In this manuscript, the authors present a substantially long time-series of isotopic and hydrographic measurements to understand carbon cycling patterns along the Sihl River and basin. The strength of this manuscript is its clear goal and approach. The study uses a mixture of "traditional" methods to measure carbon cycling (POC and DOC collection via filtration, and thorough subsequent geochemical analysis (2.3)) and new computational methods such as machine learning. Many of the terms / methods described in 2.5 and 2.6 are new to me as someone largely unfamiliar with machine learning algorithms; however, this does not negatively impact the article's "result traceability", and there is significant precedence provided for each decision via citation of previous literature. 

One minor comment / question (Line 132). I'm unfamiliar with this method of storing DOC samples, and am surprised they were not frozen. I'm wondering if the effects of acidification versus freezing was considered in the method, or interpretation of results? See Walker et al., 2016 https://doi.org/10.1017/RDC.2016.48

Overall, I find this manuscript very strong and support straightforward publication. 

This manuscript presents a large dataset from a long-term (40 months) sampling campaign of river water and suspended sediment from a subalpine catchment in Switzerland. The authors generated a substantial amount of data, including a 40-month time-series of stable carbon isotopes and radiocarbon activity of dissolved and particulate organic carbon. Time-series data sets like this are incredibly valuable to the scientific community, particularly now as our field aims to mechanistically describe the feedbacks between climate change and the global carbon cycle. This manuscript addresses relevant scientific questions (i.e., what controls the magnitude and temporal variability of river organic carbon export?). The main dataset and introduction of statistical approaches are a great contribution to the field. The methods and statistical analyses used in this manuscript are state of the art, particularly the application of EA-IRMS for high throughput 14C measurements and the application of machine learning-based statistical analyses.

Overall, the manuscript is well-written, but there several points that need clarification and revision, as noted in the major points of concern and the detailed comments below. The authors do a nice job of presenting their data and using statistics to describe the distribution of the data, however, it seems that this manuscript is lacking robust interpretation of the statistical results. Based on the introduction of the paper, I expected their results to provide a mechanistic explanation linking geomorphic and hydrologic processes to organic carbon export from small headwater rivers. However, I was not able to take away any new ideas or significant conclusions from their data interpretation and discussion. Additionally, I find that some of the analyses are not entirely appropriate (i.e. the MixSIAR analysis) and should be either removed from the manuscript or redone to reflect appropriate endmember mixing. To make this manuscript of greater interest to the scientific community, the authors should also provide a framework for integrating their statistical results into Earth system models.

In summary, a number of revisions need to be made before this manuscript can be accepted for publication in EGU Biogeosciences. Major points of concern and suggestions for revising the manuscript are detailed in the attached PDF.