Text

theses

The global carbon cycle has changed in response to climate change, and the effects of these changes, caused by anthropogenic factors such as the burning of fossil fuels and landscape alterations, are expected to be widespread. Terrestrial gross primary productivity (GPP), the largest component flux of the global carbon cycle, plays a significant role in connecting the global carbon and water cycles and the energy balance between the atmosphere, biosphere, hydrosphere and pedosphere. Despite the development of various approaches and models for estimating terrestrial GPP at different scales, large discrepancies and uncertainties remain in long-term global GPP simulations. Therefore, it is of great value and necessity to better understand and accurately estimate the spatial and temporal patterns of terrestrial GPP. In this dissertation, we improved the performance of global terrestrial GPP simulation by: 1) improving the solar radiation transfer model within a canopy by considering multiple scattering and radiation partitioning; 2) reconstructing satellite-based leaf area index (LAI) data to minimize biases and errors caused by cloud contaminations and composite technique; 3) using high performance computing of the Google Earth Engine (GEE) platform. We estimated global terrestrial GPP at 0.25° spatial resolution and 3-hour temporal intervals using our integrated process-based ecosystem model from 2001 to 2020. In Topic1, we evaluated the performance of five climate variables derived from a new reanalysis dataset - air temperature, precipitation, downward shortwave radiation, air pressure, and vapor pressure deficit (VPD) - against observations from 167 worldwide flux tower sites at both daily and annual scales. The results showed that all of the variables performed reliably, with the exception of precipitation, which had a tendency to be overestimated. In addition, we examined the temporal and spatial patterns of these variables from 2001 to2020. We found that global air temperature, solar radiation, VPD, and precipitation showed significantly increasing trends at rates of 0.7°C/decade, 3.1W/m2/decade, 0.15KPa/decade, and 49.6mm/decade, respectively, while air pressure did not show any significant changes over this time period. The climate variables also showed different spatial variations at the global scale and their changes over the past decades were not homogenous in space. In addition to evaluating the climate variables, we also assessed the performance of reconstructing MODIS LAI products in 24 typical regions, which covered a range of major climate and vegetation types. The MODIS LAI datasets were affected by cloud contamination and composite techniques and did not perform well in areas with long-term continuous cloud cover, where LAI values were severely underestimated. We developed a new clean-up algorithm to improve the LAI data by including spatiotemporal correlations of neighboring pixels and applied double logistic functions to achieve continuous LAI time series. The results showed that most of the outliers were detected and removed, and the fitted double logistic curves well characterized the variations and patterns of annual LAI, reasonably captured the timing of vegetation phenology between growing and non-growing seasons, and retained the duration of peak within the growing season for both single vegetation cycle and double vegetation cycles.
In Topic 2, we found that the good performance of the empirical radiation partitioning approach indicated that it could be used to derive the two radiation components - direct and diffuse - when only total solar radiation information was available. Additional, the absorption fraction simulated by the two stream approach, which considered multiple scattering, was lower than that estimated by Beer’s law regardless of the LAI and diffuse radiation fraction. The discrepancy in absorption fraction reached up to 73% in an overcast day. We further compared the performance of the Beer’s law (BL) model, the two-stream big-leaf (TS-BL) model, and our integrated radiative transfer (RTM) model – the two-stream two-leaf (TS-TL) model - in simulating GPP and found that our TS-TL model reduced the RMSE and bias by up to 72% and 81% based on the BL model, and up to 63% and 75% based on the TS-BL model, respectively. Overall, our integrated RTM (TS-TL model) exhibited large improvements and robust performance in estimating GPP, especially in areas with a dense vegetation cover.
In Topic 3, we developed a comprehensive process-based ecosystem model, driven by new reanalysis climate data and satellite-based LAI data, to estimate global GPP by using different biochemical photosynthesis models for C3 and C4 plants on the GEE platform. The results were evaluated by comparing the simulated GPP to observations from 167 flux tower sites, and the modeled GPP estimates were highly correlated to the flux tower observations for all vegetation types at both half-hour and annual scales. The annual global terrestrial GPP simulated by our integrated model ranged from 118 PgC to 134 PgC, with an average of 128 PgC, during 2001-2020, and showed a significantly increasing trend with an average rate of 0.71 PgC/yr globally. When compared to recent GPP estimates and products, our simulated results were within a reasonable range of global terrestrial GPP estimations but had some discrepancies due to the different models, parameters, and driving data used to simulate GPP. In addition, the sensitivity analysis exhibited that our simulated GPP was most sensitive to the biophysiological parameters V_cmax25 and LAI, highlighting the need for accurate biophysiological parameters at large scales.

http://dissertations.umi.com/gsnb.rutgers:12343

Ph.D.

Includes bibliographical references

doi:10.7282/t3-a9c3-vh37

The author owns the copyright to this work.