Open Innovation and Tools

The use of Unmanned Aerial Vehicles (UAV) is becoming a standard approach in photogrammetry, supported by a growing number of users and organizations. This is due to the relatively affordable costs for these technologies as well as to their readiness in obtaining high quality products such as detailed 3D models and orthophotos.

Nevertheless, the comprehensive price of both photogrammetric UAVs as well as proprietary software -currently adopted in the practice- often exceeds ten thousand Euros, which may still represent an obstacle to the spreading of this technology among SMEs as well as developing countries. With this in mind, we present here a framework aimed to the generation of high resolution Digital Surface Models (DSMs) by coupling low-cost UAV with Free and Open Source Software (FOSS). This framework takes advantage of the increasing availability on the market of low-cost UAVs -which anyway incorporate many kinds of reliable sensors- as well as of the recent development of powerful FOSS tools dedicated to photogrammetry and image processing.

For this study, a test area of 150 square meters -characterized by flat terrain with concrete curbs- was selected. Aerial images of the area were captured by using the 14 Mpx on board digital camera of a DJI Phantom 2 Vision+ quadcopter. Both flight plan as well as quadcopter navigation were performed by exploiting its GNSS embedded sensor through the dedicated application for mobile devices. Image pre-processing was performed using GIMP (https://www.gimp.org) while dense point-clouds were extracted from these images and georeferenced using OpenDroneMap (https://github.com/OpenDroneMap). CloudCompare (http://www.danielgm.net/cc) was then used for point-cloud post-processing and raster DSMs generation. In order to asses the quality of the UAV derived DSM, a reference DSM of the same area was created starting from 1400 points, collected by GPS survey - Real Time Kinematic (RTK) method. A statistical comparison of the two resulting raster DSM was then performed by using GRASS GIS (https://grass.osgeo.org).

The presented approach allows to obtain accurate photogrammetric products while being economically advantageous. Moreover, thanks to the continuous quality increase of both low-cost UAV hardware and FOSS tools it is candidate to further improvements in the next future.

Web portals for visualizing and exploring Earth Observation (EO) data are becoming increasingly common. Typically these portals are built upon the principles and standards of Geographic Information Systems (GIS) with tools and formats that arise from them (e.g. OpenLayers, Leaflet, KML, GeoJSON and Web Map Services). Several limitations are often found when building applications based on complex, high-resolution EO data. Such limitations include: 

A focus on loading and displaying images, rather than the “real” source data. This means that any data processing has to be done on the server, introducing complications into the architecture.

A focus on static 2D maps. EO scientists often wish to explore data in multiple dimensions (e.g. animations, time-series, vertical sections, transects and 3D renderings).

A focus on the “web Mercator” map projection. This is unsuitable for global or high-latitude data.

Information loss due to the use of simplified formats, often necessitating the use of ad-hoc workarounds that are not interoperable between portals.

 ‘Typical’ web developers may have difficulty understanding the more complex formats used in the EO community (e.g. GeoTIFF, HDF, NetCDF).

In this presentation, we will discuss two new developments that aim to address the above limitations. They take advantage of modern web browsers’ support for HTML5, which provides powerful visualization and processing capabilities.

The Web WorldWind platform (https://webworldwind.org) is a web-based virtual globe application programming interface (API) for HTML5-capable browsers, developed by NASA in partnership with ESA. It is able to display interactive globes and projected maps with accurate 3D terrain and many kinds of data layers.

The CoverageJSON format (http://tinyurl.com/covjson) is a new JSON-based format that accurately encodes many kinds of scientific data, including multidimensional grids, polygon-based representations, time-series, vertical profiles and more. It is based on the concept of a “coverage” (defined in ISO19123), which is a general data structure that maps positions in space and time to data values. CoverageJSON can be thought of as an equivalent of GeoJSON, but facilitating the exchange of scientific data (including EO data), rather than simple GIS features. The goals of CoverageJSON are to: 

encode environmental data in a readable and efficient way, using uniform concepts and structures across a variety of data types;

be usable by ‘typical’ web developers who do not necessarily have a background in EO science;

encode semantic information (through the use of URI links), enabling data to be more machine-readable and understandable. This includes encoding of uncertainty information;

be an open standard encoding, and an output format for services such as the Web Coverage Service.

When used together, Web WorldWind and CoverageJSON form a flexible and powerful platform for developing exciting, engaging and informative web portals based on multidimensional Earth Observation data. In this presentation we will demonstrate Web WorldWind and CoverageJSON in greater detail, providing several use cases. In particular, we will demonstrate the efficiency of these tools, both in terms of fast data transfer and smooth data interaction, even when tested with high-resolution data of many millions of points. 

Nansat is an open source, scientist friendly Python toolbox for processing 2D satellite EO raster data (https://github.com/nansencenter/nansat). Nansat provides high flexibility in a command line interface to facilitate easy development and testing of scientific algorithms, easy analysis of geospatial data, and efficient operational processing. To enable the use of several data sources for colocation, comparison and validation, Nansat takes benefit of the already existing open source GDAL C++ library. This is used to read geospatial data from a number of different sources and to perform basic operations on the data. A core feature of the system is that data is read from file only when needed, and it is possible to extract subsets of the full dataset. This is more efficient than reading the full dataset into memory as is necessary with, e.g., Matlab and other similar software, in particular if it is stored on a remote server and accessed via, e.g., OPeNDAP.

Nansat contains a package called mappers, which adds metadata to complement the retrieved raster data. Presently, 40 datasets from different satellite instruments and numerical models can be opened with the mappers package, in addition to any gridded datasets following the NetCDFCF standard.

A set of functions to perform common operations on the data, like georeferencing and reprojection, averaging, transect extraction and analysis, as well as visualization and generation of high quality maps, is available. In addition to this, Nansat contains a set of common export functions.

We will present the Nansat software including new and planned features.

Sentinel 1A is an European radar imaging satellite which supplies images in all light and weather conditions. It tracks many aspects of our environment, including detecting and tracking oil spills and flood areas, monitoring movement in land surfaces, and mapping changes in the land surface.            

 GeoCodis and Space SI are working together to provide services for receiving up-to-date satellite images, for the detection and analysis of water surfaces and flood areas, and to provide the right information for users.

 At present we are able to provide information about water coverage for all European countries and a significant area of Africa with a frequency of 3-5 days. The ground resolution of the data is 10 metres.

Raw Sentinel 1A Modelling

The radar can image the Earth’s surface through rain and cloud, regardless of whether it is day or night. This is particularly useful for providing imagery during extreme weather conditions. The sensor on the Sentinel-1 has a swath width of 250 km and a ground resolution of 5 x 20 m. The images are of very high quality, but are difficult to interpret without knowledge of the interaction of the radar signal with different surfaces.

Water Bodies Detection

Detection of bodies of water is a process based on machine learning and region growing image segmentation methods. Misclassifications are removed by filtering the surface area, overlap with SAR shadow mask, and an average surface slope. The best available elevation model is used during neighbourhood operations and temporal filtering to determine the presence of water at each pixel. 

Customised Services

Different professionals can derive a big advantage by using our services for their specific needs. Our system can provide online access to a last water body measurement or to an overview of a historical archive. Users can be also informed when the newest images will appear on the system. The most frequently used services are cumulative coverage for a requested period of time, and reports of water presence inside a requested area.                                                                           

The next years will seen an unprecedented explosion in environmental data produced by an increasing fleet of Earth observation (EO) satellites. The purpose of this expensive investment is to deliver global time series of observations, which reflect important environmental processes related to climate, agriculture and ecology. This constant data flow is essential to develop and eventually use the capabilities to sustainably manage a planet with 9 billion humans far away from its natural state.

The continuous stream of complex environmental remote sensing data is far too large for useful interpretation unless it is integrated in a sophisticated computational environment of models, which simulate the observed processes on the Earth’s surface. They create a comprehensible virtual world and virtual observations to compare the stream of real observations to. Together the observed and computed environment will eventually form a cyber-environmental system which will deliver local, regional and global information essential for decision making towards sustainable environmental management.

The presentation will introduce current environmental remote sensing data streams and land surface process models. It will derive the informational and computational requirements of local to global sustainable environmental management and present strategies to form a cyber-environmental system with special emphasis on global smart agriculture to sustainably secure global food supply.