Effects of 21st‐century climate, land use, and disturbances on ecosystem carbon balance in California

Abstract

Terrestrial ecosystems are an important sink for atmospheric carbon dioxide (CO2), sequestering ~30% of annual anthropogenic emissions and slowing the rise of atmospheric CO2. However, the future direction and magnitude of the land sink is highly uncertain. We examined how historical and projected changes in climate, land use, and ecosystem disturbances affect the carbon balance of terrestrial ecosystems in California over the period 2001–2100. We modeled 32 unique scenarios, spanning 4 land use and 2 radiative forcing scenarios as simulated by four global climate models. Between 2001 and 2015, carbon storage in California's terrestrial ecosystems declined by −188.4 Tg C, with a mean annual flux ranging from a source of −89.8 Tg C/year to a sink of 60.1 Tg C/year. The large variability in the magnitude of the state's carbon source/sink was primarily attributable to interannual variability in weather and climate, which affected the rate of carbon uptake in vegetation and the rate of ecosystem respiration. Under nearly all future scenarios, carbon storage in terrestrial ecosystems was projected to decline, with an average loss of −9.4% (−432.3 Tg C) by the year 2100 from current stocks. However, uncertainty in the magnitude of carbon loss was high, with individual scenario projections ranging from −916.2 to 121.2 Tg C and was largely driven by differences in future climate conditions projected by climate models. Moving from a high to a low radiative forcing scenario reduced net ecosystem carbon loss by 21% and when combined with reductions in land‐use change (i.e., moving from a high to a low land‐use scenario), net carbon losses were reduced by 55% on average. However, reconciling large uncertainties associated with the effect of increasing atmospheric CO2 is needed to better constrain models used to establish baseline conditions from which ecosystem‐based climate mitigation strategies can be evaluated.

Land-Use Change and Future Water Demand in California’s Central Coast

“…The LUCAS state-and-transition simulation model (STSM) [19,[54][55][56] was developed and modified for our study region. An STSM is a stochastic, Markov chain, empirical simulation model used to predict how defined variables transition between different specified states over a specified timeframe [58].…”

“…An STSM can also track age structure enabling age-dependent transitions as well as age-based triggering events. STSMs have been widely used to simulate landscape level vegetation change [59] and changes in land use and land cover (LULC) over time for assessing LULC scenario impacts on population and carbon dynamics [53][54][55], protected areas [57], and future water resources [19,56]. The STSM divides the landscape up into spatially discrete simulation cells, each with assigned state classes (i.e., LULC) and transition types.…”

“…We utilized the National Land Cover Dataset 2001 (NLCD01) [62], as our initial state class conditions, modified for our study region were as follows: (1) all four developed classes were collapsed into a single developed class and urban core areas defined per [63]; (2) the three forest classes were combined into a single forest class; (3) the woody and emergent wetlands classes were combined into a single wetlands class; (4) the agriculture and hay pasture classes were combined into a single annual agriculture class; (5) we used data from Sleeter et al [55], for the 2001 perennial agriculture class, described in more detail below; and (6) the "Developed-Roads" class from Landfire's Existing Vegetation Cover 2001 was used to designate a transportation class [64], ( Figure 1b). All datasets were resampled from 30-m to 270-m and re-projected into a NAD 1983 California Teale Albers map projection.…”

“…The NLCD01 does not contain a perennial orchard and vineyard class. We used a 2001 perennial cropland cover map [55], which generated orchard and vineyard cover using a gradient boosting machine algorithm framework. Any NLCD01 pixel classified as agriculture which overlapped the 2001 perennial cover estimate was classified as perennial cropland.…”

“…We simulated projected land use change and associated water demand for the years 2001-2100 at 270-m across 10 Monte Carlo simulations across these two scenarios. Our model was based on the Land Use and Carbon Scenario Simulator (LUCAS) [18,19,[53][54][55][56][57], a stochastic, spatially explicit state-and-transition simulation model. Spatial patterning of land use change was parameterized using local zoning datasets, identifying where land change would and would not occur giving current zoning ordinances and local mandates.…”