No abstract

Hyperspectral instruments provide the spectral detail necessary for extracting multiple layers of information from inherently complex coastal environments. We evaluate the performance of a semi-analytical optimization model for deriving bathymetry, benthic reflectance, and water optical properties using hyperspectral AVIRIS imagery of Kaneohe Bay, Hawaii. We examine the relative impacts on model performance using two different atmospheric correction algorithms and two different methods for reducing the effects of sunglint. We also examine the impact of varying view and illumination geometry, changing the default bottom reflectance, and using a kernel processing scheme to normalize water properties over small areas. Results indicate robust model performance for most model formulations, with the most significant impact on model output being generated by differences in the atmospheric and deglint algorithms used for preprocessing.

The biodiversity and high productivity of coastal terrestrial and aquatic habitats are the foundation for important benefits to human societies around the world. These globally distributed habitats need frequent and broad systematic assessments, but field surveys only cover a small fraction of these areas. Satellite‐based sensors can repeatedly record the visible and near‐infrared reflectance spectra that contain the absorption, scattering, and fluorescence signatures of functional phytoplankton groups, colored dissolved matter, and particulate matter near the surface ocean, and of biologically structured habitats (floating and emergent vegetation, benthic habitats like coral, seagrass, and algae). These measures can be incorporated into Essential Biodiversity Variables (EBVs), including the distribution, abundance, and traits of groups of species populations, and used to evaluate habitat fragmentation. However, current and planned satellites are not designed to observe the EBVs that change rapidly with extreme tides, salinity, temperatures, storms, pollution, or physical habitat destruction over scales relevant to human activity. Making these observations requires a new generation of satellite sensors able to sample with these combined characteristics: (1) spatial resolution on the order of 30 to 100‐m pixels or smaller; (2) spectral resolution on the order of 5 nm in the visible and 10 nm in the short‐wave infrared spectrum (or at least two or more bands at 1,030, 1,240, 1,630, 2,125, and/or 2,260 nm) for atmospheric correction and aquatic and vegetation assessments; (3) radiometric quality with signal to noise ratios (SNR) above 800 (relative to signal levels typical of the open ocean), 14‐bit digitization, absolute radiometric calibration <2%, relative calibration of 0.2%, polarization sensitivity <1%, high radiometric stability and linearity, and operations designed to minimize sunglint; and (4) temporal resolution of hours to days. We refer to these combined specifications as H4 imaging. Enabling H4 imaging is vital for the conservation and management of global biodiversity and ecosystem services, including food provisioning and water security. An agile satellite in a 3‐d repeat low‐Earth orbit could sample 30‐km swath images of several hundred coastal habitats daily. Nine H4 satellites would provide weekly coverage of global coastal zones. Such satellite constellations are now feasible and are used in various applications.

Semantic parsers map natural language statements into meaning representations, and must abstract over syntactic phenomena, resolve anaphora, and identify word senses to eliminate ambiguous interpretations. Abstract meaning representation (AMR) is a recent example of one such semantic formalism which, similar to a dependency parse, utilizes a graph to represent relationships between concepts (Banarescu et al., 2013). As with dependency parsing, transition-based approaches are a common approach to this problem. However, when trained in the traditional manner these systems are susceptible to the accumulation of errors when they find undesirable states during greedy decoding. Imitation learning algorithms have been shown to help these systems recover from such errors. To effectively use these methods for AMR parsing we find it highly beneficial to introduce two novel extensions: noise reduction and targeted exploration. The former mitigates the noise in the feature representation, a result of the complexity of the task. The latter targets the exploration steps of imitation learning towards areas which are likely to provide the most information in the context of a large action-space. We achieve state-ofthe art results, and improve upon standard transition-based parsing by 4.7 F 1 points.

No abstract