Feedbacks between the land surface and the atmosphere are expected to play a role in the response of the Earth system to climate change. Earth system models (ESMs) are designed to simulate these processes, but their implementation is only as good as scientific knowledge and technological capabilities allow. This puts ESMs in a dual role with respect to land-atmosphere feedbacks: On the one hand, they represent the most important tools researchers have for understanding land-atmosphere feedbacks and how they will affect the Earth system. On the other hand, they acquire substantial uncertainty as a result of limitations in their representation of feedback processes.

In this dissertation, I used ESMs within the context of both roles. I used them as tools for understanding how land-atmosphere feedbacks affect the Earth system and our understanding of it. I also used observational data to constrain land-atmosphere feedbacks in ESMs with the goal of evaluating and improving their ability to project future responses to climatic forcing.

The goal of my first science chapter was to evaluate how well ESMs represent seasonal-scale feedbacks between terrestrial water availability and atmospheric conditions. I developed a metric that was designed to use terrestrial water storage data from the Gravity Recovery and Climate Experiment (GRACE) satellites in conjunction with other remote sensing data sources for atmospheric temperature and humidity, precipitation, and downwelling shortwave radiation. First, I used GRACE to identify the months of the year when the land surface loses more water to evapotranspiration and runoff than it gains from precipitation. I then related the interannual water storage anomaly at the onset of this interval with subsequent atmospheric conditions as a measure of the land surface forcing on the atmosphere, as well as relating the atmospheric conditions with the subsequent water storage anomaly as a measure of the land surface response to atmospheric forcing. By calculating the equivalent quantities in an ensemble of ESMs, I demonstrated that the models tended to overestimate the strength of these relationships. These results are consistent with complementary efforts using other data sources and spatiotemporal scales, and suggest that overly simplistic representations of the heterogeneity within vegetation cover and bare soil could cause models to overestimate soil moisture feedbacks with the atmosphere.

My next chapter was partially motivated by a remaining question from the previous research: How much of the measured relationships was driven directly by interactions between soil moisture and the atmosphere, and how much is due to co-variability due to external forcing such as sea surface temperature (SST) anomalies. An additional motivation stemmed from the well-known relationship between the El Niño-Southern Oscillation (ENSO) and interannual variability in the growth rate of atmospheric CO2 concentrations. This relationship is attributed primarily to the response of tropical terrestrial ecosystems, and temperature has been widely implicated as the primary driver. However, there is an ongoing debate in the literature suggesting that hydrology is also an important driver, which, as I demonstrated in the previous chapter, is not independent of temperature.

To address these questions, I performed a set of experiments using the Energy Exascale Earth System Model (E3SM). I modified E3SM in order to decouple the interannual variability of soil moisture and SST, which allowed me to isolate the influence of each of these factors in the response of the Amazon rainforest to ENSO. I found that in E3SM, soil moisture served to amplify and extend the land surface response to ENSO in the Amazon. SST anomalies coupled with atmospheric circulation drove an immediate response, which coincides with the Amazon wet season. As the ecosystem was generally not water limited at this time, temperature played the dominant role in carbon cycle variability. However, soil moisture anomalies persisted into the dry season, intensifying and extending the response of both temperature and the terrestrial CO2 flux. This highlights the importance of considering the feedback between soil moisture and temperature when considering their relative importance as drivers of CO2 variability in the Amazon.

For my final science chapter, I used a global dataset of vertically resolved soil radiocarbon observations to evaluate the representation of soil carbon processes in the Community Land Model (CLM) and the E3SM land model (ELM). I found that while ELM slightly overestimated radiocarbon ages in temperate latitudes, it underestimated them in boreal and tropical latitudes, particularly at depth, enough to lead to a young bias globally. CLM, on the other hand, underestimated radiocarbon ages at all depths, latitudes, and vegetation types. This suggests that carbon was cycling through the soil too quickly in the models, leading them to overestimate the rate at which soils could sequester carbon in response to increasing atmospheric CO2 concentrations.

I used the observed radiocarbon profiles to constrain kinetic rate constants and transfer coefficients in ELM, which improved the young age bias and improved a low bias in the tropical soil carbon stock. Constraining the model increased the global soil carbon stock as a result of the improvements in the tropics, and reduced the contribution of soil to the terrestrial carbon sink. This suggests that soil is likely to respond very slowly to increasing atmospheric CO2 concentrations, and is unlikely to serve as a short-term carbon sink.

These three chapters illustrate the range applications for ESMs in the context of land-atmosphere feedbacks. ESMs can be evaluated for how well they reproduce observations of relationships between multiple climatic variables, as in the first chapter. ESMs can be used to understand land-atmosphere feedbacks in the Earth system, by isolating factors that are difficult or impossible to disentangle in nature, as in the second chapter. ESMs can be constrained by observations in order to reduce uncertainty in their simulation of land-atmosphere feedbacks, as in the final chapter. This dissertation represents steps toward understanding how energy, water, and carbon flow through terrestrial ecosystems, and improving the representation of these processes in ESMs.

Anthropogenic perturbation of global biogeochemical cycles, particularly through emissions of radiatively active greenhouse gases into the atmosphere—chiefly carbon dioxide (CO2), methane (CH4), and nitrous oxide (N2O)—is altering the Earth's climate and inducing feedbacks from the terrestrial biosphere and oceans on future CO2 levels and the climate system. Identifying and quantifying these feedbacks and quantifying and reducing uncertainties associated with them in process-rich Earth system models (ESMs) are important for advancing our understanding of the Earth system, predicting future atmospheric CO2 levels, informing carbon management and energy policies, and fostering the future of life on Earth. This dissertation presents three studies designed to advance our understanding of biogeochemical processes and their interactions with climate under conditions of increasing atmospheric CO2 and to offer an approach for understanding observational representativeness and for scaling up measurements.

In the first investigation, I analyzed emission-driven simulations of ESMs from the fifth phase of the Coupled Model Intercomparison Project (CMIP5) in which atmospheric CO2 levels were computed prognostically. Comparison of ESM prognostic atmospheric CO2 over the historical period with observations indicated that ESMs, on average, had a small positive bias in predictions of contemporary atmospheric CO2, due in part to weak ocean carbon uptake. I found a significant linear relationship between contemporary atmospheric CO2 biases and future CO2 levels for the multimodel ensemble, and used this emergent constraint to create a contemporary CO2 tuned model (CCTM) to estimate an atmospheric CO2 trajectory for the 21st century for the Representative Concentration Pathway (RCP) 8.5. The CCTM yielded CO2 estimates of 600 ± 14 ppm at 2060 and 947 ± 35 ppm at 2100, which were 21 ppm and 32 ppm below the multi-model mean during these two time periods, respectively. This analysis indicated that much of the model-to-model variation in projected CO2 during the 21st century was tied to biases that existed during the observational era and that model differences in the representation of concentration–carbon feedbacks and other slowly varying carbon cycle processes appear to be the primary driver of this variability.

In the second study, I extended a quantitative methodology for stratifying sampling domains and understanding the representativeness of measurements, measurement sites, and observational networks. Multivariate spatiotemporal clustering was applied to down-scaled general circulation model results and data for the State of Alaska at 4 km2 resolution to define multiple sets of ecoregions across two decadal time periods and to identify optimal sampling locations for those ecoregions. I developed a representativeness metric and used it to characterize environmental dissimilarity between potential sampling sites. This analysis provided insights into optimal sampling strategies and offered a framework for up-scaling measurements that can be applied at different spatial and temporal scales to meet the needs of individual measurement campaigns.

In the third investigation, I applied a feedback analysis framework to three sets of long-term climate change simulations from the Community Earth System Model version 1.0 (CESM1(BGC)) to quantify drivers of nonlinear terrestrial and ocean responses of carbon uptake. In the biogeochemically coupled simulation (BGC), the effects of CO2 fertilization and nitrogen deposition were expressed in the biosphere. In the radiatively coupled simulation (RAD), the effects of rising temperature and circulation changes due to radiative forcing from CO2, other greenhouse gases, and aerosols were expressed in the atmosphere. In the third, fully coupled simulation (FC), both the bigoeochemical and radiative coupling effects acted simultaneously. I found that climate–carbon sensitivities derived from RAD simulations produced a net ocean carbon storage climate sensitivity that is weaker and a net land carbon storage climate sensitivity that is stronger than those diagnosed from the FC and BGC simulations. For the ocean, this nonlinearity was associated with warming-induced weakening of ocean circulation and mixing that limited exchange of dissolved inorganic carbon between surface and deeper water masses. For the land, this nonlinearity was associated with strong gains in vegetation productivity in the FC simulation that were driven by enhancements in the hydrological cycle and increased nutrient availability. I developed and applied a nonlinearity metric for individual model variables to rank nonlinear responses and drivers. For these simulations, the overall climate–carbon cycle feedback gain at 2300 was 28% lower when estimated from climate–carbon sensitivities derived from the RAD simulation than when derived from the difference between the FC and BGC simulations. The gain estimated from compatible emissions calculations corresponded well with the gain estimated from FC − BGC climate–carbon sensitivity parameters, confirming the validity of the larger gain. This difference has direct implications for carbon management and energy policies because underestimating the climate–carbon cycle feedback gain would result in allowable emissions estimates that would be too low to meet climate change targets.

In these studies, I have shown that 1) we can reduce uncertainties in future climate projections by improving models to more closely match the long-term time series of observed atmospheric CO2; 2) we can reduce sampling biases and partition important environmental gradients to design an optimized network of sampling sites at desired scales; and 3) we can reduce uncertainties in the assessment of climate–carbon cycle feedbacks due to nonlinear terrestrial and marine responses by deriving climate–carbon sensitivities from fully coupled and biogeochemically coupled simulations.

Ecosystems in the Western US face a combination of climate-driven threats including warming temperature, drought, and wildfire. How ecosystems respond to these threats is directly relevant for human health, water availability, and carbon storage toward climate mitigation, among other services. However, there currently exist large uncertainties regarding the impact of a changing climate on ecosystems and the reliability of these ecosystems to serve as carbon sinks going forward. In my dissertation I used statistical and machine learning techniques along with large-scale geospatial datasets to explore several questions at the intersection of climate change and ecosystems. In my first study, I developed a new framework for wildfire prediction. I found that vapor pressure deficit at the time of ignition and the density of black spruce trees surrounding ignition sites could be used to predict in Alaskan ecosystems whether a fire would become large. These two pieces of information could help managers triage wildfires to protect vulnerable ecosystems given limited resources. In my second chapter, I quantified the future impacts of climate change on carbon storage in California ecosystems, finding that warming temperatures are likely to drive a net loss of carbon and increase the challenges associated with meeting the State’s climate mitigation goals. Projected losses were greatest for the mid-elevation mountains, northern coast region, and locations of current forest carbon offset projects. This study also revealed the largest sources of uncertainty to future ecosystem projections, most notably the uncertainty in future precipitation for California and uncertainty in tree species migration rates relative to the climate velocity. In my third chapter, I focused on California’s forest carbon offset projects and used remote sensing datasets to assess whether these projects have led to additional carbon sequestration in the 5-10 years since their initiation. Five lines of evidence related to carbon trends, harvest rates, and species composition in projects relative to similar forests suggested that in general our portfolio of projects has not led to detectable carbon sequestration beyond what would have otherwise occurred. Finally, in my fourth chapter, I quantified future wildfire risk across California based on climate and vegetation projections from my second chapter. I found that California is likely to see substantial increases in wildfire over this century, especially in scenarios with increased precipitation and/or shrub cover. The collective results of my research highlight key scientific uncertainties related to the future of ecosystems, particularly the uncertainty in future precipitation. The results of each chapter also offer key insights which are relevant for effective ecosystem management in a rapidly changing climate. These insights include (1) predicting large fires from the time of ignition, (2) identifying areas of ecosystem vulnerability for carbon sequestration and conservation goals, (3) building a more reliable and systematic framework for assessing additionality of carbon offsets, and (4) identifying areas of greatest future fire risk, where fuels management could have the greatest impact on reducing extreme wildfire outcomes.

Fire impacts climate over wide spatial and temporal scales via many complex, interdependent, and often poorly understood processes. For example, fires emit substantial amounts of greenhouse gases and aerosols. Vulnerable regions worldwide are becoming more prone to larger, and more intense wildfires. Global change in the form of climate change and/or land use change is impacting fire regimes across the globe. In the boreal forest, climate change is causing a combination of warmer temperatures, drought, earlier snowmelt, and favorable fire weather that encourages the development of wildfires. In the tropics, specifically Indonesia, extensive land use change involving the degradation of naturally-existing peatlands to support agricultural productivity is also causing a significant increase in wildfires and corresponding emissions. There is a need to quantify and characterize the composition of emissions in wildfire-prone regions to fully understand the climate and human health related impacts of changing fire regimes.

In my first study, I examined the influence of daily meteorology on boreal forest fire emissions and regional trace gas variability. I coupled a statistical fire combustion model with an inverse atmospheric transport model to quantify the influence of fires on trace gas variability observed at a tower in Alaska equipped with a cavity-ring down spectrometer. I discovered that basic meteorological variables, including temperature and vapor pressure deficit, explained variability in fire activity better than complex fire weather indices. Next, I analyzed environmental controls on boreal forest fire emission factors during an anomalously large fire year in Alaska (2015). I used the same coupled modeling approach to determine times when fires influenced trace gas variability at the tower. Tower observations were used to calculate 23 individual boreal fire emission factors, substantially contributing to the database of emission factors. In my final study, I analyzed wildfire aerosol emissions in Indonesia during the devastating 2015 haze event. Considerable uncertainty exists with regards to whether the particulate emissions originate from deforestation fires, agricultural burning, or peatland fires. To address this, I analyzed aerosol samples collected in Singapore, a major city downwind of Indonesian wildfires, for their radiocarbon content. The radiocarbon content of the fire-emitted aerosol was depleted in 14C, indicating the majority of fire PM2.5 emissions originated from the burning of peatlands. The collective results of the studies from this dissertation will vastly improve our characterization of trace gas and particulate emissions from fires in globally important and vulnerable ecosystems.