• Energy Research

AbstractLarge across‐model spread in simulating land carbon (C) dynamics has been ubiquitously demonstrated in model intercomparison projects (MIPs), and became a major impediment in advancing climate change prediction. Thus, it is imperative to identify underlying sources of the spread. Here, we used a novel matrix approach to analytically pin down the sources of across‐model spread in transient peatland C dynamics in response to a factorial combination of two atmospheric CO2 levels and five temperature levels. We developed a matrix‐based MIP by converting the C cycle module of eight land models (i.e., TEM, CENTURY4, DALEC2, TECO, FBDC, CASA, CLM4.5 and ORCHIDEE) into eight matrix models. While the model average of ecosystem C storage was comparable to the measurement, the simulation differed largely among models, mainly due to inter‐model difference in baseline C residence time. Models generally overestimated net ecosystem production (NEP), with a large spread that was mainly attributed to inter‐model difference in environmental scalar. Based on the sources of spreads identified, we sequentially standardized model parameters to shrink simulated ecosystem C storage and NEP to almost none. Models generally captured the observed negative response of NEP to warming, but differed largely in the magnitude of response, due to differences in baseline C residence time and temperature sensitivity of decomposition. While there was a lack of response of NEP to elevated CO2 (eCO2) concentrations in the measurements, simulated NEP responded positively to eCO2 concentrations in most models, due to the positive responses of simulated net primary production. Our study used one case study in Minnesota peatland to demonstrate that the sources of across‐model spreads in simulating transient C dynamics can be precisely traced to model structures and parameters, regardless of their complexity, given the protocol that all the matrix models were driven by the same gross primary production and environmental variables.

AbstractLarge across‐model spread in simulating land carbon (C) dynamics has been ubiquitously demonstrated in model intercomparison projects (MIPs), and became a major impediment in advancing climate change prediction. Thus, it is imperative to identify underlying sources of the spread. Here, we used a novel matrix approach to analytically pin down the sources of across‐model spread in transient peatland C dynamics in response to a factorial combination of two atmospheric CO2 levels and five temperature levels. We developed a matrix‐based MIP by converting the C cycle module of eight land models (i.e., TEM, CENTURY4, DALEC2, TECO, FBDC, CASA, CLM4.5 and ORCHIDEE) into eight matrix models. While the model average of ecosystem C storage was comparable to the measurement, the simulation differed largely among models, mainly due to inter‐model difference in baseline C residence time. Models generally overestimated net ecosystem production (NEP), with a large spread that was mainly attributed to inter‐model difference in environmental scalar. Based on the sources of spreads identified, we sequentially standardized model parameters to shrink simulated ecosystem C storage and NEP to almost none. Models generally captured the observed negative response of NEP to warming, but differed largely in the magnitude of response, due to differences in baseline C residence time and temperature sensitivity of decomposition. While there was a lack of response of NEP to elevated CO2 (eCO2) concentrations in the measurements, simulated NEP responded positively to eCO2 concentrations in most models, due to the positive responses of simulated net primary production. Our study used one case study in Minnesota peatland to demonstrate that the sources of across‐model spreads in simulating transient C dynamics can be precisely traced to model structures and parameters, regardless of their complexity, given the protocol that all the matrix models were driven by the same gross primary production and environmental variables.

AbstractAccurate assessment of forest carbon dynamics is a critical element of appraising forest‐based Natural Climate Solutions. National forest inventory and analysis (FIA) data provide valuable pool‐based estimates of carbon stocks, but have been underutilized to inform carbon cycle modeling for forest carbon dynamics with stand development. This study introduces a hierarchical data assimilation (HDA) framework to optimize modeling parameters by incrementally assimilating measured carbon pool data into the model. We found that most carbon stocks could be reproduced by constrained parameters after each HDA step. Using aboveground live biomass (AGB) alone in HDA was able to reproduce the AGB trajectories but introduced biases in estimating the downstream dead biomass and soil carbon pools. Assimilating dead biomass measurements narrowed the posterior space of parameter solutions and improved consistency between measured and modeled carbon dynamics. The HDA framework also reduced uncertainties on modeled carbon fluxes. Young stands were found to release less carbon when the model was informed by dead biomass compared to simulations guided by aboveground biomass alone. The remaining mismatches between modeled and FIA pool estimates could be attributed to wide uncertainty in some FIA estimates, differing definitions of functional carbon pools, and structural rigidity in the model. Together, this study underscores the importance of pool‐based measurements in forest carbon modeling, which improves the model‐observation fit and reduces process‐model uncertainty.

AbstractAccurate assessment of forest carbon dynamics is a critical element of appraising forest‐based Natural Climate Solutions. National forest inventory and analysis (FIA) data provide valuable pool‐based estimates of carbon stocks, but have been underutilized to inform carbon cycle modeling for forest carbon dynamics with stand development. This study introduces a hierarchical data assimilation (HDA) framework to optimize modeling parameters by incrementally assimilating measured carbon pool data into the model. We found that most carbon stocks could be reproduced by constrained parameters after each HDA step. Using aboveground live biomass (AGB) alone in HDA was able to reproduce the AGB trajectories but introduced biases in estimating the downstream dead biomass and soil carbon pools. Assimilating dead biomass measurements narrowed the posterior space of parameter solutions and improved consistency between measured and modeled carbon dynamics. The HDA framework also reduced uncertainties on modeled carbon fluxes. Young stands were found to release less carbon when the model was informed by dead biomass compared to simulations guided by aboveground biomass alone. The remaining mismatches between modeled and FIA pool estimates could be attributed to wide uncertainty in some FIA estimates, differing definitions of functional carbon pools, and structural rigidity in the model. Together, this study underscores the importance of pool‐based measurements in forest carbon modeling, which improves the model‐observation fit and reduces process‐model uncertainty.

AbstractUnderstanding vegetation recovery after drought is critical for projecting vegetation dynamics in future climates. From 1997 to 2009, Australia experienced a long‐lasting drought known as the Millennium Drought (MD), which led to widespread reductions in vegetation productivity. However, vegetation recovery post‐drought and its determinants remain unclear. This study leverages remote sensing products from different sources—fraction of absorbed photosynthetically active radiation (FPAR), based on optical data, and canopy density, derived from microwave data—and random forest algorithms to assess drought recovery over Australian natural vegetation during a 20‐year period centered on the MD. Post‐drought recovery was prevalent across the continent, with 6 out of 10 drought events seeing full recovery within about 6 months. Canopy density was slower to recover than leaf area seen in FPAR. The probability of full recovery was most strongly controlled by drought return interval, post‐drought hydrological condition, and drought length. Full recovery was seldom observed when drought events occurred at intervals of 3 months or less, and moderately dry (standardized water balance anomaly [SWBA] within [−1, −0.76]) post‐drought conditions resulted in less complete recovery than wet (SWBA > 0.3) post‐drought conditions. Press droughts, which are long term but not extreme, delayed recovery more than pulse droughts (short term but extreme) and led to a higher frequency of persistent decline. Following press droughts, the frequency of persistent decline differed little among biome types but peaked in semi‐arid regions across aridity levels. Forests and savanna required the longest recovery times for press drought, while grasslands were the slowest to recover for pulse drought. This study provides quantitative thresholds that could be used to improve the modeling of ecosystem dynamics post‐drought.

AbstractUnderstanding vegetation recovery after drought is critical for projecting vegetation dynamics in future climates. From 1997 to 2009, Australia experienced a long‐lasting drought known as the Millennium Drought (MD), which led to widespread reductions in vegetation productivity. However, vegetation recovery post‐drought and its determinants remain unclear. This study leverages remote sensing products from different sources—fraction of absorbed photosynthetically active radiation (FPAR), based on optical data, and canopy density, derived from microwave data—and random forest algorithms to assess drought recovery over Australian natural vegetation during a 20‐year period centered on the MD. Post‐drought recovery was prevalent across the continent, with 6 out of 10 drought events seeing full recovery within about 6 months. Canopy density was slower to recover than leaf area seen in FPAR. The probability of full recovery was most strongly controlled by drought return interval, post‐drought hydrological condition, and drought length. Full recovery was seldom observed when drought events occurred at intervals of 3 months or less, and moderately dry (standardized water balance anomaly [SWBA] within [−1, −0.76]) post‐drought conditions resulted in less complete recovery than wet (SWBA > 0.3) post‐drought conditions. Press droughts, which are long term but not extreme, delayed recovery more than pulse droughts (short term but extreme) and led to a higher frequency of persistent decline. Following press droughts, the frequency of persistent decline differed little among biome types but peaked in semi‐arid regions across aridity levels. Forests and savanna required the longest recovery times for press drought, while grasslands were the slowest to recover for pulse drought. This study provides quantitative thresholds that could be used to improve the modeling of ecosystem dynamics post‐drought.