Sanmei Li

A New Short-Wave Infrared (SWIR) Method for Quantitative Water Fraction Derivation and Evaluation With EOS/MODIS and Landsat/TM Data

A quantitative method is developed for deriving water fraction from coarse- to medium-resolution satellite data with visible to short-wave infrared (SWIR) channels based on the linear mixture theory. The method uses a SWIR channel (1.64 μm) by assuming that the water-surface-leaving radiance in this channel is insignificant and is thus less affected by water types and water depth than near-infrared (NIR) channels for inland water bodies. For a mixed water pixel, a dynamic nearest neighbor searching (DNNS) method is used to find the nearby land pixels to determine the average land reflectance. The nearby pure water pixels with a similar water type to the subpixel water portion of the mixed water pixel are found dynamically to derive the average water reflectance. The average reflectance in the SWIR channel from both pure land pixels and water pixels is used to calculate the water fraction from a linear mixture model. The developed method is applied to Moderate Resolution Imaging Spectroradiometer (MODIS) data and shows promising results. High-resolution satellite data from the Thematic Mapper (TM) are used to evaluate the water fraction derived from MODIS. During pixel-to-pixel water fraction evaluation, TM data are spatially aggregated to MODIS resolution. When evaluated against the high-resolution TM observations, water fractions derived from MODIS using the DNNS method with the SWIR channel show a bias of -0.021 with a standard deviation of 0.0338. Comparing lake areas between TM and MODIS data also shows consistent results with the pixel-to-pixel water fraction comparison. The DNNS method is also compared to the traditional histogram method both with SWIR channel and NIR channel. The results show that the DNNS method is more accurate than the histogram method and that the SWIR channel is better than the NIR channel to derive highly accurate water fraction from coarse- to medium-resolution satellite data.

Towards Operational Automatic Flood Detection Using EOS/MODIS Data

This study investigates how to derive water fraction and flood map from the Moderate-Resolution Imaging Spectroradiometer (MODIS) using a Regression Tree (RT) approach, which can integrate all predictors. The New Orleans, Louisiana floods in August 2005 were selected as a case study. MODIS surface reflectance with matched water fraction data were used for training. The tree-based regression models were obtained automatically through learning process. The tree structure reveals that near-infrared reflectance is more important than the difference and ratio between near-infrared and visible channels for water fraction estimate. Flood distributions were generated using the differences in water fraction values between after and before the flooding. The derived water fractions were evaluated against 30 m Thematic Mapper (TM) data from Landsat observations. Water fractions derived from the MODIS and TM data agree well (R 2 = 0.94, bias = 0.38 percent, and RMSE = 4.35 percent). The results show that the RT approach in dynamic monitoring of floods is acceptable.

Mapping floods due to Hurricane Sandy using NPP VIIRS and ATMS data and geotagged Flickr imagery

In this study, we present an approach to estimate the extent of large-scale coastal floods caused by Hurricane Sandy using passive optical and microwave remote sensing data. The approach estimates the water fraction from coarse-resolution VIIRS and ATMS data through mixed-pixel linear decomposition. Based on the water fraction difference, using the physical characteristics of water inundation in a basin, the flood map derived from the coarse-resolution VIIRS and ATMS measurements was extrapolated to a higher spatial resolution of 30 m using topographic information. It is found that flood map derived from VIIRS shows less inundated area than the Federal Emergency Management Agency (FEMA) flood map and the ground observations. The bias was mainly caused by the time difference in observations. This is because VIIRS can only detect flood under clear conditions, while we can only find some clear-sky data around the New York area on 4 November 2012, when most flooding water already receded. Meanwhile, microwave measurements can penetrate through clouds and sense surface water bodies under clear-or-cloudy conditions. We therefore developed a new method to derive flood maps from passive microwave ATMS observations. To evaluate the flood mapping method, the corresponding ground observations and the FEMA storm surge flooding (SSF) products are used. The results show there was good agreement between our ATMS and the FEMA SSF flood areas, with a correlation of 0.95. Furthermore, we compared our results to geotagged Flickr contributions reporting flooding, and found that 95% of these Flickr reports were distributed within the ATMS-derived flood area, supporting the argument that such crowd-generated content can be valuable for remote sensing operations. Overall, the methodology presented in this paper was able to produce high-quality and high-resolution flood maps over large-scale coastal areas.

Evaluation of 10 year AQUA/MODIS land surface temperature with SURFRAD observations

As the 10 year Moderate Resolution Imaging Spectroradiometer Land Surface Temperature MODIS LST becomes available, it is significant to perform a comprehensive evaluation on the long-term product before downstream users use it for climate studies and atmospheric models. In this study, a validation is carried out using observations from the US Surface Radiation budget (SURFRAD) network. Strict quality control removes cloud-contaminated samples from MODIS LST collection and decreases noise information from SURFRAD measurements, thereby making the validation more persuasive. With analysis on 19,735 valid samples, Aqua/MODIS LST from a split-window algorithm shows retrieval errors from –14 K to 17 K with a bias of –0.93 K, an RMSE of 2.65 K, and a standard deviation of 2.48 K. The errors also show strong seasonal signals. With correlation tests between LST errors and several other factors, it is disclosed that LST retrieval errors mainly come from atmospheric effects and surface emissivity uncertainties, which are closely related to relative air humidity, absolute air humidity, sensor zenith angle, wind speed, normalized difference vegetation index (NDVI), and soil moisture. In addition, the impacts from these factors may not be independent. These impact factors suggest a deficiency of the split-window algorithm in dealing with atmospheric and surface complexity and variety.

Object-based automatic terrain shadow removal from SNPP/VIIRS flood maps

Terrain shadow is a big challenge to land products such as flood extent and snow cover from moderate-resolution optical satellite data. Because terrain shadows share similar spectral features with floodwaters, they can be easily detected as floodwaters by flood detection algorithms based on spectral features in visible, near-infrared, and short-wave infrared channels, which decreases the accuracy of flood detection substantially. However, because terrain shadows appear in mountainous areas with large surface roughness while floodwaters accumulate in low-lying areas with small surface roughness, analysis on surface roughness between terrain shadows and floodwaters can be very effective to distinguish one from the other. Root-mean-square height, internal height difference, and external height difference are used as principal quantitative surface roughness parameters in this study and calculated upon water objects that are clustered from a group of adjacent water pixels. This object-based method is applied in terrain shadow removal from SNPP/VIIRS (Suomi National Polar Orbit Partnership/Visible Infrared Imager Radiometer Suite) near-real-time flood maps and shows promising results according to the tests with 10,000+ VIIRS granules across global areas. Quantitative evaluation in the northwest of the USA also indicates that more than 99% terrain shadow pixels could be removed from VIIRS flood maps by this method, which significantly improves the accuracy of near-real-time flood detection from SNPP/VIIRS imagery.

Automatic sea fog detection over Chinese adjacent oceans using Terra/MODIS data

Sea fog is a problematic weather phenomenon for marine transportation and navigation. Lacking ground observations, sea fog monitoring mainly depends on meteorological and environmental satellites because they provide large swaths of data with good spatial and temporal resolution. The Moderate Resolution Imaging Spectroradiometer (MODIS) on board the Terra and Aqua satellites provides several new features for sea fog detection because of the more available channels. In addition to the traditional variable-dual channel difference (DCDIR), which is widely used to detect sea fog and stratus clouds at night, this study uses and analyses several other variables including NDSI (normalized difference snow index), BTDback (brightness temperature difference in the thermal infrared channel between a sea fog/stratus cloud pixel and nearby clear-sky ocean surface), NWVI (normalized difference near-infrared water vapour index) and D_NWVI (NWVI difference between a possible sea fog/stratus cloud pixel and nearby clear-sky ocean surface), for all seasons. BTDback, NWVI, and D_NWVI show outstanding ability to discriminate between sea fog and stratus clouds. Automatic sea fog detection algorithms are developed using these variables for both daytime and night time with Terra/MODIS data based on a threshold scheme. During development of the algorithms, a series of data processes are also considered to maintain stable performance of the algorithms over wide areas and in all seasons. The algorithms are applied to Terra/MODIS data at a semi-operational mode from 2007 to 2013 and show promising results. Validation with data from field campaigns, one buoy station with good maintenance, 18 weather stations, and CALIOP (Cloud Aerosol Lidar with Orthogonal Polarization)/CALIPSO (Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations) demonstrates the accuracy of the algorithms. The sea fog detection results are highly consistent with fog observations from the field campaign data. The validation with the buoy station data shows an overall accuracy of 90% under all weather conditions, an accuracy of 86% during foggy weather condition, and a KSS (Hanssen–Kuiper Skill Score) of 0.81. The validation with two-year data from 18 weather stations and CALIOP/CALIPSO over the Bohai Sea and Yellow Sea shows accuracy of 76.3% and 77.9%, respectively. The promising results indicate high probability of applying the algorithms in operational systems over the oceans adjacent to China or even wider oceans using Terra/MODIS data.

Mapping coastal floods induced by hurricane storm surge using ATMS data

ABSTRACT The coastal floods induced by hurricane storm surge are frequent, costly, and deadly hazards. Accurately and quickly estimating the spatial extent of floods is highly important for relief and rescue operations. In this study, we present an approach to estimate the extent of large-scale coastal floods caused by Hurricane Sandy in late October 2012 using passive microwave remote-sensing data. The approach estimates the water fraction from coarse-resolution Advanced Technology Microwave Sounder (ATMS) data through mixed-pixel linear decomposition. Land and water sample regions generated by river density and land-cover data, the relationship of channels 3, 4, and 16, neighbourhood pixel searching, and the difference of ATMS channels 4 and 3 are all comprehensively taken into account to dynamically determine water and land end members. The difference in the water fraction at the basin scale before and after flooding is calculated to reduce the impacts of soil and vegetation and to avoid pixel-to-pixel errors. Based on the water fraction difference, using the physical characteristics of water inundation that always proceed from the lowest to the highest elevation points in a basin, the flood map derived from the coarse-resolution ATMS measurements was extrapolated to a higher spatial resolution of 100 m using topographic information. Together, these steps represent a water fraction and high-resolution flood (WFHF) mapping process. To evaluate the WFHF mapping methodology presented in this study, the corresponding ground observations (storm-tide sites and high-water-mark data) and the Federal Emergency Management Agency 3 m resolution Hurricane Sandy storm surge flooding (SSF) products are used. The results show that 88% of the storm-tide and high-water-mark sites were located within the WFHF-mapped flood area. There was also good agreement between our WFHF and the SSF areas, with an accuracy of 88% and a correlation of 0.94. Overall, the proposed WFHF methodology was able to produce high-quality and high-resolution flood maps over large-scale coastal areas.

Contributions of Operational Satellites in Monitoring the Catastrophic Floodwaters Due to Hurricane Harvey

Hurricane Harvey made landfall as a Category-4 storm in the United States on 25 August 2017 in Texas, causing catastrophic flooding in the Houston metropolitan area and resulting in a total economic loss estimated to be about

Application of suomi-npp/viirs data in near real time flood detection

125 billion. To monitor flooding in the areas affected by Harvey, we used data from sensors aboard the Suomi National Polar-Orbiting Partnership Satellite (SNPP) and the new Geostationary Operational Environmental Satellite (GOES)-16. The GOES-16 Advanced Baseline Imager (ABI) observations are available every 5 min at 1-km spatial resolution across the entire United States, allowing for the possibility of frequent cloud free views of the flooded areas; while the higher resolution 375-m imagery available twice per day from the Visible Infrared Imaging Radiometer Suite (VIIRS) aboard the SNPP satellite can observe more details of the flooded regions. Combining the high spatial resolution from VIIRS with the frequent observations from ABI offers an improved capability for flood monitoring. The flood maps derived from the SNPP VIIRS and GOES-16 ABI observations were provided to the Federal Emergency Management Agency (FEMA) continuously during Hurricane Harvey. According to FEMA’s estimate on 3 September 2017, approximately 155,000 properties might have been affected by the floodwaters of Hurricane Harvey.

Coastal flood monitoring based on AMSR-E data

As the costliest natural disaster over the globe, floods are predicted to become more frequent in most climate change forecasts. In high latitudes, floods are caused by ice jam and snow melt almost every break-up season. Floods caused by intense rainfall also threaten lives and social infrastructures. Near real-time satellite-derived flood maps are invaluable to river forecasters and decision-makers for disaster monitoring and relief efforts.