Contribution of land use to the interannual variability of the land carbon cycle Nature Communications

For GCB publication years 2013–2015, the LUH-GCB dataset was built off theLUH1 dataset and was identical to that dataset for the years 1500–2005. The datafor the years 2006 to the end of the LUH-GCB time series were based on newdata from FAO and HYDE, using the aforementioned anomaly approach (for yearsthat FAO data existed at the time of dataset creation), and extrapolationsfor years without FAO data. Table 1 provides the specific years that arebased on updated FAO and HYDE data in each of these LUH-GCB datasets. ForLUH data years without specified FAO data, extrapolations were used to filldata gaps.

Improved ORCHIDEE model with sub-grid land cohorts

The gridded fields of the BLUE simulations can be provided by the contact author upon request. Figure 3Regional FLUC between 1850 and 2015 from the two BK model estimates in GCB2019 (HN2017 in black and SBL for BLUE in dark blue), the BLUE simulations with net LUC transitions and standard parameterisation (light blue, SBL-Net) and using HN2017 parameterisations (cyan, SHNFull). The factorial simulations with only one set of parameters changed are shown in thin lines (SHNCdens in dark red, SHNt in red, SHNAlloc in yellow). The global differences between simulations result from interactions between the different factors and in the types of LUC occurring in a given point inspace and time. We first analyse the temporal evolution of regional FLUC for each simulation (Fig. 3). In order to attribute differences in FLUC between the two models to specific aspects from Table 1, we perform a set of factorial simulationswith BLUE (see Table 2), in which we replace the BLUE parameters with thosefrom HN2017 (see also schematic in Fig. 1).

3 Effects on gross FLUC component fluxes

• Differences HI-REG (b, e, h) and LO-REG (c, f, i) of BLUE-LUH2 primary land area for the same years as in panels (a, d, g).
• Transitions on primary non-forested land are mainly neglected in Europe, especially Poland, the Middle East, India and western Africa (not shown).
• In comparison to Hurtt et al. (2011), it can be noted that sensitivities might look different in other metrics like forest age or area.
• LUH2 differs from other datasets of historical land use such asHYDE3.2 (Klein Goldewijk at al., 2017), which LUH2 is based upon, andHILDA+ (Winkler et al., 2021), which incorporates additional satellite datafor the period 1960–2019.

For GCB publication years 2019–2020, a more significant update was performedto correct an error in the previously used input datasets, especially forthe country of Brazil, as described in Sect. These LUH-GCB datasetswere built off the LUH2 dataset and were identical to that dataset for allyears up to and including the year 1950. The data for the years 1951 to theend of the LUH-GCB time series were based on new data from FAO (FAO, 2020b) andcorrections from HYDE, along with extrapolations for years withoutunderlying FAO data, as outlined in Table 1. The national wood harvest reconstruction used by LUH was updated with newdata from FAO by replacing any extrapolated wood harvest data from thecurrent LUH dataset with the new wood harvest data from FAO and thenextrapolating national wood harvest rates bookkeeping model for any remaining years withoutunderlying FAO data. AB prepared the original draft; all other co-authors participated in the review and editing of the paper. The global and regional fluxes from HN2017 and the BLUE simulations are provided in the Supplement.

George C. Hurtt

• Since the other LULCC activities influence the available biomass, more or less area might be required in order to fulfil the harvested biomass demand.
• The factorial simulations with only one set of parameters changed are shown in thin lines (SHNCdens in dark red, SHNt in red, SHNAlloc in yellow).
• The IRS requires businesses making an average of $25 million or more in sales for the preceding three years to use accrual accounting.
• In addition, ORCHIDEE and HN2017 both used as input data annually harvested fuel wood volumes from FAOSTAT during 1961–2015, and from compilations of historical information19.
• In general, the IAV of Efire was underestimated in both results, because the input deforestation areas were derived from FAO statistics at a five-year interval and therefore smoothed in time19.

For the starting years presented here (850, 1700 or 1850), the spread in cumulative net LULCC flux is about the same order as that from including gross transitions but can be neglected for annual fluxes in recent years. This means that it is of little importance for estimates of the net LULCC flux over recent years when a simulation was started, but it is important for cumulative fluxes, with relevant implications for comparisons of the GCB and CMIP6 model simulations. However, not accounting for gross transitions and wood harvest, as is sometimes still the case in DGVMs, can cause even larger differences between model estimates.

2.2 Estimates of future emissions

• However, LULCC uncertainty matters less (by about a factor of 3) than the other two factors for the net LULCC flux in 2014, and historical LULCC uncertainty is negligible for estimates of future scenarios.
• At the same time, the largest estimates of the cumulative net LULCC flux comparing experiments with different StYr are produced in simulations from 1850 (second column).
• ELUC is computed by theDGVMs as the difference between two simulations, one with land use and onewithout, and as a result it includes the loss of additional sink capacityfrom reduced forest cover that is not included in the estimates ofELUC from bookkeeping models.
• Companies may use a hybrid of accrual accounting and cash accounting under IRS rules if specified requirements are met.
• If several reference experiments are given, the ordering is the same as in the column header.
• For industrial wood, in ORCHIDEE carbon release was assumed evenly distributed over a product residence time (10 years and 100 years) and in HN2017 an exponential decay was assumed with the same residence times (10 years and 100 years).

In this study, we used an improved version of the ORCHIDEE DGVM that is able to account for sub-grid cohorts for a given PFT that have different times since their establishment, so that the model has the strength to combine both bookkeeping functionality and the numerical representation of plant biophysics40. The ORCHIDEE version used here has been extensively validated for northern regions41 and applied globally in the recent annual GCP carbon budget update13. In this improved version, the carbon balances of intact and managed land (e.g., intact forest and recovering secondary forest) can be completely separated. This capability allows the quantification of ELUC and its individual components following Eq. (4), but with the advantage of accounting for the full impacts of environmental changes on ELUC, and especially the impacts of climate variations.

For both ORCHIDEE and HN2017 estimates, the simulations were started from the year 1700, but the ELUC was examined for the period of 1850–2015. The ORCHIDEE baseline simulation was driven by variable atmospheric CO2 and CRUNCEP climate data at a 2-degree resolution (prior to 1901 climate data for 1901–1920 were recycled). Historical forest area changes and wood harvest biomass were driven by exactly the same data used in HN2017 for different geographical regions of the world (see Supplementary Figs. 1 and 2; more details are provided in Supplementary Note 1). When Bookkeeping for Veterinarians agricultural area changes could not be matched by changes in forest imposed by HN2017, they were implemented as transitions with natural grassland.

Land-use-induced gross emissions (Efire, Ewood, and Elegacy) included both ledger account direct management effects and environmental effects, and exhibited greater IAV. As more IAV is attributed to managed land, intact ecosystems contribute less to the IAV of Snet. 2, the ORCHIDEE model can rigorously separate carbon fluxes of managed and intact ecosystems (Fig. 2a–c, Supplementary Figs. 3 and 4). For the period of 1990–2015, ELUC shows a net source of carbon in the tropics, driven by forest loss mainly due to the agricultural expansion, but less by industrial wood harvest20 (Fig. 2a). In contrast, over China, Europe, and part of the US, ELUC is a net carbon sink as a result of forest management, afforestation, and agricultural abandonment21,22.