See Upcoming Public Defences

PhD Defence Linlin Li | Satellite-based monitoring of surface water dynamics

satellite-based monitoring of surface water dynamics

Linlin Li is a PhD student in the department of Natural Resources. Her supervisors are prof.dr. A.K. (Andrew) Skidmore, dr. T. (Tiejun) Wang and A. (Anton) Vrieling from the Faculty of Geo-Information Science and Earth Observation.

Terrestrial surface water plays an important role in the global hydrological cycle, biodiversity conservation, and climate processes. Changes in surface water, caused by both natural and human-induced factors, strongly affect socioeconomic development, ecosystem functioning, species distributions and composition, and further influence climate change. Recently, water extent has been identified as an Essential Climate Variable (ECV) for assessing progress towards the Aichi targets for 2020 of the Convention of Biological Diversity. The changes in the extent of water-related ecosystems over time is also an indicator of the Sustainable Development Goals (SDGs). Therefore, knowledge about the spatial and temporal distribution of surface water is needed to support sustainable development and climate change assessment.

Remote sensing provides an effective way to monitor surface water in space and time. Many approaches and datasets have been developed for this purpose. However, measuring long-term changes at fine spatial and temporal resolution remains a challenge due to the trade-off between spatial and temporal resolution of remotely sensed imagery. The main objective of this thesis is to improve long-term mapping and monitoring of surface water extent at fine temporal resolution using high-frequency optical remote sensing data provided by the Moderate Resolution Imaging Spectroradiometer (MODIS), in a way that effectively accounts for small-sized (e.g., smaller than a 500x500 m MODIS cell) and dynamic water bodies. To achieve this goal, this thesis evaluates options to estimate sub-pixel surface water fraction, i.e., the percentage of surface water within a single grid cell. Several machine learning approaches that incorporate MODIS spectral information, temporal characteristics of spectral information, and topographic information were evaluated for accurately mapping and monitoring sub-pixel surface water fraction. This was explored at spatial scales ranging from a small individual wetland to the entire Mediterranean region.

In this thesis, the robustness of machine learning algorithms for mapping and monitoring sub-pixel surface water fraction at large spatial scales was demonstrated. It revealed that a single model could accurately assess water body extent and dynamics in different environmental and climatic conditions, as long as good-quality training data were collected that represent the various environmental conditions. Therefore, there is potential to scale the water fraction mapping approach to larger spatial regions, such as for the globe. 

The need for high-frequency monitoring of surface water is highlighted in the thesis. While much progress has been made recently with global Landsat-based surface water products, these can have large spatial and temporal gaps due to both the limited number of acquisitions and persistent cloud cover, preventing an accurate assessment of water resources variability. Here, using the high-frequency MODIS data and the approach developed in the thesis, a new dense 18-year surface water fraction (SWF) dataset was produced for the Mediterranean region at 500 m resolution and 8-day interval. This MODIS SWF dataset documents the long-term (2000–2017) status of surface water bodies, their location, extent, and change. MODIS SWF complements existing fine spatial resolution water datasets, especially by offering better temporal information for areas suffering from persistent cloud cover during part of the year. Moreover, it allows accurate assessment of surface water seasonality, capturing water extent fluctuations in temporary and ephemeral water bodies, including short-duration surface water that could not be captured by existing datasets of lower temporal resolution. The dataset also accurately detects small water bodies (less than one MODIS pixel) and narrow rivers.