OR/15/061 EO information services assessment

Grebby, S, Jordan, C J, and Dijkstra, T. 2015. Risk information services for Disaster Risk Management (DRM) in the Caribbean - Service utility document. British Geological Survey Technical Note, OR/15/061.

File:OR15061 topbanner.jpg

Methods

The EO service/information products were subjected to a three-stage assessment which incorporates the feedback received throughout from the WB TTL and the CHARIM project, in addition to that of the local users obtained during the initial contact at the CHARIM workshop in the Caribbean (September 2014) and the subsequent training workshop in the Netherlands (March 2015). As outlined above, one of the main phases of the assessment took place during the workshop at ITC, where the EO products were evaluated in some detail by the local users. As part of the evaluation, the users were provided with a form to capture their feedback (see Appendix B: EO product assessment feedback form). The form was designed to gain feedback relating to the four assessment attributes above, with a particular emphasis on identifying the benefits of each of the EO Services and information products over current practices, their perceived limitations and any additional information products that may be of particular use in the future. The main outcomes of the assessment are summarised below.

Service utility

Service 1: Land use/land cover mapping

Overall, the assessment for the Service 1 land use/land cover information products resulted in positive feedback, with all users agreeing that the new products provide better information than the existing land use/land cover maps. Specifically, the new products were perceived to provide much more detail than the existing maps in terms of spatial resolution, while also providing an updated representation of the current land use/land cover:

• “The new land use/land cover maps are more precise and accurate where the old one is general” — Grenada.
• “The map quality is an improvement to the previous one (higher resolution)… more classes of land cover identified” — St. Vincent and the Grenadines.
• “Generally seems to be good and properly represented… there is no new information, but it shows an updated land cover map” — St. Lucia.

With some minor refinement, the land use/land cover maps were used to successfully support the landslide and flood risk mapping within the CHARIM project (e.g. https://www.charim.net/sites/default/files/handbook/maps/SAINT_LUCIA/FINAL_LANDSLI DE_SAINT_LUCIA.pdf). Moreover, many of the local users recognised that the timely delivery of information through the Service could enable them to undertake new work that was not previously possible based on current practices and information:

• “The land cover maps for Grenada, St. Lucia and St. Vincent are very important datasets for us, and they will certainly contribute a lot to improving the quality of the landslide and flood maps for these islands” — CHARIM project team.
• “This will result in on time delivery of maps which can assist in decision making and planning. When a map is done using other methods it sometimes takes too long to materialise and sometimes the land cover may have changed before even producing a map” — St. Lucia.
• “Can help with development decision making and help in preserving of watersheds” — Grenada.
• “The map can be useful in monitoring changes in forest cover… gives a good idea as to where development is taking place” — St. Vincent and the Grenadines.

The majority of users stated that more detailed information on the land use/land cover types would be beneficial:

• “Some of the types in the legend should be separated” — Grenada.
• “It will be nice if the cultivated land class could be further split into other uses” — St. Vincent and the Grenadines.
• “Classes for new land use/land cover maps need to be separated to be used for informing decisions for watershed management” — Grenada.

The land use/land cover classes used in the mapping were selected to ensure full compatibility with the existing maps. In the future these classes could be easily refined with input from the users to further enhance the information content of the maps.

Service 2: Hazard mapping to support landslide risk assessment

The landslide inventories delivered by Service 2 were primarily of interest to the CHARIM project because these were essential inputs to the landslide risk mapping. The St. Lucia inventory delivered by this Service was particularly useful because it provided the most comprehensive and up-to-date representation of recent landslide activity:

• “The landslide inventory map for St. Lucia is a very important dataset for us, and it will certainly contribute a lot to improving the quality of the landslide map for the island” — CHARIM project team.

Relatively little feedback on the national-scale DEMs was received from the local users, possibly due to unfamiliarity in working with elevation data or the similarity in spatial resolution between the delivered DEMs and those that already exist. Nevertheless, one local user expressed some degree of satisfaction with the DEMs, but suggested that a higher resolution product would be more desirable:

• “The DEM was more or less OK. A higher resolution DEM will be appreciated though” — Grenada.

Although the DEMs delivered by the Service met the specifications outlined in the SOW, the intention was to produce 10 m DEMs for St. Lucia and Grenada. Unfortunately, this was not possible due to the inability to acquire suitable cloud-free imagery during the project timeframe because of the hurricane season. Nevertheless, this Service has successfully demonstrated that it is possible to produce DEMs from stereo optical satellite imagery, irrespective of the resolution.

Service 3: Digital Elevation Model

Based on the feedback received, the 20 m DEM delivered for part of Belize under this Service was deemed to provide information that is significantly better than the existing data, in terms of both spatial resolution and accuracy:

• “The 20 m map is better than we currently have, which is a 30 m” — Belize.

The precise 1 m DEM demonstration was considered to be the most useful by the users because it would enable much more accurate flood risk mapping than currently possible using the existing 30 m DEM:

• “The maps are useful, especially the 1 m, this is really what we need” — Belize.

Service availability

Service 1: Land use/land cover mapping

The land use/land cover maps for the islands were generated using a combination of automated image classification and visual interpretation. As a result, each map was produced in approximately three days, which is significantly quicker than possible through conventional field-based mapping. The ability of the Service to produce land use/land cover maps in a timely manner and the benefits offered by this were recognised by local users:

• “I believe that the land cover maps generated through satellite takes so little time as doing it on site. This will result in on-time delivery of maps which can assist in decision making and planning” — St. Lucia.

The land use/land cover classes are based on those in the existing maps, ensuring that compatibility is maintained. The methodology used the generate the maps is predominantly objective, meaning that it can be employed to produce consistent land use/land cover maps both in to the future, as well as from archived satellite imagery. The ability to produce consistent land use/land cover maps makes this Service particularly advantageous to the local users going forward:

• “Images on a 2-year interval from 2000 to future years would be appreciated” — St. Lucia.

The maps produced under this Service provide an order of magnitude increase in spatial resolution in comparison to the existing 30 m land use/land cover maps. As highlighted above, all users commented that the new 2 m maps would enable them to undertake new work that was not previously possible using the existing data. Nevertheless, one local user did believe that higher resolution information would be more advantageous for monitoring purposes:

• “A 2 m resolution is good, but for on the ground monitoring a 1 m or 50 cm resolution would be better” — St. Lucia.

The maps were produced at a 2 m resolution because this is the native resolution of the multispectral bands on which land use/land cover discrimination is dependent. Nevertheless, it is possible to perform pan-sharpening using the panchromatic band to increase the spatial resolution of the multispectral imagery to 50 cm. However, this could potentially pose technical challenges for processing the imagery for relatively large areas because the file size will increase dramatically.

Service 2: Hazard mapping to support landslide risk assessment

The landslide inventories were initially produced from the very-high resolution RapidEye and Pleaides satellite imagery before being validated in the field. This combined approach to mapping was used to generate comprehensive landslide inventories in a more time-efficient manner than possible through field mapping alone. For St. Lucia, the multi-temporal approach employed also demonstrates the ability to produce consistent landslide information both historical imagery and imagery acquired in the future. In this particular case, the multi-temporal mapping approach made it possible to monitor the activity of individual landslides throughout the 5-year period. This approach can also be used to continue to monitor landslide activity into the future.

Only one landslide was identified for Grenada during the 4-year period covered by the satellite imagery. This lack of information on previous landslide activity therefore made it difficult for the CHARIM project to model the future hazard. However, this was not an issue with the mapping approach employed in this Service, but rather the restriction on the time period for the mapping that was defined in the SOW (i.e., using imagery no older than 2010). To gain a better understanding of the previous landslide activity, the CHARIM team acquired archived imagery that pre-dated 2010 and used the same methodology to derive a landslide inventory.

The DEMs for St. Lucia and Grenada appeared to be of lower importance to the users. Nonetheless, one local user did express an interest in having access to a higher resolution DEM for Grenada. As discussed above (see Service 2: Hazard mapping to support landslide risk assessment), the acquisition of fresh stereo imagery was tasked, which would have enabled the generation of 1 m DEMs for St. Lucia and Grenada. However, all attempts to acquire suitable imagery within the timeframe of the project were hampered by the persistent cloud cover during the Atlantic hurricane season. Despite this, the Service demonstrated the ability to generate DEMs from optical stereo satellite imagery for the entire area of interest in a timely manner. Moreover, the largely automated nature of the processing means that consistent elevation information can be generated from both archived and future stereo imagery, if required.

Service 3: Digital Elevation Model

This Service demonstrates the ability to generate DEMs for Belize from stereo and tri-stereo imagery a several types of satellite imagery with different spatial resolutions (30 m, 20 m, and 1 m). Again, owing to the largely automated technique used to generate the DEMs, consistent elevation data can be rapidly derived from both future and historical imagery. As in Service 2, the main constraint affecting the ability to produce the DEMs in this Service is the successful acquisition of stereo or tri-stereo imagery with acceptable levels of cloud cover.

The 1 m and 20 m DEMs were both deemed to represent a significant enhancement over the existing 30 m DEM. Although the 1 m DEM demonstration met the 100 km² coverage specified in the SOW, the local users thought it was preferable to have national-scale coverage at this level of detail and quality. Whilst this is possible, it would require additional imagery and more processing time. Furthermore, one of the local users considered bathymetric data for the wetland areas to also be important for assessing flood risk:

• “The limitation is the ability to effectively resolve the areas covered by water; most of the areas of work are in such areas” — Belize.

Although not within the remit of this project, it is technically possible to extract bathymetric data from high-resolution optical satellite imagery in relatively shallow water. This is one aspect of the Service that could potentially be upgraded in the future.

Service reliability

Service 1: Land use/land cover mapping

The land use/land cover maps were validated using a combination conventional confusion matrices and limited field verification. Information about this validation procedure was provided to the users. This was sufficient information for most users to accept the EO products with a high level of confidence, although others felt that it would be necessary for them to conduct their own ground-truthing before they could incorporate the products into their work activities:

• “I believe that the BGS has the ability to provide maps for the region in a timely manner and is also accurate” — St. Lucia.
• “It makes a great base map for further ground work/field verification” — St. Vincent and the Grenadines.
• “Some more ground work can be done to verify information on the ground” — Grenada.
• “Ground truthing to verify represented information on the maps would allow for monitoring of development and growth” — St. Lucia.
• “The accuracy will help in certain developments” — Grenada.

The main issue associated with this Service was the incompleteness of the rivers and roads layers across the entire area of interest:

• “For some instances the roads and rivers seem to disappear” — St. Lucia.
• “Rivers and streams in some areas appear to be incomplete” — Grenada.

This is due to a combination of cloud cover in the imagery and obscuring by dense vegetation. Although the issue with cloud cover can be overcome by acquiring imagery at a different time of year, the problem with vegetation obscuring roads and rivers is a problem that persists in tropical settings. The use of airborne LiDAR to generate ‘bare-earth’ to create digital terrain models is the only effective way of overcoming the obscuring effects of vegetation. Furthermore, existing road/river network maps could be used as a baseline, and updated using satellite imagery. This is an exercise that the Planning/Surveying Ministries could undertake.

Some minor land use/land cover misclassifications were noted by the CHARIM team. This specifically related to misclassification between some vegetation types, and between some buildings, bare ground and roads. This is due to similarities in their spectral characteristics in the imagery at visible to near-infrared wavelengths detected by the sensors. The target thematic accuracies specified in the SOW were achieved, but such classification confusion can be reduced through manual refinement or the use of satellite imagery with more spectral wavebands. Although for current satellite sensor technology, it should be noted that imagery with additional spectral information generally has a lower spatial resolution.

Service 2: Hazard mapping to support landslide risk assessment

The landslide inventories initially produced from very-high resolution satellite imagery were subsequently validated on the ground and updated appropriately. The vast majority of the mapped landslides visited in the field were verified, thus providing an extra level of confidence in the EO products. The use of multi-temporal imagery ensured full spatial and temporal coverage for the entire areas of interest, with the remote sensing approach enabling mapping of areas that were inaccessible on the ground (Jordan et al., 2015). No issues regarding the landslide inventories for St. Lucia were raised, and these were used successfully by the CHARIM team for hazard mapping. The only issue for Grenada was the lack of landslide activity during the time period specified in the SOW. With hindsight, the specification needed adjusting to incorporate imagery that was acquired prior to 2010, when a climate event caused significant landsliding on the island.

Little feedback on the reliability of the DEMs was received, but discussions with the local users revealed that they were generally satisfied with the quality DEMs and the validation procedure. The only specific comment was that higher resolution DEMs would be preferable.

Service 3: Digital Elevation Model

Due to a lack of accurate elevation data (e.g., GPS), the DEMs for Belize were subjected to a preliminary validation procedure. Understandably, this has implications for the level of user confidence in the EO products:

• “The maps are useful…but verification needs to be done first” — Belize.

Nonetheless, the potential of this approach to generating DEMs that are comparable to other state-of-the-art techniques in terms of accuracy and detailed is clearly recognised:

• “We have been asking for LiDAR, but for some projects these maps that you produce can substitute for the accuracy needed” — Belize.

Service affordability

Service 1: Land use/land cover mapping

The existing land use/land cover maps for the countries were produced from freely-available Landsat satellite imagery acquired ca.2000. Although free, the data have only a moderate spatial resolution of 30 m. In contrast, the new land cover maps were predominantly generated from very-high resolution Pleaides satellite imagery, which provides an order of magnitude increase in the spatial resolution (2 m) of the resulting maps. Access to the Pleiades satellite imagery was provided through the ESA TPM scheme, but commercially such imagery typically costs upwards of €17 per km². Taking into account the overwhelmingly positive feedback from users, it is apparent that the enhanced information content, quality and subsequent benefits of the new maps fully justify the costs. Moreover, land use/land cover mapping using EO data is generally more time and cost efficient than conventional field-based mapping. This was recognised by one local user:

• “EO products will enable me to reduce the overall expenditure of data collection” — Grenada.

Service 2: Hazard mapping to support landslide risk assessment

Landslide inventory mapping was based largely on the visual interpretation of multi- temporal Pleaides and RapidEye satellite imagery. This imagery enabled the extent of landslides to be mapped in detail and their failure mechanisms to be determined across the entirety of the St. Lucia and Grenada areas of interest. In this project the imagery was obtained free-of-charge through the ESA TPM scheme, but the commercial cost of such imagery is typically €17 per km² and €0.95 per km², respectively. The cost benefit of the Service is fully justified just on the basis that it can be utilised to produce national-scale landslide inventory maps more efficiently than possible using conventional field mapping techniques. Moreover, previous attempts to produce landslide inventories have focussed only on field mapping in areas accessible from the roadside, whereas this EO-based Service provides a means of overcoming accessibility issue to generate more comprehensive inventories. Furthermore, EO data provides the user with the opportunity to undertake historical mapping from archived data and extend this into the future for long term monitoring activities. It should also be noted that the same Pleiades data used for land use/land cover mapping in Service 1 was also used to generate the landslide inventories. The ability to utilise the imagery for multiple applications only acts to further justify the costs.

The national-scale DEMs for St. Lucia and Grenada were generated using ASTER stereo satellite imagery with a spatial resolution of 30 m and vertical accuracies (RMS) better than 5 m. Although elevation data with a higher spatial resolution and accuracy can be acquired through traditional ground-based GPS surveys, this approach is extremely inefficient when covering large areas. Airborne LiDAR can provide accurate and high resolution elevation data, although surveys have to be commissioned and can therefore be very costly. At a cost of approximately $60US for a scene covering 3600 km², ASTER satellite imagery proves a cost effective means of generating a DEM for applications such as national-scale landslide and flood hazard mapping. If temporal coverage of the DEM is not an issue, then the 30 m ASTER Global DEM, generated using imagery acquired in 2000–2011, is available free of charge. As was initially planned for this project, optical stereo imagery acquired by the Pleiades satellites can be used to generate DEMs with a much higher spatial resolution (ca.1 m) and vertical accuracy. Such imagery can be acquired for a cost in the region of €29 per km², and would enable the landslide and flood risk to be mapped more accurately.

Service 3: Digital Elevation Model

Although offering a slightly enhanced spatial resolution and vertical accuracy over the 30 m ASTER-derived DEM for Belize, the 20 m DEM generated from SPOT satellite imagery is more expensive at a cost in the region of €2 per km². Accordingly, in the absence of a DEM produced through traditional means, the use of stereo satellite imagery is the most cost effective method of producing a DEM of Belize for national-scale flood risk mapping purposes.

The precise 1 m DEM for 100 km² of Belize provides a demonstration of the full potential in using optical satellite to generate accurate and high-resolution DEMs. In this case, the DEM was generated using state-of-the-art Pleiades tri-stereo satellite imagery. Again, the imagery was acquired free of charge through the ESA TPM scheme. Commercially, this type of imagery costs in the region of €50 per km². Whilst this is quite costly, the feedback from the users suggested that a DEM of this quality and accuracy is urgently required for accurate flood risk mapping in Belize because it far exceeds that of the existing DEM. Based on this alone, the cost associated with this Service appears fully justified with respect to the potential benefits and opportunities it can bring. The 1 m DEM delivered by the Service is comparable to the resolution and accuracy achievable using airborne LiDAR. In generally, the generation of DEMs from tri-stereo optical imagery is probably more cost effective than airborne LiDAR for mapping areas up to 200 km².

Overall appraisal

Overall, the results of the assessment provide clear evidence of acceptance of the Service and EO information products by the users. The ability to utilise EO data to produce land use/land cover maps that are considerably more detailed than the existing data was greatly appreciated by all users:

• “The use of satellite data for generating land cover maps is pretty accurate as it relates to what is really on the ground. Hence I support the use of satellite imagery to generate land cover maps” — St. Lucia.

The users also recognised the benefit of using EO data to map land use/land cover and landslides efficiently, thus enabling them to reduce their overall expenditure on data collection in comparison to traditional fieldwork. Another major strength of the EO approach is the ability to produce consistent land use/land cover data in a timely manner, which would provide local users with the means to undertake various monitoring activities that are not possible or viable using current practices. This is permitted by the development and application of largely automated and objective processing techniques in the Service. Use of EO data was also deemed to be particularly advantageous in helping to generate more comprehensive landslide inventory maps, thus overcoming the accessibility issues that limit traditional field-based mapping approaches. The ability to generate DEMs from the optical imagery was perceived to be another major strength of the Services. This was particularly true for the precise 1 m DEM for Belize, with one local user adamantly stating that data of that quality was what they urgently required to enable accurate flood risk mapping.

A major limitation of the EO Services was the dependency on cloud free imagery. Many users commented on the incompleteness of the rivers and roads layers owing to clouds and associated shadows in the imagery. Cloud cover is arguably one of the main restrictions on the use of EO data in tropical environments such as the Caribbean region, largely because of the Atlantic hurricane season. The probability of acquiring useable cloud-free imagery can be maximised by increasing the time duration of the acquisition window or to avoid tasking during known climatic events.

Although information on the validation of the EO products using conventional procedures was provided, it is apparent that some local users preferred the opportunity to undertake their own field validation. This may be partly due to the relative unfamiliarity of these users with EO-based products and conventional validation procedures. Although not possible in this project, more interaction with the local users prior to and during the assessment phase would have helped to improve the level of confidence these user have in the reliability of the EO products. Despite this, all users were able to recognise the substantial benefits provided by the use of EO data in their activities.

References

File:OR15061 bottombanner.jpg