Multiscale Drought Monitoring and High-Resolution GPP Estimation Using Remote Sensing and Data-Driven Approaches

Abstract

This thesis addresses the challenge of monitoring and quantifying drought-induced variability in terrestrial carbon uptake by integrating satellite observations with data-driven approaches. Drought can strongly affect ecosystem productivity, yet its impacts remain difficult to quantify. This difficulty arises from the multidimensional nature of drought processes. It is further compounded by limitations in existing gross primary productivity (GPP) products, including restricted temporal coverage and coarse spatiotemporal resolution. Chapter 1 introduces the research background and outlines key limitations of existing approaches for drought monitoring and terrestrial carbon-uptake assessment. In Chapter 2, a novel drought monitoring framework—Vector Projection Analysis and the Vector Projection Index of Drought (VPID)—is developed to address limitations of conventional single-scale drought indices. The framework integrates multiple satellite-derived drought factors to capture shared characteristics across multiscale drought indicators. The resulting integrated multiscale index (VPIDinte) demonstrates strong agreement with the U.S. Drought Monitor and historical disaster records. Chapter 3 presents the UNified, high-resolution Intelligent Carbon QUantification and Explanation (UNIQUE) framework for estimating GPP at high spatial and temporal resolution. UNIQUE first trains machine learning models separately for Landsat and MODIS observations. The resulting GPP estimates are then combined using a deep learning–based spatiotemporal fusion model to produce daily GPP at 30-m resolution. By addressing the long-standing trade-off between spatial and temporal resolution, UNIQUE outperforms existing global GPP products and captures fine-scale variability in heterogeneous and cloud-prone environments. In Chapter 4, a case study in Saunders County, Nebraska, evaluates the ability of UNIQUE-derived GPP estimates to capture drought-driven variability in GPP at local site scales during the 2012 U.S. extreme drought. Using a time series from 2010 to 2012 that encompasses the 2012 U.S. drought, the high-resolution GPP estimates distinguish contrasting responses between irrigated and rainfed agricultural systems and reveal spatially heterogeneous crop-specific GPP declines consistent with flux-tower observations. Overall, this dissertation develops flexible, observation-driven frameworks for monitoring drought conditions and terrestrial carbon uptake across diverse spatial and temporal scales. The results demonstrate that the proposed methods not only improve drought and GPP monitoring capabilities but also effectively characterize drought–carbon interactions, particularly under extreme drought conditions and in landscapes with pronounced local heterogeneity. These contributions support improved understanding of climate–carbon feedbacks and enable more robust drought and ecosystem monitoring under a changing climate.