The Sendai Declaration and the Sendai Framework for Disaster Risk Reduction 2015-2030 recognise that one of the priorities for action is an understanding of disaster risk for prevention and mitigation and for the development and implementation of appropriate preparedness and effective response to disasters, which continue to undermine efforts to achieve sustainable development. Every year fires occur on several hundred million hectares of forests and other vegetation types worldwide. Consequently, a robust Global Fire Danger Rating System is of ultimate importance for the prevention of destructive fires and should be an integral element of land-use policies and fire management strategies. Live Fuel Moisture Content (FMC) is the amount of water in leaves per unit of dry matter, and this influences fire danger and rate of fire spread. Vegetation with high FMC takes longer to ignite and leaf water acts as a heat sink, slowing down fire spread and reducing fire intensity. Fire management can benefit from comprehensive spatial and temporal information on FMC, which can be derived from remote sensing estimates validated against comparatively limited field sampling. This study compares MODIS (or Moderate Resolution Imaging Spectroradiometer), FMC products to measured FMC in validation sites around the globe. Empirical models and models based on Radiative Transfer Modelling (RTM) were used to estimate FMC using MODIS product MCD43A4 collection 6 and associated Quality Flags (MCD43A2). A network of 1181 FMC field sampling sites with a total of 59629 observations in Argentina, Australia, China, France, Italy, Senegal, Spain, South Africa, the republic of Tunisia, the United Kingdom and the United States provided a validation data set for these models. Among the empirical models, Enhanced Vegetation Index (EVI) and Visible Atmospherically Resistant Index (VARI) performed better than others and in many cases outperformed RTM. Predictors of FMC are consistent across field sites for grasslands (R2>0.6, RMSE<52%). Shrublands benefit from relative predictor values to assure they are comparable across sites and reached R2>0.4 and RMSE<25%. The lower number and increased structural complexity of forest sites resulted in poor correlations (R2<0.21, RMSE>30%). A multi-model ensemble is finally proposed to provide a final global product of FMC with superior prediction power.

Area A - Board 134

TASKS:
I– A new Mechanism for decentralized prevention planning and management, through Regionally Integrated Networks of national units with high operational capacity and regular monitoring of changes, challenges and risk reduction;
II– Significant added-value services for EU citizens, nature resources and society;
III- Integrated approach on disaster risk management&disaster risk reduction, including territory/infrastructure/resources management and prevention modeling - performance-results-resilience;
IV- Real-time Earth Observation Big Data applications, based on integrated remote and in-situ data acquisition and interpretation capacity;
Details:
I.1. Core action is the elaboration of methodology for a REFERENCE LAND COVER/LAND USE Dataset, based on the principles of ISO 19144-2. It will provide stable territorial units of management allowing monitoring of the land dynamics using COPERNICUS Sentinels
II.2.The South East Europe(later for other macro regions) Big Data Cloud – involving existing IT infrastructure using a new methodology - “MULTI-DIMENSIONAL NUMBERED DATA BASE”™(MDNDB™) providing powerful and adequate management of big data massifs; Link to EU/EC SDB platforms – COPERNICUS, EMM, EFAS, EFFIS, DRDSI, national; ;
III.1. Regular monitoring of changes, trans-border challenges, critical infrastructure/assets status, fostering the application of regionally integrated COPERNICUS SDB's and in-situ services, support to EU DG’s monitoring mechanism (CAP, ENV, HOME, etc); Possible target – support of CAP monitoring with respect to handling extreme weather events/biodiversity challenges through Regional Integrated Risk&Territory Monitoring Network.
IV.1. - Demonstratioon of a pilot application on two trans-border sites - Bulgarian-Romania and Bulgaria-Turkey. Real-time value-added services, based on integrated remote (ESA&GAF&COPERNICUS support) and in-situ (ASDE&RESAC&RUSE and BOLYAROVO municipalities) Data and Operational Capacity; Application of an integrated approach, based on Earth Observation(EO), including surface and terrain data (DTM/DSM) provided by standardized pan-European datasets and services (LC layer on ISO 19144-2), enhanced with detailed information at local level. Selective monitoring of vulnerable sites - trends, tendencies and emerging risks. Transition from the traditional approach based on static maps to systematic tracking of the area of interest through monitoring, combined with a strengthening of the local operational capacity through transfer of knowledge and responsibilities;
IV.2. Proposal for an IMPORTANT PROJECT OF COMMON EUROPEAN INTEREST (IPCEI) ON TRANS BORDER/TRANS NATIONAL DISASTER RISK MANAGEMENT/DISASTER RISK REDUCTION); New quality of EU collaboration – Direct links between EU programs as COPERNICUS, with regional networks of MS operational capacity, providing Stakeholder&Users based prevention measures

Area A - Board 135

International Civil Agencies operate a number of satellites that are generating petabytes of synthetic aperture radar (SAR) data to observe Earth densely in space and time. SAR observation provides disaster information quickly and widely even if under bad weather condition (rainy and cloudy) as well as night time. However, to maxmize the use of internatinal SAR virtual constellation to disaster quick response, it is very important to have a good coordination across the international community of SAR data providers with respect to efficient data acquisition, planning, science algorithm development, data product standards, framework for science data processing methodologies, data requests and processing resources should, in principle, be possible and would be highly advantageous to all parties. This paper describes the good success story of virtual constellation of SAR under international cooperatoin of multiple space agencies for Japanese disaster moniotoring with lessons learned to future requirements of disaster monitoring by SAR virtual constellation.

Area A - Board 136

PLEASE SEE ATTACHED FILE
----

Research in (semi-)automatic marine target detection using Synthetic Aperture Radar (SAR) data mainly concentrates on the detection and classification of large, metallic targets - mainly ships. This work focuses on small, non-metallic targets, in particular inflatable rubber vessels which are regularly used by migrants in the attempt to cross the Mediterranean from Africa to Europe. In most cases they are 12m long and 3.5m wide and accommodate between 100 and 200 people. 127 people were rescued in 2015 from this particular boat which we used for our data campaign. The physical attributes of such kind of targets, namely its small size and height and the absence of materials of a high dielectric constant such as metals, decrease the detection capabilities of commonly known vessel detection systems. In this work we applied and tested a range of detectors and different methodologies to gain a better understanding of the target’s backscattering properties. The final goal is to identify the best way to process data in the effort of developing a new, specially tailored detection algorithm.

We use multi-platform SAR data from TerraSAR-X, accompanied by Sentinel-1 & Sentinel-2. The data were collected in the Müggelsee, a lake near Berlin, Germany, which functioned as a test bed. A lake scenario was selected because it allows clear identification of the scattering mechanisms of small boats since the background’s radar response is generally lower than for the open ocean. The collection comprises single and dual-polarimetric data and covers a variety of different sensor and geometry parameters. Different combinations of incidence angles, acquisition modes, targets orientation and movement relative to the Line of Sight (LoS) support the effort of identifying the main scattering mechanism and likelihood of its detectability. In the realistic scenario there are 100 to 200 people found on these boats. To examine the exact radar backscatter of fully loaded vessels, we invited an adequate number of volunteers to go out with us on the lake during some data acquisitions.

In this preliminary work we applied the CA-CFAR detector (Figure 3), Polarimetry-based and sub-look-based detectors following a qualitative evaluation of their detection capabilities regarding the special target. Amongst the polarimetric detectors are the intensity Dual-Pol Ratio Anomaly Detector (iDPolRAD) (Marino et al. 2016), the Geometrical Perturbation-Polarimetric Notch Filter (GP-PNF) (Marino, 2013), the Polarimetric Match Filter (PMF) (Novak, 1989) and the polarimetric symmetry and entropy detectors. The sub-look detectors include the sub-look coherency (Marino et. al. 2015), the sub-look entropy (Schneider et.al. 2006) and the sub-look product.

As an additional step, the iDPolRAD was modified in order to focus the detector on surface scattering anomalies instead than volume anomalies. This is useful when the water has a low backscattering (characterised by thermal noise, i.e. volume-like scattering) and the target is mostly represented by surface scattering. We finally considered a combination of the two iDPolRAD that is able to evaluate which detector to use based on the backscattering from the water.

Hence, the novelty of this work lies in two points: a) it is the first time that a broad variety of vessel detectors are benchmarked in detecting this kind of special inflatables target and analysing its backscatter behaviour; and b) the iDPolRAD was modified to detect in addition to volume scattering also the surface scattering anomalies in volume scattering background.


The well-established CA-CFAR is capable of detecting the target in several cases but can be biased by strong sea surface clutter and reduced target radar response. Main factors reducing the target’s signal can be (a combination of) a low incidence angle, the use of cross polarimetric channel and the vessel’s orientation.

The adapted version of iDPolRAD, using the combination of the two scattering mechanisms volume and surface scattering, increases detection capabilities for very small targets. To reduce the rate of false positives of such a detector, a preliminary analysis of the sea surface scattering has to be done. In case of low clutter, the main signal is produced by thermal noise. This state of high Entropy allows for vessel identification if this has a strong surface scattering component. In case of high surface scattering levels, the detector only searches for volume scattering.
The notch filter, employing a geometrical perturbation analysis, shows clear differences in the polarimetric signature between water and the target (Fig. 4). The PMF shows the best results in terms of contrast (Fig. 5). The polarimetric symmetry detector reveilles the target’s asymmetric behaviour (Fig. 6). Very promising results can be seen from the sub-look product algorithm (Fig 7). The entropy detector on the other hand was not working properly because the water had low backscattering and the thermal noise produced very high entropy. A quantitative analysis of the different detectors’ capabilities is still under progress (and it will be ready for the conference) and this abstract only includes first results.

The primary motivation of this research is to mitigate the ongoing humanitarian crisis at Europe's southern Sea border. The applicability of the project’s results to the setting of the open sea, where stronger winds and seas could interfere with radar detection, will be discussed. This project builds a foundation to develop satellite based detection systems for inflatable rubber boats. Such systems could be integral to search and rescue infrastructure in reducing the number of lives lost at sea.





References:

Marino, A. (2013). A notch filter for ship detection with polarimetric SAR data. Selected Topics in Applied Earth Observations and Remote Sensing, IEEE Journal Of, 6(3), 1219–1232.
Marino, A., Sanjuan-Ferrer, M. J., Hajnsek, I., & Ouchi, K. (2015). Ship Detection with Spectral Analysis of Synthetic Aperture Radar: A Comparison of New and Well-Known Algorithms. Remote Sensing, 7(5), 5416–5439. https://doi.org/10.3390/rs70505416
Schneider, R. Z., Papathanassiou, K. P., Hajnsek, I., & Moreira, A. (2006). Polarimetric and interferometric characterization of coherent scatterers in urban areas. IEEE Transactions on Geoscience and Remote Sensing, 44(4), 971–984. https://doi.org/10.1109/TGRS.2005.860950
Marino, A., & Hajnsek, I. (2014). A Change Detector Based on an Optimization With Polarimetric SAR Imagery. IEEE Transactions on Geoscience and Remote Sensing, 52(8), 4781–4798. https://doi.org/10.1109/TGRS.2013.2284510
Marino, A., Dierking, W., & Wesche, C. (2016). A Depolarization Ratio Anomaly Detector to Identify Icebergs in Sea Ice Using Dual-Polarization SAR Images. IEEE Transactions on Geoscience and Remote Sensing, 54(9), 5602–5615. https://doi.org/10.1109/TGRS.2016.2569450
Novak, L. M., Sechtin, M. B., & Burl, M. C. (1989). Algorithms for Optimal Processing of Polarimetric Radar Data, 99.

Area A - Board 137

The Gorkha earthquake (MW 7.8) occurred on 25th April 2015 in the higher Himalayas. The earthquake’s mainshock and several aftershocks triggered numerous huge and small mass movements in this area. The mass movements happened mostly in the hilly and mountainous region which killed more than 9000 persons and destroyed a vast number of manmade infrastructures (Kargel et al. 2016). Several earthquake‐triggered mass movement hazard and risk analyzing studies have been done for this case study area. In addition to use an appropriate hazard and risk analyzing methodology, having access to accurate mass movement inventory plays an important role in the resulted accuracies of analysis. Therefore, there is a growing demand for more reliable mass movement inventory maps for modelling and mapping of this phenomenon. Although, it is a quite challenging task to map and separating the earthquake‐triggered mass movement from old rainfall-induced ones, using SAR data which acquired before and after the earthquake is supportive for this aim.
We generated the surface displacement map by using the differential synthetic aperture radar interferometry (DInSAR) of Sentinel-1 images before and after the earthquake. The derived displacement map from this approach was considered as an important factor of earthquake‐triggered mass movement detection. The high-resolution Planet satellite images were also used in this study. Spectral information of R, G, B, NIR bands and the normalized difference vegetation index (NDVI) were considered as the other factors for detection of this type of mass movement.
Both object-based and pixel-based image analysis have implemented along with different machine-learning models and used in several remote sensing applications. These models have considered effective solutions in image classification and object recognition (Wessel, Brandmeier, and Tiede 2018). Different machine-learning models and classifiers have been used in current studies for mass movement detection. During the past decade, the deep learning models and the convolution neural networks (CNNs) in particular were successfully used for a wide range of applications in computer vision and remote sensing domains. The CNNs have also shown remarkable results in VHR images classification and segmentation, scene annotation and object detection. However, the potential of CNN for mass movement detections have not been fully explored yet. The CNNs are constructed by supervised machine learning, in which lots of labelled data are used to push learnable, i.e. adaptive and feature extractor filters to minimize a loss function (Yang, Lunga, and Yuan 2017). The recent good performance of the CNNs is mostly because of the availability of larger training datasets, optimized network architectures and faster GPUs (Ghorbanzadeh et al. 2018).
In this study, we defined two independent training and testing zones for evaluating the performance of our proposed CNN model. The training process was twofold: firstly, the training included spectral information of R, G, B, NIR and NDVI; secondly, we added to the previous data to the surface displacement map derived from DInSAR. The second trained model illustrated the impact of using SAR data on the earthquake‐triggered mass movement detection. Accuracies were assessed based on three different metrics including precision, to investigate how much detected areas were mass movement triggered by the earthquake; Recall was used to discover how much actual earthquake‐triggered mass movement areas were recognized. The F1 measure was used to find the balance between precision and recall metrics. The results showed that the first model could not separate the mass movements triggered by the earthquake with the old ones. However, the second model detected mostly those which caused by the earthquake.

References:

Ghorbanzadeh, Omid, Dirk Tiede, Zahra Dabiri, Martin Sudmanns, and Stefan Lang. 2018. "DWELLING EXTRACTION IN REFUGEE CAMPS USING CNN-FIRST EXPERIENCES AND LESSONS LEARNT." International Archives of the Photogrammetry, Remote Sensing & Spatial Information Sciences 42 (1).
Kargel, JS, GJ Leonard, Dan H Shugar, UK Haritashya, A Bevington, EJ Fielding, K Fujita, M Geertsema, ES Miles, and J Steiner. 2016. "Geomorphic and geologic controls of geohazards induced by Nepal’s 2015 Gorkha earthquake." Science 351 (6269): aac8353.
Wessel, Mathias, Melanie Brandmeier, and Dirk Tiede. 2018. "Evaluation of Different Machine Learning Algorithms for Scalable Classification of Tree Types and Tree Species Based on Sentinel-2 Data." Remote Sensing 10 (9): 1419.
Yang, Hsiuhan Lexie, Dalton Lunga, and Jiangye Yuan. 2017. "Toward country scale building detection with convolutional neural network using aerial images." Geoscience and Remote Sensing Symposium (IGARSS), 2017 IEEE International.

Area A - Board 138