Mirela Tulbure
Grants
Spatiotemporal quantification of surface water and flooding is essential for research on hydrological cycles. Satellite remote sensing is the only means of monitoring these dynamics across vast areas and over time. Several regional to global surface water data sets have been developed using optical time-series, either from MODIS-type sensors with coarse spatial resolution but daily frequency, or based on the entire Landsat archive. Despite its high spatial resolution, the 16-day repeat frequency of Landsat means that short-lived hazardous flooding and the maximum extent of large floods are likely missed. Meanwhile, spatially coarser MODIS-type sensors may miss small water bodies and floods entirely. In addition, two limitations when mapping inundation with optical data have been detecting water under vegetation and cloud obscuration, which often coincides with floods. Both issues can be overcome by fusing multiple optical with synthetic aperture radar (SAR) data, taking advantage of complementary observation properties including SAR??????????????????s ability to penetrate through clouds. Thus, combining observations and spectral properties of the newly available Sentinel 1 SAR (S1) and Sentinel 2 (S2) series of satellites with Landsat 8 (L8) holds promise for global surface water and flood mapping with improved spatial and temporal resolution and accuracy. To accurately capture maximum extent of all floods in near real time, our key objectives are to (1) map flooding dynamics globally, using machine learning applied to time-series of multi-sensor optical (L8, S2) and radar (S1) time series data, (2) assess the accuracy of the mapped flood extent, and (3) test the ability of our algorithms to map (a) ephemeral floods in a dynamic dryland river system (b) a complex delta including inundated vegetation in Western Canada (leveraging field validation data on extent of inundated vegetation collected during NASA??????????????????s Arctic Boreal Vulnerability Experiment), (c) extreme flooding in North Carolina (during hurricanes in 2016, 2018 and 2019), and (d) small water bodies (< 5ha) in irrigated areas (i.e. Arkansas, the U.S. state with the 3rd largest irrigated area, where hundreds of small reservoirs have been constructed since 2015). We will use NASA??????????????????s 30m Harmonized L8/S2 (HLS) Products that seamlessly combine L8 and S2 observations, and S1 as input to machine learning-based mapping of surface water and flooding. As training data, we will use the freely available USGS Spatial Procedures for Automated Removal of Cloud and Shadow dataset, which contains ????????????????water??????????????? and ????????????????flooded??????????????? masks. We will further augment flood labels via active learning, by evaluating initial model results and adding labels on misclassified areas. To assess the accuracy of our flood maps we will use a stratified sampling design, with flooding and water as the rare classes used as strata to improve precision of the accuracy estimates. We will assess whether the increased temporal frequency resulting from multiple/fused data streams will result in improved detections of small and short-lived flooding events, and maximum extent of large floods compared to the use of L8, S2 or S1 alone over a dynamic dryland basin (i.e., Australia??????????????????s Murray-Darling Basin), and over small farm dams of Arkansas. To test the improved capacity of flood mapping when adding SAR to HLS during cloudy conditions we will focus on 3 hazardous floods in North Carolina. We will assess the ability of C-band S1 combined with optical image time series to detect water under vegetation in Canada??????????????????s Peace-Athabasca Delta, where detailed validation data will be available. This proposal is significant to this NASA solicitation as it will enable improved quantification of flood extent dynamics and water quantity. The algorithms and maps produced can be used for better mapping of floods during hazardous conditions and assessment of how changes in land cover and land use and climate impact surface water and flood dynamics.
Fresh water stored by on-farm reservoirs (OFRs) is a fundamental component of surface hydrology and is critical for meeting global irrigation needs. Farmers use OFRs to store water during the wet season for crop irrigation during the dry season. There are more than 2.6 million OFRs in the US alone, and many of these OFRs were constructed during the last 40 years. Despite their importance for irrigating crops, OFRs can contribute to downstream water stress by decreasing stream discharge and peak flow in the watersheds where they are built, thereby exacerbating water stress intensified by climate change and population growth. However, modeling the impact of OFRs on surface hydrology remains a challenge because they are so abundant and have frequent fluctuations in surface area and water volume. Prior to the recent availability of satellite data, widespread monitoring of OFRs?????????????????? surface area and water volume across space and time was impossible due to temporal latency of satellite observations. The goal of this project, therefore, is to harness a multi-sensor satellite imagery approach to reduce observation latency and improve surface hydrology modeling, with the aim of supporting more efficient management of OFRs and mitigation of their downstream impacts. Our objectives are: 1) Develop a multi-sensor imagery approach to reduce latency and obtain sub-weekly OFRs surface area and volume change; and 2) Input sub-weekly OFRs volume change into the Soil Water and Assessment Tool (SWAT) model to estimate OFRs?????????????????? impact on surface hydrology. Specifically for Objective 1, a novel method based on the Kalman filter will be used to harmonize data from multiple optical sensors and to provide sub-weekly OFRs surface area change, which will be converted to volume change using area-elevation equations. Then for Objective 2, we will carry out hydrological simulations in SWAT to quantify OFRs?????????????????? impact on simulated daily and monthly stream discharge, simulating stream discharge with and without the OFRs. We will perform yearly simulations, based on satellite imagery availability, to measure OFRs?????????????????? impact during low and peak flows in each watershed of our study region, which will account for both intra- as well as inter-annual variability in flows. This project will monitor OFRs?????????????????? surface area and volume change to enable better assessment and management of water quantity, and further the use of Earth system science to inform decisions and provide benefits to society regarding preservation of surface water resources, both of which are overarching science goals that guide NASA??????????????????s Earth Science Division program.