Review of ‘Modeling the topographic influence on aboveground biomass using a coupled model of hillslope hydrology and ecosystem dynamics’
This study presents a unique effort to link an earth systems model, a hillslope hydrology model and a structured forest and vegetation function simulation model at a well known tropical forest study site in Panama.
Primary findings in my view are that while the model has not been tuned for the site and has unrealistic mortality and biomass responsponese (quantitatively), qualitatively it shows an interesting and promising ability to produce biomass, soil moisture, WTD and biomass turnover gradients over hillslopes. This is the result of coupled hydrological ecosystem and forest regeneration dynamics played out over periods of water stress expected in this seasonal tropical forest, unique outcomes for a unique modeling effort. This in and of itself represents key progress for the field. The value of this work at BCI will only increase from here if further steps are taken to improve model structure, tune the model and compare it more rigorously against the rich datasets available, particularly for biometrics/forest dynamics inside and (if possible) outside of the 50 ha plot. A rigorous validation analysis through time could even be used to test mortality models against one another in a model testing framework (rather than to show, as this study does as a first step, that different water stress mortality formulations have different outcomes).
The study may however be limited by a few important pieces, even when recognizing its place as a stepping stone. One is that the primary variation of interest appears to occur between the slope outside of the plot and the state inside the plot, which is relatively homogeneous (an issue familiar to 50 ha plot network research, an advantage of the data for cross site work and perhaps disadvantage for within site work). Nevertheless, this study provides model evidence that including hillslope hydrology creates WTD and associated seasonal soil moisture gradients that can make slopes and topographically lower areas more resilient to water stress, and enhance biomass in these areas. That is interesting and addresses current hypotheses and debate (see suggested consideration of Costa et al 2022 New Phytologist and Sousa et al 2022 GEB references below). Nevertheless, it seemed that this finding--of biomass differences over hillslopes--could have been a greater point of focus for the discussion in the manuscript particularly given that the Mascaro et al 2011 citation appears, which supports this with remote sensing analysis (this citation concludes that there is lidar based evidence with a map of biomass increasing on the slopes). I was also perplexed by the following: Is the spatial correlation between modeled and observed biomass non significant inside the 50 ha plot for all of the simulations? That might be expected due to low driver variation. However, I would clarify this as a result, and will return to the issue below.
A concern that I saw in relation to the particular hillslope modeled results was the issue of the impermeable boundary conditions/no lateral flow at the boundary. If I understood this, there is no outflow on the edges of the modeled scene, which all fall just below the plateau … ? Is there any chance that the modeled water is unrealistically ‘pooling’ in the soil, increasing WTD artificially here..? If it is easy to refute this, great! I would add a small statement to such an effect. I know that computation is a limitation (as is geography of the island), but I wondered if a larger scene would serve the analysis better in the future for this reason.
I see the value of the regression tree analysis to summarize and capture the predictability of drivers of the simulation outcomes. However, I felt that the specific details and discussion could have been de-emphasized relative to other issues, such as those related to improving the model realism to improve predictions across demographic, flux, and soil moisture variables. Specific RT results are not likely to remain of interest if the model is improved, suggesting that the focus should be on what RT analysis can reveal and how eventually it can complement empirically driven analysis.
The discussion of the dependency on water stress mortality kernels ended up somewhat superficial relative to my expectation from the presentation of the different models (which included some specification and description of the mechanisms). Why was it the M1 mortality function that led to such relatively weak dependency of AGB on water table depth while the others showed such drastic (and likely unrealistic) dependencies? That these models have different impacts overall is not surprising.
What is the role of vertical soil water stratification and how was this implemented in the model? In a few cases it seemed averages over vertical soil water profiles were taken for analysis (but not totally clear). I realize that this is not a focus and thus does not justify great expansion, however, the paucity of discussion seemed incomplete with negative consequences for understanding. For example, FATES hydraulic redistribution capacity seemed potentially important, which made sense to me when mentioned I think only because I know a little bit about FATES structure. I would suggest a clear sentence or two about vertical soil gradients in the methods. Related to this, it was unclear how the model produced water table depth predictions and how this was related to soil moisture. Finally, I will add that while it seems not a direction of this study at the moment, a lot of attention is being given to vertical stratification of water uptake, which may differ by tree size and function strategy (e.g. Chitra‐Tarak et al 2021. New Phytologist)
Soil structure is a huge issue that was not clearly addressed in the ‘factors limiting inference of field data’ discussion. Soil structure is an essential complexity related to hillslopes in tropical forests. In much of the tropics uplands are more clayrich than lowlands, which are sandier or siltier in general. I do not know about BCI. This has a big effect on flow dynamics obviously. See discussion of this in the Costa et al 2022 paper.
In general, I appreciated that the discussion was concise, but I would have foregone some discussion of the Random Forest results and focused on what is needed next to make the model produce more realistic results! The RF model does make clear and interesting points, but would have been stronger if it was comparable to, or could be extended to, the empirical data. Perhaps utilizing the broader lidar derived estimates of Mascaro et al 2011 (or another updated effort) would offer some hope for this approach in the future. As it stands, the detailed discussion of RF would be more satisfying if the model were performing more realistically. I am not sure what to take away from the relative importance comparisons without such grounding.
The plant functional type comparison is a nice additional component. It is understandable to start with something simple like a few present functional groupings (PFTs). That said, it would perhaps be worth noting that these should, in the future, also include specifications for differences in water stress responses. There are citations suggesting that tropical pioneers are more hydraulically vulnerable (cited/discussed in Costa et al 2022). Generally, this is necessary to get the best quantitative mortality response, and incorporate it into the model to produce ultimately realistic mortality rates. On that point, I think that M2 and M3 just are sending mortality too high too fast with soil water stress, and that is driving your rapidly asymptoting to too low level AGB with increasing WTD responses. Trait/PFT composition will impact these rates and should be a key part of advancing these models too.
A few more specific points…: There were some shortcuts in model description that would only be clear to insiders; e.g., talking about target biomass, carbon storage, etc. That would be out of the blue and not make much sense to someone that had no knowledge of flavors of ED I think… I would try to include some descriptive topic sentences about how that is important for determining plant relative performance and is impacted by resources and competition, etc. when first mentioned
Why would different mortality functions not impact AGB variation as is suggested?
I have not said much about the time series of ecosystem function components. That is a nice analysis, and a benefit for the paper. A note though, I would be a little careful in explaining what this is and how it operates clearly in your model experiment description. It took me a little while to understand the need for an extra 16 years of run time (i.e. when the met. drivers were available, right?). Maybe I missed something earlier but this section required a little head scratching before I realized that this pertained to comparison of time series ecosystem data.
In sum, I have raised some concerns and suggested action points for improvement. I want to step back though and say that this manuscript encompasses a nice progress report on new modeling developments and applications in the field. I just think that shifting the focus a little could do a better job to highlight what this paper is showing effectively that is novel and good, and should be further developed, i.e., that we can model/capture hillslope function and biomass responses. This will, in my view, entail stepping away from some detail about the specific predictions of RF that will end up less relevant when a better version of the model is built and running. I really hope that such future work is a serious plan; this could bring great leaps towards model-based testing for model structure comparisons. Furthermore, in discussion, I think there could be fewer generalities about the merits of such modeling, and more about how this will be improved by addressing its limitations for BCI or other tropical sites (while addressing workers rights).
Some more points on potentially overlooked citations:
Sousa, T.R., Schietti, J., Ribeiro, I.O., Emílio, T., Fernández, R.H., ter Steege, H., Castilho, C.V., Esquivel‐Muelbert, A., Baker, T., Pontes‐Lopes, A. and Silva, C.V., 2022. Water table depth modulates productivity and biomass across Amazonian forests. Global Ecology and Biogeography.
Costa, F.R., Schietti, J., Stark, S.C. and Smith, M.N., 2022. The other side of tropical forest drought: do shallow water table regions of Amazonia act as large‐scale hydrological refugia from drought?. New Phytologist.
Mascaro et al 2011 is cited but the finding that AGB increases on BCI hillslopes is surprisingly not mentioned as far as I noted. Instead it is stated that biomass spatial structure is not known for evaluation in the results section. I am a bit confused by this. Within the 50ha plot it certainly is known, and this lidar analysis suggests that there is a ready source for exploration in the literature for quantitative BCI hillslope biomass effects.
Chitra‐Tarak, R., Xu, C., Aguilar, S., Anderson‐Teixeira, K.J., Chambers, J., Detto, M., Faybishenko, B., Fisher, R.A., Knox, R.G., Koven, C.D. and Kueppers, L.M., 2021. Hydraulically‐vulnerable trees survive on deep‐water access during droughts in a tropical forest. New Phytologist, 231(5), pp.1798-1813. |
