Composted manure and green waste amendments have been shown to increase net carbon (C) sequestration in rangeland soils and have been proposed as a means to help lower atmospheric CO2 concentrations. However, the effect of climate change on soil organic C (SOC) stocks and greenhouse gas emissions in rangelands is not well understood, and the viability of climate change mitigation strategies under future conditions is even less certain. We used a process-based biogeochemical model (DayCent) at a daily timestep to explore the long-term effects of potential future climate changes on C and greenhouse gas dynamics in annual grassland ecosystems. We then used the model to explore how the same ecosystems might respond to climate change following compost amendments to soils and determined the long-term viability of net SOC sequestration under changing climates. We simulated net primary productivity (NPP), SOC, and greenhouse gas fluxes across seven California annual grasslands with and without compost amendments. We drove the DayCent simulations with field data and with site-specific daily climate data from two Earth system models (CanESM2 and HadGEM-ES) and two representative concentration pathways (RCP4.5 and RCP8.5) through 2100. Net primary productivity and SOC stocks in unamended and amended ecosystems were surprisingly insensitive to projected climate changes. A one-time amendment of compost to rangeland acted as a slow-release organic fertilizer and increased NPP by up to 390–814 kg C ha^-1 y^-1 across sites. The amendment effect on NPP was not sensitive to Earth system model or emissions scenario and endured through the end of the century. Net SOC sequestration amounted to 1.96 ± 0.02 Mg C ha^-1 relative to unamended soils at the maximum amendment effect. Averaged across sites and scenarios, SOC sequestration peaked 22 ± 1 years after amendment and declined but remained positive throughout the century. While compost stimulated nitrous oxide (N₂O) emissions, the cumulative net emissions (in CO₂ equivalents) due to compost were far less than the amount of SOC sequestered. Compost amendments resulted in a net climate benefit of 69.6 ± 0.5 Tg CO₂e 20 ± 1 years after amendment if applied to similar ecosystems across the state, amounting to 39% of California’s rangeland. These results suggest that the biogeochemical benefits of a single amendment of compost to rangelands in California is insensitive to future climate change and could contribute to decadal-scale climate mitigation goals alongside emissions reductions.

The DayCent Biogeochemical Model (Parton et al. 1998) was used to simulate the effects of climate and management on C and greenhouse gas dynamics in each rangeland ecosystem. DayCent is a widely utilized and well-established complex process model, developed using ecological concepts of grassland soil C and N dynamics (Parton et al. 1994). The model is parameterized for initial conditions using site-specific historical climate data, annual net primary productivity (NPP), and depth-specific measured values for soil texture and bulk density. The calibrated model provides a baseline from which the model can calculate trends with time and differences under changing conditions (e.g., climate and compost amendments in this study). DayCent partitions existing and added C into discrete soil pools based on estimated C turnover time: active (< 1 year), slow (decadal), and passive (millennial). Dead plant material is partitioned into active or slow cycling pools initially, depending upon tissue chemistry (e.g., lignin:N ratio), using first-order kinetics. Carbon can move among pools through decomposition and stabilization. The movement among pools mimics microbial activity and the mineral association of organic matter; it includes a separate pool for microbial biomass, but DayCent does not explicitly model specific mechanisms of microbial interactions or mineral stabilization (Parton et al. 1994). Modeled SOC flows and NPP are both strongly dependent on soil water availability in DayCent, which has been shown to be an important driver of ecosystem C dynamics in grasslands (Burke et al. 1997, Harpole et al. 2007). The nitrogen gas sub-model of DayCent uses a daily timestep to simulate N₂O fluxes from nitrification and denitrification based on diffusivity parameters of soil (water-filled pore space, texture, bulk density, field capacity, temperature), pH, and soil NH₄⁺ and NO₃^- concentrations (Parton et al. 2001). The grassland CH₄ oxidation sub-model simulates methanotrophy at a daily timestep as a function of soil water content, field capacity, porosity, and temperature (Del Grosso et al. 2000). DayCent also models soil respiration and microbial respiration of CO₂; here we report on total soil respiration which is more comparable with field data. DayCent facilitates the simulation of explicit management practices including grazing and compost amendments, and was originally developed, and has been used extensively, for modeling managed grassland and cropland ecosystems (Kelly et al. 2000; Parton et al. 1993; Parton et al. 1998; Ryals et al. 2015).

Biogeochemical model inputs

Field observations of soil texture, total organic C, bulk density, and biomass production from pre-treatment plots were used for the initial parameterization of the model for each site (Table 1). Total organic C was measured on five replicate soil cores at four depths down to 1 meter or point of refusal. The point of refusal was below one meter except for a minority of cores in Mendocino, Marin, and Tulare, where the mean points of refusal were 95.7 ± 2.7 cm, 92.2 ± 3.4 cm, and 99.5 ± 0.5 cm, respectively. Soil texture was measured on 3 samples from each transect (first, third, and fifth core) from the 0-10 cm depth at each site. Soil texture data for 10-100 cm soil depths were obtained from the SSURGO database (Soil Survey Staff et al. 2017). Bulk density samples were taken using a 6.35 cm diameter metal corer at 10 cm depth increments to 1 m or point of refusal from two soil pits per site. Aboveground NPP was measured by clipping vegetation at peak biomass from eight replicate 200 cm² subplots for both amended and unamended plots, oven drying at 65 ºC, and weighing; belowground NPP was measured in Marin and Yuba only by Ryals et al (2013). Soil subsamples were analyzed in duplicate for total C concentration at U.C. Berkeley on a Carlo Erba Elantech elemental analyzer (Lakewood, NJ, USA) using atropine as a standard at a rate of one per ten samples. Samples were re-run in if duplicates varied by more than 10%. Soils were tested for carbonates using 2M HCl; as no carbonates were found, results reported reflect only organic C concentrations. Bulk density was determined by calculating the rock volume and determining the oven dry (105°C) mass of soil per unit volume. Soil organic C contents were calculated by multiplying the C concentrations (%) by the oven-dry mass of the fine fraction (< 2 mm) and dividing by the bulk density and depth (Throop et al. 2012). Additional details can be found in Silver et al. (2018).  Livestock effects on biomass and biogeochemical cycling were represented using scheduled time- and intensity-specific grazing events. Grazing management was simulated to reflect site-specific historic and current practices (Appendix S1).

Simulations of future conditions were driven by daily climate data from 2006 to 2100 extracted from the CanESM2 (Canadian Centre for Climate Modeling and Analysis, Canada) and HadGEM2-ES (Met Office Hadley Centre, UK) Earth System Models (ESMs). We chose not to simulate CO2 fertilization in order to isolate the role of climate. There remains debate as to which ESM most accurately represents future weather in California. We used CanESM2 and HadGEM2-ES because they yielded contrasting projections for future precipitation (see below). We used two Representative Concentration Pathway (RCP) scenarios: RCP4.5 that assumes some emissions reductions, and RCP8.5 that assumes business as usual societal behavior with minimal emissions reductions. We chose these two scenarios because California used RCP4.5 and RCP8.5 for emissions reduction targets in their 2018 assessment report (Franco et al. 2018). Data were extracted for the site-specific (2.8°x 2.8°) geographical grid of CanESM2 and HadGEM2-ES.

Output units are described in the ReadMe.txt file.