. Capturing within-reach spatial variability in decomposition across habitats and channel depths would complement the among-reach approach used here, and may lead to stronger relationships between decomposition rates and reach-scale ER. Nonetheless, models applied to any stream network that aim to predict spatiotemporal variation in decomposition rates would be well served by considering processes throughout the integrated watershed system.