Remote Sensing

A deep transfer learning-based damage assessment on post-event very high-resolution orthophotos

Abstract

Post-disaster building damage assessment is an important application of remote sensing. The increasing resolution of remote sensing imaging systems and state-of-the-art deep learning networks has facilitated damage assessment. However, most existing methods in the literature concentrate on damage/non-damage classification only in specific disaster types/areas using pre- and post-event images. Furthermore, site visits are inevitable to assess the level of damage to structures. Therefore, the main objective of this study was to utilize deep transfer learning over a pre-trained network and extend it to a damage assessment framework. The network is fine-tuned to identify four different damage levels: non-damage, minor damage, major damage, and collapsed, using only post-event images taken from different disaster types/areas. To evaluate the proposed framework, we carried out three experiments on Hurricane Irma in Sint Maarten, Hurricane Dorian in Abaco Islands, and Woolsey Fire using post-event orthophotos derived from unmanned aerial vehicle (UAV) images. The results of over 80% overall accuracy confirm that with a structured learning scenario, it is possible to use transfer learning on very high-resolution remote sensing images to classify the level of structural damage.

Full document can be requested here: https://cdnsciencepub.com/doi/10.1139/geomat-2021-0014

Date
2022
Data type
Scientific article
Theme
Research and monitoring
Journal
Geographic location
St. Maarten

High spatial resolution mapping identifies habitat characteristics of the invasive vine Antigonon leptopus on St. Eustatius (Lesser Antilles)

On the Caribbean island of St. Eustatius, Coralita (Antigonon leptopus) is an aggressive invasive vine posing major biodiversity conservation concerns. The generation of distribution maps can address these conservation concerns by helping to elucidate the drivers of invasion. We test the use of support vector machines to map the distribution of Coralita on St. Eustatius at high spatial resolution and use this map to identify potential landscape and geomorphological factors associated with Coralita presence. This latter step was performed by comparing the actual distribution of Coralita patches to a random distribution of patches. To train the support vector machine algorithm, we used three vegetation indices and seven texture metrics derived from a 2014 WorldView-2 image. The resulting map shows that Coralita covered 3.18% of the island in 2014, corresponding to an area of 64 ha. The mapped distribution was highly accurate, with 93.2% overall accuracy (Coralita class producer's accuracy: 76.4%, user's accuracy: 86.2%). Using this classification map, we found that Coralita is not randomly distributed across the landscape, occurring significantly closer to roads and drainage channels, in areas with higher accumulated moisture, and on flatter slopes. Coralita was found more often than expected in grasslands, disturbed forest, and urban areas but was relatively rare in natural forest. These results highlight the ability of high spatial resolution data from sensors such as WorldView-2 to produce accurate invasive species, providing valuable information for predicting current and future spread risks and for early detection and removal plans.

 

Referenced in BioNews publication (BioNews Article). 

 

Related Resources:

1. Supplementary Infromation (Report)

2. Topographic Wetness Index raster layer for Statia developed for use in the Coralita mapping publication (Raster Layers). 

3. Raster layers: High spatial resolution mapping identifies habitat characteristics of the invasive vine Antigonon leptopus on St. Eustatius (Lesser Antilles) (Raster Layers).

Date
2021
Data type
Scientific article
Theme
Research and monitoring
Journal
Geographic location
St. Eustatius

High-resolution prediction of plant species richness in the Christoffel national park

Previous attempts at mapping the vegetation of the Christoffel national park on the island of Curaçao were done in times of intense grazing pressure and are likely not valid anymore after the removal of goats from the park because grazers have a significant effect on the native vegetation of the island ecosystems. In 2018, a 2-year fieldwork campaign was started to revisit the sampling points of Bokkestijn & Slijkhuis (1987) with the aim of remapping the vegetation communities and studying the change that occurred in the last decades. This thesis aims to assess the changes in vegetation distribution and use the newly acquired data to predict plant species richness across the entire national park at a high resolution using a macroecological modeling strategy. A trend of secondary vegetation succession has been found since 1985, with an increase in the coverage of trees, orchids, and bromeliads and a decrease in grasses and herbs. The large-scale recovery of the native vegetation is found especially on the coast and midland of the park, while the Christoffel mountain and its surroundings have remained relatively stable. An aerial photograph interpretation of the vegetation communities found significant dependence of vegetation communities on elevation and slope aspects. High-resolution plant species richness prediction models were built and it was found that elevation and slope aspects have the most predictive weight. Little research has been done on high-resolution species richness prediction models; however, it is shown that these models can be utilized to characterize the variables influencing species distribution at high resolution and local scale, with comparable accuracy to coarser prediction models.

Date
2020
Data type
Research report
Theme
Research and monitoring
Report number
Thesis report
Geographic location
Curacao
Image

Remote Sensing Tools to support NEXUS challenges

Smalls islands are especially vulnerable to climate change and land  use changes due to the competing needs for limited resources. To support the NEXUS approach we need evidence based monitoring tools that can provide policy makers, conservation managers, entrepeneurs, scientists and the general public with information on the state, pressures and associated changes in the environment. Satellite imagery can provide synoptic information at appropriate
spatial and temporal resolutions that can support evidence based monitoring. Only at very detailed levels information might be added by using airplanes or drones. Remotely sensed information can help to provide information on e.g. land cover and associated dynamics such as urban sprawl, mapping habitats such as mangroves and coral reefs, surveying terrain conditions such as soil moisture conditions and erosion hazards associated within catchments, sea level rise and changing coastlines, and on many aspects of the vegetation (natural and agriculture), such as plant traits, phenology and plant growth. Remotely sensed information can in general make field surveys and monitoring more effective, and can thoroughly support decision making.

Date
2019
Data type
Research report
Theme
Education and outreach
Research and monitoring
Geographic location
Bonaire

Classifying benthic habitats and deriving bathymetry at the Caribbean Netherlands using multispectral Imagery

Benthic habitats (habitats occurring at the bottom of a water body) and coral reef ecosystems provide many functions. Currently, however, coral reefs are threatened by a number of factors and degrade rapidly. Benthic maps are important for management, research and planning. Coral communities in the Caribbean Dutch island of St. Eustatius are generally in a good condition, but the benthic communities around St. Eustatius have not been yet accurately mapped.
Remote sensing imagery has been found to be a very useful tool in providing timely and up-to-date information for benthic mapping and offers an effective approach to complement the limitation of field sampling. Remote sensing in water, however, presents challenges mainly due to the complex physical interactions of absorption and scattering between water and light. Shorter wavelengths (-450 nm) penetrate deepest into the water column and longer wavelengths (-500-750 nm) are more rapidly absorbed and scattered. Therefore, the potential extent of use of remote sense imagery in the oceans relies more on shorter wavelengths (blue band), which have inherently noisier signals due to atmospheric effects.
This research explores the utility of multispectral imagery to identify and classify marine benthic habitats in the Dutch Caribbean island of St Eustatius. These include the comparison of two sensors with different spatial and spectral resolution, QuickBird (2.4m, 4 bands) and WorldView-2 (2.0m, 8 bands) for mapping benthic habitats. The study first investigates the existing methodologies for benthic habitat classification. The benefits of atmospheric correction, sun glint effect correction and water column attenuation correction on the accuracy of classification maps are also assessed. Then, an object and pixel based supervised classifications for the characterization of sea grass, sand and coral are performed. This research also evaluates the possibility to extract water depth from multispectral satellite imagery by the use of a ratio transform method. Bathymetric data is important for water column correction, to improve the classification accuracy and for the study of the ecology of the habitats.
Results showed that the best results for pixel-based image classification in QuickBird and WoldView-2 imagery were obtained after deglinting the image, with accuracies of 49.3% and 51.9% respectively. The sunglint removal method improved the total accuracy of benthic habitat mapping, by increasing before and after deglinting 3.4% for QuickBird and 6.3% for WorldView-2. Object-based classification provided slightly better classification results, with a 53.7% accuracy for QuickBird and 56.9% accuracy for WorldView-2. Therefore, it can be concluded that an object-oriented approach to image classification shows potential for improving benthic mapping. The classification accuracy did not increase after compensation for water column effects.
The effectiveness of the ratio method to calculate the bathymetry using multispectral imagery has been confirmed. The coefficients of determination (r2) achieved are statistically significant, 0.66 for QuickBird, and 0.41 for WorldView-2 (BG ratio) for a linear relation. The root mean square errors are 4.02 m for QuickBird and 5.11 m for WorldView-2. It has been proved that this method works better for shallow areas, with a root mean square error of 2.32 m and 2.47 m, respectively. Results also indicate that the ratio method is sensitive to variable bottom type. Overall, better bathymetric values were obtained with QuickBird than with WorldView-2.
This research provides a baseline for future benthic habitat classification of the Dutch Caribbean islands using remote sensing. The results of this study are a good example of how remote sensing data can be a useful and cost effective method to map benthic habitats and calculate bathymetry.

Date
2013
Data type
Research report
Theme
Research and monitoring
Report number
GIRS-2013 -18
Geographic location
St. Eustatius
Author

Classifying benthic habitats and deriving bathymetry at the Caribbean Netherlands using multispectral Imagery

Benthic habitats (habitats occurring at the bottom of a water body) and coral reef ecosystems provide many functions. Currently, however, worldwide coral reefs are threatened by a number of factors and are degrading rapidly. Benthic maps are important for management, research and planning, but the benthic communities around St. Eustatius have not yet been accurately mapped or described.
Remote sensing imagery has been found to be a useful tool in providing timely and up-to-date information for benthic mapping and offers an approach that may complement the limitations of field sampling. Remote sensing in water, however, presents challenges mainly due to the complex physical interactions of absorption and scattering between water and light. Shorter wavelengths (-450 nm) penetrate deepest into the water column and longer wavelengths (-500-750 nm) are more rapidly absorbed and scattered. Therefore, the potential extent of use of remote sense imagery in the oceans relies more on shorter wavelengths (blue band), which have inherently noisier signals due to atmospheric effects.
This research explores the utility of multispectral imagery to identify and classify marine benthic habitats in the Dutch Caribbean island of St Eustatius. These include the comparison of two sensors with different spatial and spectral resolution, QuickBird (2.4m, 4 bands) and WorldView-2 (2.0m, 8 bands) for mapping benthic habitats. The study first investigates the existing methodologies for benthic habitat classification. The benefits of atmospheric correction, corrections for sun-glint effect and water column attenuation on the accuracy of classification maps are also assessed. Then, an object and pixel-based supervised classifications for the characterization of sea grass, sand and coral are performed. This research also evaluates the possibility to extract water depth from multispectral satellite imagery by the use of a ratio transform method. Bathymetric data is important for water column correction, to improve the classification accuracy and for the study of the ecology of the habitats.

Date
2013
Data type
Research report
Theme
Research and monitoring
Report number
IMARES rapport C143/13; Alterra rapport 2467
Geographic location
St. Eustatius