MONITORING WATER-RELATED LAND COVERS WITH MULTI-SOURCE REMOTE SENSING, BIOPHYSICAL KNOWLEDGE, AND MACHINE LEARNING

Abstract

Water-related land cover types, including natural wetlands, paddy rice, and surface water bodies, play vital roles in biodiversity conservation, climate regulation, food production, and public health. However, these ecosystems have undergone significant changes due to climate change and human activities, particularly in high and middle latitudes such as Northeast Asia, which exhibits extensive wetlands, paddy rice, and transboundary water systems. Existing large-scale water-related land cover data products are often limited by moderate accuracy, outdated observations, and incomplete representation of water-related types. Thus, accurate, high-resolution, and regularly updated maps are needed to support sustainable land and water management, biodiversity protection, and disease surveillance. Landsat satellites provide a 30 m, 8–16-day global record since the 1980s, offering a long-term foundation for historical land surface analysis. Sentinel-2A/B offers 10 m optical observations every 5 days, enabling detailed monitoring of vegetation and surface dynamics. Sentinel-1A/B provides 10 m, 6-day all-weather microwave data that capture surface inundation and vegetation structure, and MODIS and VIIRS thermal imagery adds temperature context for plant growth and agricultural practices. The goal of this dissertation is to develop robust, scalable mapping algorithms for water-related land covers and to produce high-resolution regional and continental maps by integrating optical, microwave, and thermal satellites.Chapter 2 defines natural wetland subtypes by considering remote sensing capabilities and characterizing them in terms of plant growth form, life cycle, leaf seasonality, and canopy type. Unique and stable spectral and/or microwave features of individual wetland types were identified, and knowledge-based algorithms were developed to map these wetlands in Northeast China at 10 m spatial resolution using integrated multi-source satellite data in 2020, including Landsat, Sentinel-2, PALSAR, Sentinel-1, and MODIS. The resultant wetland map achieved an overall accuracy exceeding 95%, revealing a total of 154,254 km2 of wetlands in the region, comprising 27,219 km2 of seasonal open-canopy marsh, 69,158 km2 of yearlong closed-canopy marsh, and 57,878 km2 of deciduous forest swamp. This chapter demonstrates the strong potential of knowledge-based algorithms combined with multi-source imagery for accurate and consistent wetland mapping and monitoring. Chapter 3 improved the knowledge-based mapping of paddy rice by integrating optical (Sentinel-2 and Landsat) and microwave (Sentinel-1) time series data and introducing a confidence-based classification framework to address challenges of cloud contamination and spectral confusion between paddy rice and natural wetlands. By incorporating an additional Land Surface Water Index (LSWI) threshold and Sentinel-1 microwave backscatter dynamics to detect flooding signals, along with a post-flood closed-canopy check, the resultant 10 m paddy rice map for Northeast China in 2020 achieved an overall accuracy of 98%. The total estimated paddy rice area was 60.83 ± 0.86 × 103 km2, with 62% of paddy rice pixels having a confidence level of 1 (detected by both optical and microwave images) and 38% having a confidence level of 0.5 (detected by either optical or microwave images). These findings indicate that knowledge-based paddy rice mapping algorithms and a combination of optical and microwave images hold great potential for timely, accurate paddy rice mapping in large-scale complex landscapes. Chapter 4 investigated the spatial–temporal dynamics of surface water area (SWA) and total water storage (TWS) across Northeast Asia from 2000 to 2023 by integrating Landsat and Sentinel-2 imagery with GRACE/GRACE-FO data. Annual surface water maps were generated to quantify long-term changes and assess their drivers and impacts. The results revealed a net SWA loss of 16 × 103 km2 in Far East Russia, primarily driven by rising temperature and evaporative demand, and a net SWA gain of 3 × 103 km2 in Northeast China, mainly due to increasing precipitation and agricultural irrigation infrastructure. Approximately 1004 0.5° gridcells (1.4 × 106 km2) exhibited concurrent losses of both SWA and TWS. Approximately 185 million people resided in watersheds experiencing SWA or TWS loss. The observed reductions in SWA and TWS, especially in densely populated regions such as Japan, Northeast China, and South Korea, highlight growing water stress under climate change and human pressure, emphasizing the need for integrated management strategies that consider both environmental and demographic changes. Chapter 5 further developed an integrated framework that combined knowledge-based and deep learning algorithms to map comprehensive water-related land covers. Representative and temporally consistent training samples were automatically generated using knowledge-based rules, which were then applied in deep neural network models with multi-source time series imagery (VIIRS, Sentinel-1, Sentinel-2, and Landsat). The resultant 30 m water-related land cover map for Northeast Asia in 2021 achieved an overall accuracy of 93.8 ± 0.2%, identifying 1,239,234 km2 of water-related land covers, including 17% yearlong surface water, 76% natural wetlands (15% seasonal open-canopy marshes and 61% yearlong closed-canopy marshes), and 7% paddy rice. Our map captured extensive yearlong closed-canopy marshes that have been underrepresented or omitted in previous datasets. This study provides a comprehensive and transferable mapping approach, supporting continental-scale environmental monitoring and land management. Together, these studies demonstrate the power of combining multi-source satellite imagery and integrating physical knowledge with data-driven learning for high-accuracy, large-scale mapping of dynamic water-related ecosystems. The approaches developed here provide critical data and methodologies for understanding environmental change, promoting sustainable resource management, and supporting regional and continental ecosystem monitoring.