Learning a River Network Extractor Using an Adaptive Loss Function

Isikdogan, Furkan; Bovik, Alan C.; Passalacqua, Paola

doi:10.1109/lgrs.2018.2811754

Cited by 23 publications

(19 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Because of the difficulty in extracting delta networks and the manual labor involved, studies up to date have analyzed metrics only in small systems or have focused on bulk metrics that do not capture how the characteristics of 5 the delta may vary spatially and temporally. We expect that further development in automatic approaches for delta network extraction (Isikdogan et al, 2017b(Isikdogan et al, , 2018 and for the analysis of network change over time (Jarriel et al, 2019) will enable similar analyses at the global scale and over time.…”

Section: Metrics Importance and Applicability Of The Approach To Othementioning

confidence: 99%

“…Applications span a wide range of topics, such as: streamflow modeling and forecasting (Asefa et al, 2006;Rasouli et al, 2012;Shortridge et al, 2016), runoff modeling 25 (Gudmundsson and Seneviratne, 2015), flood risk assessment (Dibike and Solomatine, 2001;Tehrany et al, 2014;Mojaddadi et al, 2017), sediment yield variability (Tamene et al, 2006), and sediment transport (Bhattacharya et al, 2007;Melesse et al, 2011;Schmelter et al, 2011;Choubin et al, 2018). Particularly relevant to the analysis of river networks is the work on surface water extraction (Pekel et al, 2016;Donchyts et al, 2016;Isikdogan et al, 2017aIsikdogan et al, , 2019 and on delta network extraction (Isikdogan et al, 2018).…”

mentioning

confidence: 99%

See 1 more Smart Citation

Dominant process zones in a mixed fluvial-tidal delta are morphologically distinct

Perignon¹,

Adams²,

Overeem³

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

Abstract. The morphology of deltas is determined by the spatial extent and variability of the geomorphic processes that shape them. While in some cases resilient, deltas are increasingly threatened by natural and anthropogenic forces, such as sea level rise and land use change, which can drastically alter the rates and patterns of sediment transport. Quantifying process patterns can improve our predictive understanding of how different zones within delta systems will respond to future change. Available remotely sensed imagery can help but appropriate tools are needed for pattern extraction and analysis. We present a method for extracting information about the nature and spatial extent of active geomorphic processes across deltas from ten parameters quantifying the geometry of each of 1239 islands and the channels around them using machine learning. The method consists of a two-step unsupervised machine learning algorithm that clusters islands into spatially continuous zones based on the ten morphological metrics extracted from remotely sensed imagery. By applying this method to the Ganges–Brahmaputra–Meghna Delta, we find that the system can be divided into six major zones. Classification results show that active fluvial island construction and bar migration processes are limited to relatively narrow zones along the main Ganges River and Brahmaputra and Meghna corridors, whereas zones in the mature upper delta plain, with smaller fluvial distributary channels stand out as their own morphometric class. The classification also shows good correspondence with known gradients in the influence of tidal energy with distinct classes for islands in the backwater zone and in the purely tidally-controlled region of the delta. Islands at the delta front, under the mixed influence of tides, fluvial-estuarine construction, and local wave reworking have their own characteristic shape and channel configuration. The method does not distinguish between islands with embankments (polders) and natural islands in the nearby mangrove forest (Sundarbans), suggesting that human modifications have not yet altered the gross geometry of the islands beyond their previous natural morphology. These results demonstrate that machine learning and remotely sensed imagery are useful tools for identifying the spatial patterns of geomorphic processes across delta systems.

show abstract

Section: Metrics Importance and Applicability Of The Approach To Othementioning

confidence: 99%

mentioning

confidence: 99%

Dominant process zones in a mixed fluvial-tidal delta are morphologically distinct

Perignon¹,

Adams²,

Overeem³

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Because of the difficulty in extracting delta networks and the manual labor involved, studies up to now have analyzed metrics only in small systems or have focused on bulk metrics that do not capture how the characteristics of the delta may vary spatially and temporally. We expect that further development in automatic approaches for delta network extraction (Isikdogan et al, 2017b(Isikdogan et al, , 2018 and for the analysis of network change over time (Jarriel et al, 2019) will enable similar analyses at the global scale and over time. Hierarchical clustering of delta islands according to their common characteristics can also allow the identification of areas of the landscape that would be affected by different forecasted scenarios of future environmental conditions.…”

Section: Metrics Importance and Applicability Of The Approach To Othementioning

confidence: 99%

“…Applications span a wide range of topics, such as streamflow modeling and forecasting (Asefa et al, 2006;Rasouli et al, 2012;Shortridge et al, 2016), runoff modeling (Gudmundsson and Seneviratne, 2015), flood risk assessment (Dibike and Solomatine, 2001;Tehrany et al, 2014;Mojaddadi et al, 2017), sediment yield variability (Tamene et al, 2006), and sediment transport (Bhattacharya et al, 2007;Melesse et al, 2011;Schmelter et al, 2011;Choubin et al, 2018). Particularly relevant to the analysis of river networks is the work on surface water extraction (Pekel et al, 2016;Donchyts et al, 2016;Isikdogan et al, 2017aIsikdogan et al, , 2019 and on delta network extraction (Isikdogan et al, 2018).…”

Section: Introductionmentioning

confidence: 99%

Dominant process zones in a mixed fluvial–tidal delta are morphologically distinct

et al. 2020

Self Cite

View full text Add to dashboard Cite

Abstract. The morphology of deltas is determined by the spatial extent and variability of the geomorphic processes that shape them. While in some cases resilient, deltas are increasingly threatened by natural and anthropogenic forces, such as sea level rise and land use change, which can drastically alter the rates and patterns of sediment transport. Quantifying process patterns can improve our predictive understanding of how different zones within delta systems will respond to future change. Available remotely sensed imagery can help, but appropriate tools are needed for pattern extraction and analysis. We present a method for extracting information about the nature and spatial extent of active geomorphic processes across deltas with 10 parameters quantifying the geometry of each of 1239 islands and the channels around them using machine learning. The method consists of a two-step unsupervised machine learning algorithm that clusters islands into spatially continuous zones based on the 10 morphological metrics extracted from remotely sensed imagery. By applying this method to the Ganges–Brahmaputra–Meghna Delta, we find that the system can be divided into six major zones. Classification results show that active fluvial island construction and bar migration processes are limited to relatively narrow zones along the main Ganges River and Brahmaputra and Meghna corridors, whereas zones in the mature upper delta plain with smaller fluvial distributary channels stand out as their own morphometric class. The classification also shows good correspondence with known gradients in the influence of tidal energy with distinct classes for islands in the backwater zone and in the purely tidally controlled region of the delta. Islands at the delta front under the mixed influence of tides, fluvial–estuarine construction, and local wave reworking have their own characteristic shape and channel configuration. The method is not able to distinguish between islands with embankments (polders) and natural islands in the nearby mangrove forest (Sundarbans), suggesting that human modifications have not yet altered the gross geometry of the islands beyond their previous “natural” morphology or that the input data (time, resolution) used in this study are preventing the identification of a human signature. These results demonstrate that machine learning and remotely sensed imagery are useful tools for identifying the spatial patterns of geomorphic processes across delta systems.

show abstract

“…Convolutional neural networks have become increasingly popular in image analysis due to their ability to automatically learn relevant contextual features. Initially devised for natural images, these networks have been revisited and adapted to tackle remote-sensing problems, such as road extraction (Cheng et al, 2017), cloud detection (Chai et al, 2019), crop identification (Ji et al, 2018), river and water body extraction (Chen et al, 2018b;Isikdogan et al, 2018), and urban mapping (Diakogiannis et al, 2019). As such, they seem particularly wellsuited to extract field boundaries but this has yet to be empirically proven.…”

Section: Introductionmentioning

confidence: 99%

Extracting field boundaries from satellite imagery with a convolutional neural network to enable smart farming at scale

Waldner¹,

Diakogiannis²

2020

Preprint

View full text Add to dashboard Cite

Applications of digital agricultural services often require either farmers or their advisers to provide digital records of their field boundaries. Automatic extraction of field boundaries from satellite imagery would reduce the reliance on manual input of these records which is time consuming and error-prone, and would underpin the provision of remote products and services. The lack of current field boundary data sets seems to indicate low uptake of existing methods, presumably because of expensive image preprocessing requirements and local, often arbitrary, tuning. In this paper, we address the problem of field boundary extraction from satellite images as a multi-task semantic segmentation problem. We used ResUNet-a, a deep convolutional neural network with a fully connected UNet backbone that features dilated convolutions and conditioned inference, to assign three labels to each pixel: 1) the probability of belonging to a field; 2) the probability of being part of a boundary; and 3) the distance to the closest boundary. These labels can then be combined to obtain closed field boundaries. Using a single composite image from Sentinel-2, the model was highly accurate in mapping field extent, field boundaries, and, consequently, individual fields. Replacing the monthly composite with a single-date image close to the compositing period only marginally decreased accuracy. We then showed in a series of experiments that our model generalised well across resolutions, sensors, space and time without recalibration. Building consensus by averaging model predictions from at least four images acquired across the season is the key to coping with the temporal variations of accuracy. By minimising image preprocessing requirements and replacing local arbitrary decisions by datadriven ones, our approach is expected to facilitate the extraction of individual crop fields at scale.

show abstract

Learning a River Network Extractor Using an Adaptive Loss Function

Cited by 23 publications

References 16 publications

Dominant process zones in a mixed fluvial-tidal delta are morphologically distinct

Dominant process zones in a mixed fluvial-tidal delta are morphologically distinct

Dominant process zones in a mixed fluvial–tidal delta are morphologically distinct

Extracting field boundaries from satellite imagery with a convolutional neural network to enable smart farming at scale

Contact Info

Product

Resources

About