hess-2022-342 “A comprehensive assessment of in situ and remote sensing soil moisture data assimilation in the APSIM model for improving agricultural forecasting across the U.S. Midwest” M.Kivi, N.Vergopolan, H.Dokoohaki

Kivi et al (2022) is cited 25 times in the text! While this work is distinctly different to that work, perhaps a single mention at the start of each section, rather than every second and third level sub-section, would be adequate (such as the existing introductory paragraph in each major section). As far as I can tell, and I am no expert in data assimilation techniques or applications, the work presented is solid and thoroughly describes the proper use of the techniques. The authors honestly presented when the technique improved certain predictions, when it made little difference, and when the model performance degraded. While I can see why the extension was required in Figure 4b, as it’s only 3 of the results and the big picture is shown well with the box-and-whiskers, the extension can be placed in the Appendices and mentioned in the text as with free-model results.

Was the “free run” calibrated against any data, or was it just an ensemble of model runs with the parameters randomly assigned from prior distributions? Were they a single run with an arbitrary (or literature) set of values assigned to parameters? How do the modelled results compare to the “final” set of parameters after the full SDA? I did not get the appendices so don’t know if this is covered.

There are a number of questions that the work raises, that may be answered here (or later maybe).

• As the parameters are nudged successively with the SDA procedure, how much do they need to vary until the researcher considers that (a) their a priori range is incorrect, (b) that they cannot be considered a constant, or (c) that some process is too simplified or missing in the numerical model?
• Further to the case of a non-constant model parameter, is there a pattern in the parameter adjustments, e.g., always too high in winter and too low in summer, that indicates a systematic misrepresentation within the model?
• Is there a form of a posteriori distribution of parameter values, e.g., normal versus log-normal versus bi-modal versus uniform, that may indicate systematic model or data errors?
• Is there parameter bias (or trends?) associated with larger underlying groupings, such as soil texture, vertical layering (duplex, gradational, uniform), crop type, or management, that indicate model structure or data limitations?
• Given the use of numerical models is primarily predictive, i.e., what are future potential grain yields or nitrate loss or deep drainage under different management or climate conditions, which set of parameters (or reduced range) do researchers consider stable enough to make such computations?

The References and citations need a lot of purely technical corrections.

• Following reference not cited in text: Akhavizadegan et al (2021); Archontoulis et al (2014, 2020); Balboa et al (2019); Crane-Droesch (2018); Das et al (2020); Dietze et al (2013); Dietzel et al (2016); Flathers and Gessler (2018); Guerif and Duke (2000); Hoffman et al (2020); Jeong et al (2016); Kang et al (2020); van Klompernburg et al (2020); Leng and Hall (2020); Li et al (2014); Martinez-Feria et al (2019); Pasley et al (2021); Puntel et al (2016); Shahhosseini et al (2021); Spijker et al (2021); Wallach et al (2021).
• Following citations not in references: Lu et al (2019); Lu and Steele-Dunne (2019).
• The two articles of Vergopolan et al (2021) are cited with (a) and (b) in the text, but not indicated as such in the references.
• Citation for Chakrabarti et al (2014) is misspelled in the text line 83 and 605.

General comments

The study “A comprehensive assessment of in situ and remote sensing soil moisture data assimilation in the APSIM model for improving agricultural forecasting across the U.S. Midwest”,  M.Kivi, N.Vergopolan, H.Dokoohaki  comprises of a set of new approaches and  techniques of data assimilation for improving agricultural forecasting. In this study authors integrated in situ and remote sensing soil moisture observations with APSIM model through sequential data assimilation and assessed the extent to which soil moisture data assimilation can improve APSIM model forecasts. Therefore, paper addresses relevant scientific questions within the scope of HESS. The scientific methods and assumptions present are valid. The title clearly reflects the contents of the paper and abstract provides a complete summary of the research.

The work showed that assimilation of in situ surface soil moisture is not as powerful as the assimilation of in situ root-zone soil moisture in terms of model constraint. It is shown that high temporal resolution due to multisensor satellite availability and accurately estimated observation uncertainty are critical components for optimal  system performance. More frequent assimilation helps mitigate the impact of such model errors and improve overall crop model predictions by correcting errors more often.  Assimilating in situ observations, the accuracy of soil moisture forecasts in the assimilation layers was improved  by an average of 17% for 10 cm and ~28% for 20 cm depth soil layer across all site-years and the crop yield was improved by an average of 23%.