Improved process representation of leaf phenology significantly shifts climate sensitivity of ecosystem carbon balance
Abstract. Terrestrial carbon cycle models are routinely used to determine the response of the land carbon sink under expected future climate change, yet these predictions remain highly uncertain. Increasing the realism of processes in these models may help with predictive skill, but any such addition should be confronted with observations and evaluated in the context of the aggregate behavior of the carbon cycle. Here, two formulations for leaf area index (LAI) phenology are coupled to the same terrestrial biosphere model, one is climate agnostic and the other incorporates direct environmental controls on both timing and growth. Each model is calibrated simultaneously to observations of LAI, net ecosystem exchange (NEE), and biomass using the CARbon DAta-MOdel fraMework (CARDAMOM), and validated against withheld data including eddy covariance estimates of gross primary productivity (GPP) and ecosystem respiration (Re), across six ecosystems from the tropics to high-latitudes. Both model formulations show similar predictive skill for LAI and NEE. However, with the addition of direct environmental controls on LAI, the integrated model explains 22 % more variability in GPP and Re, and reduces biases in these fluxes by 58 % and 77 %, respectively, while also predicting more realistic annual litterfall rates, due to changes in carbon allocation and turnover. We extend this analysis to evaluate the inferred climate sensitivity of LAI and NEE with the new model, and show that the added complexity shifts the sign, magnitude, and seasonality of NEE sensitivity to precipitation and temperature. This highlights the benefit of process complexity when inferring underlying processes from Earth observations and in representing the climate response of the terrestrial carbon cycle.
Alexander J. Norton et al.
Alexander J. Norton et al.
Alexander J. Norton et al.
Viewed (geographical distribution)
This manuscript seeks to evaluate the influence of different representations of leaf phenology on modeled terrestrial carbon cycle estimates. The manuscript compares two LAI phenology formulations---one with no climate controls (CDEA, the default in DALEC), and one where timing and growth are influenced by climate (Knorr et al. 2010, with some DALEC-specific modifications). This manuscript uses the CARDAMOM terrestrial ecosystem modeling and data assimilation framework, calibrated jointly against LAI (Copernicus EO 1km product) and NEE (FLUXNET2015) measurements and validated against tower-based GPP and RE (FLUXNET, based on night-time partitioning) and in-situ biomass measurements (with site-specific allometric scaling). The analysis is performed at 6 FLUXNET sites spanning a variety of biomes. Results show that the climate-driven phenology scheme improved predictions of GPP, RE, and litterfall. The climate-driven phenology scheme also led to different NEE sensitivity to precipitation and temperature.
Overall, I found this to be a solid, well-executed study. The science topic --- representations of LAI phenology in vegetation models --- is important and relevant. The modeling approach, and the methods for calibration, validation, and sensitivity analysis, are well-explained and sound. The results are compelling and well-interpreted and contextualized in the literature. I have a few minor comments related to presentation (see detailed comments below), but I think the overall quality of this study is good.
[Line 7, "biomass"]
Based on the methods, I think the model is *validated* against biomass but only calibrated against LAI and NEE (i.e., only LAI and NEE appear in the likelihood).
Somewhere in here, you might also consider citing Wheeler & Dietze 2021 (DOI: 10.5194/bg-18-1971-2021).
[Line 56, "Bayesian data assimilation"]
Although technically not inaccurate, I find the terms "data assimilation" and "Model data fusion" to be somewhat vague and potentially misleading in this context. Here and elsewhere, I suggest more precise terminology such as (Bayesian) "calibration", "optimization", or "parameter data assimilation", to distinguish what is done here (tuning of model *parameters* that affect the entire course of the simulation) from *state* data assimilation (a stepwise process in which model *states* at a particular time and place are tuned to better match observations, e.g., via Kalman filter, as is done in reanalysis products). (Admittedly, Macbean et al. 2016 and many others also use "data assimilation" this way, so this is not a problem unique to this study.)
[Equations 3-5, 10, others]
You might consider using explicit multiplication symbols (x or dot), spacing, fonts (e.g., non-italic font for symbols like LAI), different brackets (e.g., hard brackets for indexes), or different kinds of symbols (e.g., Greek vs. Latin, capital vs. lowecase) to more clearly distinguish between multiplication, function calls, indexing, and multi-letter acronyms (e.g., in equation 2, Phi refers to the Normal CDF called on the fraction in parentheses, whereas in equation 3, the lowercase chi is presumably multiplied by the LAI difference; WLAI isn't immediately obvious as W x LAI).
This probably needs the (t) index for the terms on the right?
C(lab) here probably needs a time index (t-1?)
[Line 476, "positive ST_LAI"]
This is slightly misleading, since the Knorr formulation predicts near-zero ST_LAI in the warmer sites (which is what one would hope!). I suggest tweaking this sentence to highlight the differences across formulations and sites.
[Line 634, "available upon request"]
EGUsphere doesn't have a data use policy (or at least, I couldn't find one) so this is technically not a violation. However, I personally feel that "availability upon request" is unacceptable data sharing policy for modern scientific publications. Unless there is a clear and compelling reason (e.g., government mandate, conservation risk, etc.; if there is such a limitation, it should be explicitly stated), data should to be deposited in a publicly available repository such as Dryad, FigShare, or Open Science Framework. The importance and benefits of open data have been widely documented over the last decade; among the most recent examples is Noy and Noy 2019 (https://doi.org/10.1038/s41563-019-0539-5), and journals are increasingly requiring code and data sharing as a precondition for publication (e.g., AGU data policy -- https://www.agu.org/Publish-with-AGU/Publish/Author-Resources/Data-and-Software-for-Authors; GMD data policy -- https://www.geoscientific-model-development.net/policies/code_and_data_policy.html).