An official website of the United States government
Abstract Classifying images using supervised machine learning (ML) relies on labeled training data—classes or text descriptions, for example, associated with each image. Data‐driven models are only as good as the data used for training, and this points to the importance of high‐quality labeled data for developing a ML model that has predictive skill. Labeling data is typically a time‐consuming, manual process. Here, we investigate the process of labeling data, with a specific focus on coastal aerial imagery captured in the wake of hurricanes that affected the Atlantic and Gulf Coasts of the United States. The imagery data set is a rich observational record of storm impacts and coastal change, but the imagery requires labeling to render that information accessible. We created an online interface that served labelers a stream of images and a fixed set of questions. A total of 1,600 images were labeled by at least two or as many as seven coastal scientists. We used the resulting data set to investigate interrater agreement: the extent to which labelers labeled each image similarly. Interrater agreement scores, assessed with percent agreement and Krippendorff's alpha, are higher when the questions posed to labelers are relatively simple, when the labelers are provided with a user manual, and when images are smaller. Experiments in interrater agreement point toward the benefit of multiple labelers for understanding the uncertainty in labeling data for machine learning research.
more »
« less
An Active Learning Pipeline to Detect Hurricane Washover in Post-Storm Aerial Images
We present an active learning pipeline to identify hurricane impacts on coastal landscapes. Previously unlabeled post-storm images are used in a three component workflow — first an online interface is used to crowd-source labels for imagery; second, a convolutional neural network is trained using the labeled images; third, model predictions are displayed on an interactive map. Both the labeler and interactive map allow coastal scientists to provide additional labels that will be used to develop a large labeled dataset, a refined model, and improved hurricane impact assessments.
more »
« less
- PAR ID:
- 10222367
- Date Published:
- Journal Name:
- AI for Earth Sciences Workshop at NeurIPS 2020
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Scientists continue to study the red tide and fish-kill events happening in Florida. Machine learning applications using remote sensing data on coastal waters to monitor water quality parameters and detect harmful algal blooms are also being studied. Unmanned Surface Vehicles (USVs) and Autonomous Underwater Vehicles (AUVs) are often deployed on data collection and disaster response missions. To enhance study and mitigation efforts, robots must be able to use available data to navigate these underwater environments. In this study, we compute a satellite-derived underwater environment (SDUE) model by implementing a supervised machine learning model where remote sensing reflectance (Rrs) indices are labeled with in-situ data they correlate with. The models predict bathymetry and water quality parameters given a recent remote sensing image. In our experiment, we use Sentine1-2 (S2) images and in-situ data of the Biscayne Bay to create an SDUE that can be used as a Chlorophyll-a map. The SDUE is then used in an Extended Kalman Filter (EKF) application that solves an underwater vehicle localization and navigation problem.more » « less
-
Hurricane evacuations require fast updates of coastal inundation predictions based on the update of hurricane forecasting track. NOAA usually updates the hurricane track at about 6 hr interval. This paper presents a multi-scale nested modeling method for faster simulations of storm surge and coastal inundations. A medium-resolution model with minimum mesh size of 1200 m for the Gulf of Mexico is used to simulate the storm surge in the Gulf of Mexico. A high-resolution model with 120m-150m mesh sizes is used to predict coastal inundations in the area of potential hurricane landfall. A nested modeling method has been developed to transfer boundary conditions from the large-scale storm surge model to the nested local-scale high-resolution model. The nested models have been satisfactorily validated and applied in the case study of Hurricane Michael. Results indicate that, by applying the nested models, it takes about 85 minutes for the simulation of one hurricane track for a 5-day forecasting, which will provide sufficient time before the next NOAA forecast update of hurricane’s track in 6 hour interval. The nested model application to the case study of Hurricane Michael demonstrates the coastal inundation patterns in the city of Mexico Beach with the root-mean-square error of 0.12 m from all measurement stations. Results of the nested model inundations on coastal critical infrastructure and roadways are further used with models that investigate risk assessments to support hurricane mitigation planning and evacuation operations sufficiently in advance.more » « less
-
Abstract The world’s coastlines are spatially highly variable, coupled-human-natural systems that comprise a nested hierarchy of component landforms, ecosystems, and human interventions, each interacting over a range of space and time scales. Understanding and predicting coastline dynamics necessitates frequent observation from imaging sensors on remote sensing platforms. Machine Learning models that carry out supervised (i.e., human-guided) pixel-based classification, or image segmentation, have transformative applications in spatio-temporal mapping of dynamic environments, including transient coastal landforms, sediments, habitats, waterbodies, and water flows. However, these models require large and well-documented training and testing datasets consisting of labeled imagery. We describe “Coast Train,” a multi-labeler dataset of orthomosaic and satellite images of coastal environments and corresponding labels. These data include imagery that are diverse in space and time, and contain 1.2 billion labeled pixels, representing over 3.6 million hectares. We use a human-in-the-loop tool especially designed for rapid and reproducible Earth surface image segmentation. Our approach permits image labeling by multiple labelers, in turn enabling quantification of pixel-level agreement over individual and collections of images.more » « less
-
Floods are often associated with hurricanes making landfall. When tropical cyclones/hurricanes make landfall, they are usually accompanied by heavy rainfall and storm surges that inundate coastal areas. The worst natural disaster in the United States, in terms of loss of life and property damage, was caused by hurricane storm surges and their associated coastal flooding. To monitor coastal flooding in the areas affected by hurricanes, we used data from sensors aboard the operational Polar-orbiting and Geostationary Operational Environmental Satellites. This study aims to apply a downscaling model to recent severe coastal flooding events caused by hurricanes. To demonstrate how high-resolution 3D flood mapping can be made from moderate-resolution operational satellite observations, the downscaling model was applied to the catastrophic coastal flooding in Florida due to Hurricane Ian and in New Orleans due to Hurricanes Ida and Laura. The floodwater fraction data derived from the SNPP/NOAA-20 VIIRS (Visible Infrared Imaging Radiometer Suite) observations at the original 375 m resolution were input into the downscaling model to obtain 3D flooding information at 30 m resolution, including flooding extent, water surface level and water depth. Compared to a 2D flood extent map at the VIIRS’ original 375 m resolution, the downscaled 30 m floodwater depth maps, even when shown as 2D images, can provide more details about floodwater distribution, while 3D visualizations can demonstrate floodwater depth more clearly in relative to the terrain and provide a more direct perception of the inundation situations caused by hurricanes. The use of 3D visualization can help users clearly see floodwaters occurring over various types of terrain conditions, thus identifying a hazardous flood from non-hazardous flood types. Furthermore, 3D maps displaying floodwater depth may provide additional information for rescue efforts and damage assessments. The downscaling model can help enhance the capabilities of moderate-to-coarse resolution sensors, such as those used in operational weather satellites, flood detection and monitoring.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.31223/X5JW23
