A Forest Model Intercomparison Framework and Application at Two Temperate Forests Along the East Coast of the United States
Round 1
Reviewer 1 Report
General Comments:
In the manuscript by Erickson and Strigul, two forest biogeochemistry models (LANDIS-II NECN and PPA-SiBGC) were applied to two long-term research sites in the eastern U.S. Model performance is evaluated through comparison with data on carbon and nitrogen stocks and fluxes. The models performed well for C and N stocks but poorly for fluxes. This type of model-data comparison is needed and model intercomparison is a useful tool for improving models. The modeling work done in this manuscript is of high quality. However, I feel that an opportunity was missed here because of the lack of a thorough evaluation and comparison of model performance for fluxes. PPA-SiBGC correctly predicted the magnitude of NEE but not the temporal pattern while the opposite was true of LANDIS-II. Relating this difference to structural differences between the models (even if it is hypothesized) is the key piece of analysis that would make this a strong publication. The manuscript currently includes a qualitative discussion of model characteristics such as ease of parameterization and training new users, which could be shortened or removed. In my opinion, it is much less useful to model development than a careful analysis of differences in model performance. This type of material may be more appropriate as a commentary written for the forest management and modeling community.
The introduction is primarily a detailed history of forest biogeochemistry models. While this is interesting, it does little to set up the need for and motivation behind the current study. This section should be shortened substantially. A more detailed discussion is needed in the introduction of the previous forest model intercomparison projects (ISIMIP, CoFoLaMo). It is unclear why there is the need to establish a foundation for forest model intercomparisons when these projects exist. The lack of existing model intercomparisons in North America is given as justification, but it is not clear why the results from the European projects are not transferable.
The clarity of the two sections describing the models used (2.1.1 and 2.1.2) could be improved by using the same organizational structure for the two sections, with descriptions of the different model characteristics occurring in the same order using similar language. Certain things are emphasized for one model and barely mentioned for the other (e.g. model code structure for LANDIS-II and parameter estimation for PPA-SiBGC) and it is unclear why this is done. A table could also be a good way to highlight key model similarities and differences.
The poor performance of both models in representing fluxes deserves more attention. Both models are often used for multi-decadal modeling scenarios in which error in the fluxes could compound over time to errors in the stocks, creating problems that are not evident in the relatively short model runs (11 and 5 years) used in this study. The results presented suggest more substantial errors in the flux results that are not presented. Figure 4 shows that the modeled NEE with LANDIS-II predicted the HF site was a C source, but the modeled aboveground C stocks were increasing at a rate similar to observed values. This would suggest substantial overestimation of soil respiration in the model. It should also be considered that model performance was better for metrics that were measured through vegetation surveys than metrics based on eddy covariance data. It might be worth investigating if there are any issues with how the tower footprint matched up with the spatial domain of the models.
Specific Comments:
139: What was the reason for the sharp decline in NEE in 1998?
163: “soil respiration rates here again appear to be complex,” only ecosystem respiration is discussed for the other site, which is not the same thing as soil respiration.
175: A site map should be included in the main manuscript, not an appendix.
198: I’m confused by the use of “Perhaps unknowingly” here, implying that the model doesn’t know what it is based on.
Fig. 1 & 2: Make sure all the acronyms used in the figures are defined in the text or caption.
Table 2: Recommend giving full species names rather than abbreviations as this would require minimal additional space and improve clarity.
290 & L. 317: For the data described in these paragraphs, explain where data collection methods can be accessed or provide in an Appendix.
L. 366: What is the time step of the time series used for model intercomparison?
Fig. 3: Recommend giving the sites as sub-figure titles and the color key in a legend rather than in the figure caption.
Fig. 4: Why is the time series figure only given for the Harvard Forest site?
Fig. 4b: If the models were initialized with field data, why is there such a large initial difference between modeled and observed aboveground C pools?
Author Response
Dear Reviewer,
We graciously thank you for your insightful comments, which have strengthened the manuscript. We have related the differences in model performance to the model architectures in new paragraphs in the discussion. We agree that this provides valuable new insights. We have left the section on the practical details of model implementation, as this is a deliverable required for our project. In addition, we have removed whole paragraphs on previous work at the model intercomparison sites from the introduction, which are now found within the site descriptions. In addition, we have reorganized the introduction with new subsections that clarify the background for our work and have added new paragraphs detailing previous MIP efforts in the US and Europe. We also added substantial text regarding the performance of both models in simulating fluxes. Finally, we have gone over all simulation code with a fine-tooth comb, recomputed all simulations to ensure reliable results, and produced new plots with more information in less space for the model outputs, which will be made accessible through a simple object-oriented interface in our Erde library.
Specific Comments:
139: What was the reason for the sharp decline in NEE in 1998?
The 1998 anomaly was attributed to weak GEE due to a disturbance even lasting until 2001. It is hypothesized that this may be due to snow or ice storms (Urbanski et al., 2007). Spring of 1998 was unusually warm, resulting in earlier leaf emergence (producing a record tie at nearby Hubbard Brook), which can yield frost damage given a subsequent dip in temperatures.
163: “soil respiration rates here again appear to be complex,” only ecosystem respiration is discussed for the other site, which is not the same thing as soil respiration.
We note in the new discussion section that soil respiration is not ecosystem respiration and we discuss related problems in connecting the two.
175: A site map should be included in the main manuscript, not an appendix.
We have included the site maps in the main document.
198: I’m confused by the use of “Perhaps unknowingly” here, implying that the model doesn’t know what it is based on.
To remove any confusion, we have removed this phrase.
Fig. 1 & 2: Make sure all the acronyms used in the figures are defined in the text or caption.
All acronyms are clearly defined throughout the manuscript, including a full list in the Appendix.
Table 2: Recommend giving full species names rather than abbreviations as this would require minimal additional space and improve clarity.
For consistency with our code and parameter files and to therefore ease interpretation of these files, we would prefer to use the PLANTS database codes.
290 & L. 317: For the data described in these paragraphs, explain where data collection methods can be accessed or provide in an Appendix.
Data collection methods can be accessed through the websites provided. Both sites provide metadata for each file, which we note.
L. 366: What is the time step of the time series used for model intercomparison?
This is an annual time-step or integral, which we mention.
Fig. 3: Recommend giving the sites as sub-figure titles and the color key in a legend rather than in the figure caption.
We have modified the figures and improved them substantially.
Fig. 4: Why is the time series figure only given for the Harvard Forest site?
We have provided time-series figures for both sites.
Fig. 4b: If the models were initialized with field data, why is there such a large initial difference between modeled and observed aboveground C pools?
This is due to differences in the calculation methods for each model in comparison to the site data.
Reviewer 2 Report
As a microbial ecologist with an interest in measuring microbe-mediated biogeochemical process rates across ecosystems and ultimately finding a way to model these processes with time, given measured abiotic and biotic variables, I was interested in the comparisons provided by in this submission. Although the scope was larger, I found it relevant and think it deserves merit. Overall, in as much as the submission indicates that both models work well for their intended purposes, it also highlights their limitations, as well as the need for more difficult to obtain empirical data in order to improve model efficiency and transferability.
Author Response
Dear Reviewer,
We graciously thank you for your insightful comments. We are delighted to hear that our manuscript caught your interest as a microbial ecologist. We hope that our manuscript will similarly capture the interest of wide variety of readers from other disciplines.
Round 2
Reviewer 1 Report
The manuscript has been improved considerably from the previous version and is now suitable for publication.
