OR/15/001 Executive summary

From Earthwise
Jump to navigation Jump to search
Jordan, J J, Grebby, S, Dijkstra, T, Dashwood, C and Cigna, F. 2015. Risk information services for Disaster Risk Management (DRM) in the Caribbean. British Geological Survey Internal Report, OR/15/001.

The World Bank project context

The primary objective of this ESA project is to raise awareness within the World Bank (WB) of the capabilities of Earth Observation (EO) data and specialist service providers to supply information customised to the specific needs of individual projects. This project was set up within the ESA/WB eoworld initiative to contribute to the WB Caribbean Risk Information Program that is operating under a grant from the ACP-EU Natural Risk Reduction Program.

The Caribbean is heavily affected by natural (and geo-) hazards with over 5 billion US$ in losses in the last 20 years (source: CRED database) Figure 1 illustrates the division of natural disaster by occurrence in the region over the last 30 years, providing an insight into the impact in the region over a significant time period. A specific example of the environmental, social, economic and political issues that the project is addressing is highlighted by the effects of Hurricane Tomas on St Lucia in October 2010. The hurricane resulted in seven deaths with 5952 people severely affected, while the cost of the damage was estimated at US$336.2 million, representing 43.4% of GDP (ECLAC, 2011[1]). Understanding and mitigating these 'geo-environmental disasters' (as they are termed in ECLAC, 2011[1]) is a primary concern in the region.

Figure 1 Natural disaster occurrences in the Caribbean region. Source: CRED database.

The ITC-led Caribbean Handbook for Disaster Information Management (CHARIM) is operating in five Caribbean countries (Belize, Dominica, Saint Lucia, St Vincent & the Grenadines and Grenada). CHARIM has several objectives including developing a structure for landslide and flood hazard and risk assessment. This ESA project is focussed on 'risk information services for disaster risk management in the Caribbean' a title that has been abbreviated to EO-RISC (Earth Observation for Risk Information Services in the Caribbean) internally by BGS. EO-RISC deliverables directly contribute to the CHARIM objectives e.g. by providing data and services that enable certain hazards such as landslides to be identified directly.

EO-RISC is addressing various issues in the Caribbean. In broad terms, the Latin America and Caribbean Regional Urban and Disaster Risk Management Unit (GPSURR) with funding from the ACP-EU Natural Disaster Risk Reduction Program, managed by the Global Facility for Disaster Reduction Recovery (GFDRR) has begun the 'Caribbean Risk Information Programme to support the Integration of Disaster Risk Management Strategies in Critical Sectors' project. This has been initiated in order to strengthen the regional and national capacity to create and use hazard and risk information for planning and development processes, and consists of four components: (a) creation of a geospatial information basis, focusing on the collation, quality control and adequate storing, management and sharing of existing geospatial data in a spatial data infrastructure, (b) development of a methodological framework for the development of hazard and risk information for development and planning processes, (c) implementation of five national pilot hazard studies aimed at implementing the methodological framework in partnership with Caribbean countries, and (d) integrating institutional strengthening as a cross-cutting activity to all components. The Caribbean Risk Information Programme forms part of the Probabilistic Risk Assessment (CAPRA) Program whose objective is to enhance the capacity of targeted sectors in Latin America and the Caribbean region to develop and mainstream disaster risk information into development programs and policies by providing knowledge products and services. Counterpart agencies are the Ministries of Works and Physical Planning in the following countries: Belize, Dominica, Grenada St. Lucia and St. Vincent and the Grenadines. With a focus on national-level landslide and flood hazard assessments, country-wide baseline data and information are required. They span a broad range such as: Land Use/Land Cover, updating of river and stream courses, extent of lakes, water bodies, and watersheds, basic road network, landslide inventory, Digital Elevation Models, geology including fault lines, geomorphology, soil maps, etc.

Table 1 lists the hazard characteristics of the four islands that EO-RISC is covering, and provides an overview of the issues and problems being encountered by agencies in the region. CHARIM introduced BGS to the local stakeholders in the region at the WB workshop in St Vincent in October 2014. Presentations by BGS introduced the EO data and the preliminary derivative services, and feedback was received from the stakeholders/potential users on the services and their formats.

Table 1 General disaster management and hazard characteristics for the four countries in EO-RISC (Source: CDEMA website, and modified from van Westen, 2014[2]).
Belize Saint Lucia St. Vincent and the Grenadines Grenada
Area 22 806 km2 606 km2 389 km2 (Saint Vincent 344 km2) with 32 islands and cays 344 km2
Coastline 386 km 158 km 84 km 121 km
Terrain Flat, swampy coastal plain; low mountains in south. Max. elevation 1160 m Volcanic and mountainous with some broad, fertile valleys. Max. elevation: 950 m Volcanic, mountainous. Max. elevation: 1234 m Volcanic in origin with central mountains. Max elevation: 840 m
Natural hazards Frequent, devastating hurricanes (June to November) and coastal flooding (especially in the south) Hurricanes and volcanic activity, landslides, debris flows, flashfloods Hurricanes; Soufriere volcano on the island of Saint Vincent is a constant threat. Flashfloods and landslides Lies on edge of hurricane belt; hurricane season lasts from June to November. Flashfloods and landslides.
Hazard characteristics Hurricanes and tropical storms are the principal hazards, causing severe losses from wind damage and flooding due to storm surge and heavy rainfall. Hurricanes Keith (2000), and Iris (2001) caused some of the worst damage ever, reaching 45% (US$280 million) and 25% of GDP, respectively. Saint Lucia’s mountainous topography coupled with its volcanic geology means that it experiences landslides, particularly in the aftermath of heavy rains. Much of the island’s housing is distributed along steep slopes and poorly engineered and constructed housing is particularly at risk. Additionally, the island periodically experiences earthquakes of generally lower magnitudes. Also storm surge and flash floods are among the other risks regularly faced by the island. Landslides, particularly on the larger islands, are a significant hazard and the risk is increased during the seasonal rains. Coastal flooding is a major concern particularly relating to storm surge and high wave action. The Grenadines are more susceptible to drought. The active volcano La Soufriere, located on the north end of St. Vincent is another risk factor, posing threats from shallow earthquake and eruption events. Since 1900, St. Vincent has been hit by 8 named storms, the strongest being Hurricane Allen (Category 4), which passed between St. Lucia and St. Vincent in 1980. The 1939 eruption of the volcano Kick‐‘em‐Jenny located some 100 km S of Grenada, generated a 2‐meter high tsunami. The country was heavily affected by Hurricane Ivan in 2004, and Hurricane Emily in 2005. There are two active volcanoes in Grenada, Mount St. Catherine in the center of the island and the submarine volcano Kick‐‘em‐Jenny is located 8 km north of the island and has led to tsunami in the past. Flood risk in Grenada is largely associated with storm surge in low lying coastal areas. Flash flooding from mountain streams coupled with storm surge events are the primary causes of flood events and effects are generally limited to communities located in the coastal margins along stream passages. Landslides are a common event in Grenada, with much of the impact experienced along the roadway network.
Population 334 297 (2013) 174 000 (2010). 104 574 (2009) 110 000 (‐)

Of the natural disasters listed above, BGS has been tasked with producing an inventory of landsides; over 1200 landslides were identified in St Lucia alone from the satellite imagery. The impact of this geohazard on the island was noted at first hand during the fieldwork when travel was curtailed e.g. roads were closed due to landslides (Figure 2).

Figure 2 Road closure due to landslides — the type of problem the project is addressing.

Section 3 outlines the products that were delivered in response to the issues affecting the region.

Requirements for Geo-Spatial Information

The geospatial data listed in Table 2 are the primary ones identified by the users that are utilised by WB and local users through a variety of projects and initiatives. EO-RISC has been tasked with providing landslide inventories, DEMs and landcover maps for selected territories in order to contribute missing information, to update older information or to increase the (spatial/temporal) resolution of existing information.

In Grenada mapping and GIS capability is managed predominantly by the Ministry for Agriculture, but progress is limited. A school landslide vulnerability assessment has been (http://www.oas.org/CDMP/document/schools/vulnasst/gre.htm). No comprehensive multi-hazard map compilation has been prepared. The WB is implementing a Disaster Vulnerability Reduction Programme (DVRP). Component 2 (Disaster and Climate Risk Reduction) of the Disaster Vulnerability Reduction Project which would consist of new construction and rehabilitation of existing infrastructure in order to reduce their vulnerability to natural hazards and climate change. Included within the activities are consultancy services to undertake soil investigation mitigation measures for landslip sites in several sites.

In St. Vincent and the Grenadines, progress in preparation of hazard maps is limited. To date, risk mapping has been limited to volcanic risks and some coastal vulnerability analyses. Basic GIS-ready maps of roads, contours, rivers, coastlines, agricultural & urban land use have been prepared — primarily available through the Ministry of Planning and the National Emergency Managements Organisation (NEMO). The WB is implementing a Disaster Vulnerability Reduction Programme (DVRP). Components include identification and creation of required baseline data for hazard assessment; development of institutional systems for the collection, sharing and management of geospatial data among national agencies and with regional institutions; training and education in applications integrating geospatial data systems, hazard and risk assessment to support decision making within various sectors and mainstream the use of these tools as a standard practice in development planning.

In Saint Lucia, hazard maps have been produced for debris flows, but these may not reflect current conditions. Furthermore, NEMO is not equipped to support GIS data and there is no program to support additional hazard mapping. The WB is implementing a DVRP. Component 2 (Technical Assistance, Regional Collaboration Platforms for Hazard and Risk Evaluation, Geospatial Data Management, and Applications for Improved Decision-Making) would finance: a series of capacity-building, knowledge-building and technical assistance interventions at the national and regional levels to support disaster risk management and climate change adaptation. There are specific areas that have been identified and proposed as high priorities for intervention. At the national level, activities would include, inter alia: i) enhancement of national hydro-meteorological monitoring networks; ii) development of an integrated watershed management plan for flood mitigation; iii) technical assistance for the establishment of maintenance monitoring systems for bridges and public buildings that would integrate natural hazards and extreme events considerations; iv) establishment of geo-spatial data sharing and management platform and related training activities; and v) climate change adaptation public education and awareness campaigns. The GeoNode platform for Saint Lucia http://sling.gosl.gov.lc is accessible.

In Belize, no nationwide flood hazard maps have been made for the country based on hydrological modelling, and the source of the only flood map identified by van Westen (2014)[2] was unclear. However, hazard mapping has been completed in several areas with GIS datasets covering landslide risk, volcanic hazard assessment and storm hazards amongst others. Belize is participating in the Central American Probabilistic Risk Assessment (CAPRA) platform but the initiative remains modest in Belize.

Table 2 Geospatial information sources currently used by WB teams and/or local users (derived and updated from van Westen, 2014[2]). The EO-RISC services are highlighted in green.

Current geospatial information sources


St Vincent and the Grenadines

Saint Lucia


DEM 10m raster DEM (source unknown) and partial LiDAR coverage 5m raster DEM (higher parts are not covered). There are LiDAR data of St.V but the format is incorrect so they cannot be analysed 50m raster maps and contours with 2.5m intervals ASTER and SRTM. Higher resolution data are urgently required for flood risk modelling.
30m DEM 30m DEM National DEM at 30m, 40% of territory at 20m and sub-area at 1m
Landcover USDA 30m raster map Polygon map exists with 11 land use classes 1:50 000 raster maps. Vegetation information is in vector format
Landcover derived from 2m satellite data Landcover derived from 2m satellite data Landcover derived from 2m satellite data
Landslide inventory and hazard map 1988: OAS study for selected towns. 2006: CBD/CDERA study — limited inventory of 40 landslides, but not available digitally Landslide footprints are available, but there is no detail 2010 inventory map has been produced from satellite imagery Not applicable
Landslide inventory at 1:20 000 Landslide inventory at 1:20 000 with key areas (no more than 50%) at 1:10 000
Elements‐at‐risk Non‐attributed building footprints Not available Available for the country, including building footprints — though occupancy and structural type is unavailable Not available
Building footprints may be derived from 2m satellite imagery, but this is not a priority for EO-RISC Building footprints may be derived from 2m satellite imagery, but this is not a priority for EO-RISC Building footprints may be derived from 2m satellite imagery, but this is not a priority for EO-RISC
Geological map A very general one is available, made by USGS A very general one is available, made by USGS Vector map is available
Soil map A 1959 soils report exists but ITC have not been able to obtain the 1959 map General soil map from USAID from 1990 Vector map is available General map has been scanned by ITC
Discharge data Continuous stream flow data do not exist None available None available None available
Geotechnical data None available to date None available None available
Rainfall data Approx 50 rainfall stations. Data is not continuous. Data available from the Land Use Division, Ministry of Agriculture, Lands, Forestry and Fisheries None obtained thusfar, but rainfall stations do exist Hourly rainfall data for 24 stations Missing
Socio‐economic data Missing Missing Missing Missing

The fieldwork and discussions with local users has highlighted two main future geo-spatial information requirements:

  1. The landslide inventory and land cover maps are vitally important to gain a full understanding of the events and the associated risk. However landslides are highly dynamic systems, and the land cover is also constantly changing so the requirement is to update these maps on an annual or bi-annual basis. The annual imagery used from 2010 to 2014 to create the landslide inventory highlighted patterns e.g. related to weather systems, however more data from inventories gathered in successive years would help to clarify potential thresholds for weather fronts and their impacts on the environment;
  2. Those missing datasets (such as geotechnical data) or ones that are identified as ‘general’ (such as geology) are obligatory for the users to gain a true understanding of the geohazard processes. For example EO-RISC identified that there were far fewer landslides in Grenada than in St Lucia, despite similar topography and weather conditions on both islands. Therefore it is proposed that increased knowledge of geology, geomorphology, and soils coupled with geotechnical data are mandatory if the associated risks and the potential to provide forecasts for landslides is desired.

Interpretation of the results

The WB actively supports Disaster Risk Reduction in the Caribbean and has received a grant from the European Union for the Caribbean Risk Information Programme regarding the development and use of risk information in critical sectors. The primary Risk Information Programme objective is to strengthen the regional and national capacity of governments in the Caribbean region to develop or procure the development of landslide and flood hazard and risk information. EO-RISC is contributing to the WB requirements via the CHARIM project by providing them with the products, which are outlined in products description. Specifically, CHARIM is developing national hazard mapping studies in the five target countries (Belize, Dominica, St. Lucia, St. Vincent and the Grenadines and Grenada), one on Belize related to floods and two on each island for landslides and floods.

Contact with the Users was required to define the user requirements that were outlined in the Service Readiness Document (Jordan and Grebby, 2014[3]) and subsequently to refine the services to ensure fitness-for-purpose. The sole face-to-face contact with users (to date) was via the CHARIM workshop in St Vincent (29 September – 3 October 2014) which was attended by users including Chief Engineers, Chief Planners and Geospatial Experts from each Caribbean country. Preliminary EO-RISC services / results were presented to the users assembled as an entire group (30th September) and subsequently to a focused meeting of the geospatial experts during a technical meeting (1st October). This enabled us to interpret how the results relate to the user requirements.

Service 1: Land use/land cover mapping

The objective of Service 1 is to generate land use/land cover maps for St. Lucia, Grenada, and St. Vincent & the Grenadines by exploiting recent high-resolution or very high-resolution optical satellite imagery. As well as land use/land cover, the objective was to utilise the imagery to produce a vector layer of water features (e.g. lakes, ponds, rivers) present in each of the areas of interest (AOI). With a preference on utilising imagery acquired using European and Canadian sensors, a set of recent images with acceptable levels of cloud cover was identified and obtained from the relevant archives for each of the three AOIs. The satellite data comprised Pleiades imagery (acquired between 2013–2014) and RapidEye imagery (acquired 2010–2014). These datasets have a spatial resolution (pixel size) of 2m and 5m, respectively, for the multispectral waveband images. Additionally, the Pleiades datasets includes a very high-resolution 0.5m panchromatic image.

Existing land use/land cover maps for the three AOIs were produced as part of The Nature Conservancy’s Mesoamerica and Caribbean Region project (Helmer et al., 2007[4]; 2008[5]). These maps were derived at 30m resolution from satellite imagery with the aid of extensive field knowledge and observations. Accordingly, these maps represent useful baseline data to build upon for the land use/land cover mapping under this service.

To enable the most detailed information to be resolved, the Pleiades imagery was used as the primary dataset for generation of the new land use/land cover maps for the three AOIs; thus achieving a spatial resolution of 2m, which is equivalent to a mapping scale of 1:10 000. For each of the AOIs, land use/land cover was mapped using a combination of automated image classification, rule-based refinement and manual digitisation. The existing 30m maps were used to define the different land use/land cover types and identify representative areas in the imagery to help guide the initial automated classification and to subsequently validate the mapping. Water features and the basic road networks were manually digitised at 1:10 000-scale from Pleiades imagery that had been pan-sharpened to 0.5m resolution using the panchromatic image. Wherever available, existing vector layers were utilised as baseline information during mapping.

Cloud and associated shadow coverage in the Pleiades imagery was quite significant, typically varying in the region of 20–40% for the AOIs. Wherever the ground was obscured by cloud and shadow in the Pleiades imagery, the land use/land cover maps were patched using the RapidEye imagery and the existing land use/land cover maps in order to provide complete areal coverage of the AOIs. Unfortunately, it was generally not possible to map water features (particularly rivers and streams) and roads in the areas obscured by cloud and shadows in the Pleiades imagery because the alternative RapidEye imagery lacked the spatial resolution required to resolve such features.

The land use/land cover maps were validated using a standard remote sensing approach, which involves comparing the land use/land cover class identities of a sample of pixels in the map with their ‘true’ land use/land cover class. The ‘true’ land use/land cover classes of these pixels were determined using a combination of the pan-sharpened Pleiades imagery and existing maps. Consequently, the maps for St. Lucia, Grenada, and St. Vincent and the Grenadines were found to have accuracies of 84.9%, 84.8% and 80.8%, respectively; which are within the desired target accuracy of 80–90%. Additional validation of the maps for St. Lucia and Grenada was achieved using point-sampled field observations at a number of locations.

Compared to the existing maps, the new land use/land cover maps derived for this service represent an order of magnitude increase in terms of spatial resolution — increasing from 30m to 2m. As a result, the new maps provide much more detailed information on the distribution of the different land use/land cover types in the AOIs. Moreover, a number of significant errors (due to misclassification) in the existing maps have been corrected in the new maps. Nevertheless, a relatively minor degree of confusion between inherently similar land use/land cover types (particularly vegetation) does persist. Overall, the results demonstrate the potential to utilise high-resolution satellite imagery to produce detailed and accurate land use/land cover maps more efficiently than through an equivalent field-based survey.

Service 2: Hazard mapping to support landslide risk assessment

The objective of Service 2 is to generate ground-truthed landslide inventories and digital elevation models for both St. Lucia and Grenada from optical satellite imagery.

Landslide inventory mapping

The establishment of landslide inventories for St Lucia and Grenada is based on the interpretation of satellite images covering the period 2010–2014. For most of this period RapidEye images are available. Images from the Pleiades satellite are only available for 2014.

Landslide activity can result in the disturbance of vegetative cover and exposure of soils at the surface. This spectral signature is combined with an assessment of other information such as position in the landscape, slope morphology, vegetation cover, etc. to interpret the satellite images and create outlines of landslide events. The distribution of landslides for each image (year) was captured manually by skilled operators and the results stored in one event database. The attributes stored for each event are shown in Table 3. The interpretation of potential landslide sites was analysed throughout the complete image sequence. This enhanced confidence in the mapping process, particularly if polygons are visible in several images. This approach also helped to reduce the negative effects of cloud cover and occasional poor image quality (e.g. the SE quadrant of the 2013 RapidEye satellite image of St Lucia).

It is theoretically possible to capture a landslide spectral signature automatically, but our experience has shown that this leads to an over-representation of cultivated fields necessitating supervised re-classification of every polygon. It was therefore decided not to pursue this approach.

The RapidEye images are available at a resolution of 5 m while the Pleiades image was pan- sharpened to a resolution of 0.5m. Determination of landslide events at 5 m resolution is not very reliable and results therefore in rather low confidence mapping. However, when polygons persist into the 2014 Pleiades images, much more detailed interpretation can be achieved leading to greater confidence in the mapped product.

Table 3 Attributes stored for each landslide event in the multi-temporal cumulative inventory.
Description Name Type Length Information
Landslide ID LID number


polygon identifier that can be related to landslide database entry point
Location District DISTR text


district name
Location Locale LOCAL text


locality name
Movement type TYPE code


.. (not entered), FL (flow), SR (rotational slide), SP (planar slide), SU (undifferentiated slide), FA (fall), TO (topple), SP (spread), UN (undefined)
Morphology MORPH code


L, S, T, A (Landslide undifferentiated, Scarp, Transport zone, Accumulation zone)
Confidence CONF code


H, M, L
2010 2010 code


N, I, A (Not present — no slide visible, Inactive — the slide can be recognized but no activity suggested by disturbed vegetation or bare surfaces, Active — slide shows clear signs of recent activity in the form of disturbed vegetation, etc.)
2011 2011 code


N, I, A (Not present — no slide visible, Inactive — the slide can be recognized but no activity suggested by disturbed vegetation or bare surfaces, Active — slide shows clear signs of recent activity in the form of disturbed vegetation, etc.)
2012 2012 code


N, I, A (Not present — no slide visible, Inactive — the slide can be recognized but no activity suggested by disturbed vegetation or bare surfaces, Active — slide shows clear signs of recent activity in the form of disturbed vegetation, etc.)
2013 2013 code


N, I, A (Not present — no slide visible, Inactive — the slide can be recognized but no activity suggested by disturbed vegetation or bare surfaces, Active — slide shows clear signs of recent activity in the form of disturbed vegetation, etc.)
2014 2014 code


N, I, A (Not present — no slide visible, Inactive — the slide can be recognized but no activity suggested by disturbed vegetation or bare surfaces, Active — slide shows clear signs of recent activity in the form of disturbed vegetation, etc.)


free text

The project imposed scale limitations pre-determined that the landslide inventory should be established at a scale of 1:20 000 with key areas (no more than 50%) at 1:10 000 for St Lucia, and at 1:20 000 scale for Grenada. This scale limitation affects the mappable minimum size of landslides. Based on the experience of mapping landslide polygons in St Lucia using the Pleiades high resolution images as guidance the following limitations apply. At a scale of 1:20 000 the minimum mappable size is approximately 30 by 30m or 900m2. At 1:10 000 this enhances to mappable polygons of about 15m side lengths (225m2). At a scale of 1:5 000 it is possible to map elements with effective minimum dimensions of about 10m (100m2), depending on the terrain. Closely grouped small events were sometimes visible and these have been mapped as landslide clusters. The database therefore contains some polygons that contain several events (too small to map individually).

Considering the difficulties encountered in mapping landslide polygons at 1:20 000 scale this project adopted a pragmatic (though time-consuming) approach where the landscape was interpreted at 1:5 000 scale (or an even more detailed scale where features were uncertain). Outlines were then up-scaled (i.e. generalized) to be representative of polygons at 1:10 000-scale (this is a standard BGS approach). As a consequence of this practice it was possible delineate landslide events in the size range smaller than 1000 m2 (approximately 100 events) and this has resulted in a more ‘complete’ landslide inventory.

The clarity and detail offered by the high resolution (0.5m) Pleiades has been used to carry out detailed investigations of a limited number of individual sites and events. In combination with other data (landuse, topography, etc.) it is possible to generate highly detailed geomorphological maps that not only show the spatial extent of an event, but also can be attributed with information on the likely nature of deformation and, in combination with other images, a timeline of event progression. This level of detailed interpretation falls outside the scope of work for this project but is discussed in case studies in Section 3 of this report to highlight the significant additional value of these new, high resolution products.

The opportunities to capture landslide event outlines is strongly linked with the time period between trigger and image capture. The use of a multi-temporal image stack therefore provides the best opportunity to achieve ‘completeness’ of the database. The database can then be used to evaluate how quickly a landscape recovers and what the consequences are of subsequent trigger events (Figure 3). For St Lucia, the time series captured two major landslide trigger events; the 2010 Hurricane Tomas event and the 2013 December Trough. This has led to important insights into changes in the annual inventories as the landscape firstly recovers and then gets disturbed at a later date.

Figure 3 Landscape recovery versus trigger event recurrence. When recurrence exceeds recovery, relatively stable system vulnerability can be assumed (A), but when recovery exceeds recurrence, system vulnerability is likely to increase significantly. (From Dijkstra et al., 2014).

Two previous inventories were available for comparison with the present dataset. In 1995 some 712 events were identified, whilst an inventory created following 2010 Hurricane Tomas captured 1132 landslide events. The current multi-temporal inventory covered the years of 2010–2014 and contained 1233 landslide polygons that have been classed as active (fresh signs of landsliding) or inactive (no evidence of active movement, but still recognisable landslide features) at least once during this period. Generally, each polygon represents a single event. However, where clusters of very small events (dimensions smaller than about 5m) are encountered, a single polygon can represent more than one landslide. There are considerable benefits offered by a sequential analysis covering several years, including a reduction in the effects of cloud cover, a better insight into persistence of features and a more comprehensive capture of events. Any year looked at in isolation is likely to result in fewer events being recorded.

The 1995 landslide event database is largely based on field observations. This resulted in the capture of many events along roads and relatively few events in the forested areas where access is very limited. This database was checked against the multi-temporal inventory and relatively little overlap was encountered. Even taking into consideration the large time gap between the 1995 and 2010–14 inventories it is an indication that field capture and satellite image interpretation of events result in different populations and should be regarded as complementary activities.

The 2010 inventory captured the events generated by Hurricane Tomas and is based on satellite image interpretation. It appears that the use of ‘bare earth’ automated classification provided an important contribution to the establishment of the landslide inventory. As a consequence, the difference between this inventory and our multi-temporal inventory is considerably larger than the numerical difference between the two.

Field verification

During a 6-day field visit to St Lucia more than 650km were covered and as many landslides as feasible were visited. Many rural roads were still blocked as a consequence of landslides generated during Hurricane Tomas and the 2013 December Trough hampering access to landslides in the interior of the Island. Field verification in Grenada was limited to two days only, but during this brief period of time it became apparent that only one of the mapped polygons did in fact represent a landslide. The false positives were generally the result of newly cultivated fields on hillsides, many of which involved small banana plantations.

The field verification emphasised the importance of satellite image interpretation. Many of the mapped landslides are some distance away from roads. Gaining access to these sites in the field is very laborious and the road network does not reach very far inland. In addition, particularly in the case of St Lucia, many of the smaller rural roads in the interior were dramatically affected by landslide events triggered by Hurricane Tomas and the December 2013 Trough. It was therefore often impossible to reach landslide sites beyond those that cut off the roads.

In conclusion the following observations are drawn regarding the use of satellite images for landslide inventory establishment for St Lucia and Grenada:

  • Landslide signatures in St Lucia are recognisable by skilled operators
  • Observations are limited by clouds, not by road access, enabling much more comprehensive coverage
  • The current database is constrained by scale (1:20 000 and 1:10 000) and identification of events at more detailed scales is possible, particularly with recent Pleiades images at the resolution of 0.5m
  • Minimum landslide dimensions in the database are approximately 200 m2 and many smaller events are known to have occurred
  • Very small (<5m) and obscured (in the shade, on steep slopes, below overhanging vegetation) landslides are difficult to capture
  • Small events can still have a significant impact on lives and livelihoods and recording these through different means will complement the database
  • Automatic classification is, at present, not conducive to establishing a reliable record
  • Temporal proximity of trigger event and satellite image acquisition affects the number of events that can be captured
  • Multi-temporal inventory establishment enhances the number of events captured and can be used to establish derived products such as landscape resilience and hazard assessments
  • Landslides triggered by Hurricane Tomas (2010) were rapidly covered by vegetation indicating a rapid rate of recovery of the landscape, but many events were re-activated during the 2013 December Trough indicating a still heightened sensitivity of the landscape to disturbance
  • Extending the multi-temporal record with ongoing acquisitions will create further insights into landscape response and this will be vital in establishing relevant hazard and risk assessments
  • Landslides in Grenada are much more difficult to establish due to widespread cultivation practices resembling landslide signatures
  • Field verification remains an essential tool to ascertain validity of image interpretations

Digital Elevation Models

The desired spatial resolution of the digital elevation models (DEMs) to be produced from optical satellite imagery for St. Lucia and Grenada was 30m or better. Originally, it was intended to utilise stereo Pleiades satellite imagery to derive DEMs with significantly higher resolution of ca. 1m. With no suitable imagery available in the appropriate archive, a request to have new stereo imagery acquired for the two AOIs was submitted in early August 2014. However, due to the timing of the request coinciding with the hurricane season in the Caribbean region, no suitable cloud-free stereo imagery for either AOI has yet been acquired.

As an alternatively, national 30m DEMs for both AOIs were produced using stereo imagery acquired by the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) sensor. Specifically, the nadir and backward-looking ASTER band 3 stereo images can be used to extract DEMs through the application of standard photogrammetric processing techniques.

This approach was utilised by NASA and Japan’s Ministry of Economy, Trade and Industry (METI) in order to produce the ASTER Global DEM (ASTER GDEM); of which version 2 is the most recent (released in 2011).

For St. Lucia, the ASTER-derived elevation data were first vertically calibrated with a subset of contour height data using regression analysis. Next, the calibrated ASTER elevation dataset was merged with the contour height data to form a single file containing x-y-z coordinates of all points. These points were then interpolated to generate the 30m DEM. The vertical accuracy of the DEM was determined using 18 GPS control points provided by the Physical Planning Office, and found to have a root-mean-square (RMS) vertical accuracy of 1.4m.

Similarly, for Grenada, the ASTER-derived elevation data were vertically calibrated with a subset of airborne Light Detection And Ranging (LiDAR) data using regression analysis. Sizeable ‘pit’ artefacts in the DEM were subsequently rectified with the aid of elevation data acquired by the Shuttle Radar Topography Mission (SRTM). The vertical accuracy of the DEM was determined using an independent subset of 500 LiDAR data points and found to have an RMS vertical accuracy of 4.9m.

Service 3: Digital Elevation Models of Belize

The objective of Service 3 is to generate a national DEM of Belize and a precise DEM for a subset of Belize to support hazard/risk assessment.

National Digital Elevation Model

The desired spatial resolution of the national DEM to be produced from optical satellite imagery for Belize was 30m or better. A 20m DEM derived from high-resolution SPOT-5 stereo satellite imagery for 40% of Belize was also acquired from Airbus Defence & Space. This DEM is reported to have an RMS vertically accuracy of 7m.

With no other suitable imagery available, a 30m DEM covering the entirety of Belize was generated based on the ASTER-derived elevation data. Firstly, ASTER-derived elevation data were first vertically calibrated with a subset of elevation data from the 20m DEM using regression analysis. Next, a 3×3 pixel moving average filter was applied in order to reduce 'noisy' data values and notable 'spike' artefacts present in the calibrated DEM. In the absence of any GPS or other control data, the accuracy of the 30m DEM was determined using an independent subset of 32 000 points extracted from the 20m DEM. The national 30m DEM was found to have an RMS vertical accuracy of 9.8m.

Precise Digital Elevation Model

The second aspect of Service 3 was to generate an accurate 1m resolution DEM from very high-resolution optical satellite imagery for a subset of Belize encompassing an area to the north of Belize City. To meet these requirements, a request to have new tri-stereo (triplet) Pleiades imagery acquired was submitted in early August 2014. However, due to the timing of the request coinciding with the hurricane season in the Caribbean region, no suitable cloud-free tri-stereo imagery for has yet been acquired.

Guidelines for use

The land use/land cover maps produced under Service 1 provide detailed information in St. Lucia, Grenada, and St. Vincent & the Grenadines. The products are equivalent to a map scale of approximately 1:10 000. The accompanying vector layers of water bodies, rivers and streams and the basic road network were also mapped at 1:10 000 scale. Accordingly, the intended scale for use of these data is 1:10 000, and so may not be representative at finer scales. The land use/land cover maps provide 100% coverage of the AOIs, however the vector layers are fragmented due to cloud and shadow obscuring the ground in the Pleiades satellite imagery. Overall, the new land use/land cover maps represent a considerable improvement on the existing 30m maps.

The land use/land cover maps can be used to determine spatial correlations with landslide occurrences documented in Service 2. Such analysis is useful in establishing whether specific land use/land cover types are more prone to landslide events, and can thus be used as input, alongside the DEMs (also generated in Service 2) to derive landslide susceptibility maps for St. Lucia and Grenada. Additionally, the DEMs generated in Service 2 can also be used in conjunction with the water surface features mapped under Service 1 to model the flood risk in both St. Lucia and Grenada. Furthermore, the land use/land cover maps could be readily turned into impervious layers, which can also be incorporated in flood risk analysis. The 20m and 30m DEMs produced for Service 3 can also be used to model flood risk in Belize.

Beyond the scope of this project, the land use/land cover information can be used for a wide spectrum of uses. For example, the maps could be used for planning purposes, asset management and in developing forestry management strategies. The data can also be used to monitor change over time. Some broader applications of the DEMs could include forestry management, the planning of new transport infrastructure (i.e. roads and railways), and natural resource exploration.

Landslide inventories for St Lucia and Grenada contain polygons that represent the maximum extent of events mapped during the period 2010–2014. Each polygon is attributed with landslide type, morphology, confidence level of the mapped outline and a statement of activity for each of the five years in the sequence 2010–2014. Because of the prevalence of small events, the inventories were established on a scale of 1:10 000 (a pragmatic limit considering 5m resolution of RapidEye images). However, high 0.5 m resolution Pleiades images will allow landslide inventory establishment at even more detailed scales.

The inventory provides a clear indication of landscape response to trigger events and this information can provide context to future studies of landslide hazard and landslide risk reduction.

Interpretation of satellite images requires well-trained operators with a good understanding of mass movement processes and local conditions.

It is also good practice to populate each polygon with additional information that can be stored in a relational database for further analysis. This was not within the remit of this project and requires further work, but will result in an invaluable tool for future landslide hazard and risk assessments. Examples of the additional information are included in Guidelines for use.

The current multi-temporal sequence can be extended both using older images and through regular updates, at least annually and when major trigger events such as extreme rainfall or earthquakes occur.


  1. 1.0 1.1 ECLAC Report (2011). Saint Lucia: Macro socio-economic and environmental assessment of the damage and losses caused by Hurricane Tomas: a geo-environmental disaster. Towards Resilience. Economic Commission for Latin America and the Caribbean. 183pp.
  2. 2.0 2.1 2.2 Van Westen C J (2014). Preliminary Assessment Report: CHARIM Caribbean Handbook on Risk Information management. ITC, University of Twente.
  3. Jordan, C J and Grebby, S. (2014) Risk Information Services for Disaster Risk management (DRM) in the Caribbean: Service Readiness Document. British Geological Survey Open Report ) OR/14/064 and ESA Technical Report.
  4. Helmer, E H, Schill, S, Pedreror, D H, Kennaway, T, Cushing, W M, Coan, M J, Wood, E C, Ruzycki, T and Tieszen, L L. (2007). Forest formation and land cover map series-Caribbean Islands. U.S. Geological Survey/Earth Resources Observation and Science (EROS), Land Cover Applications and Global Change, International Land Cover and Biodiversity, Caribbean Land Cover Analyses.
  5. Helmer E H, Kennaway T A, Pedreros, D H, Clark, M L, Marcano-Vega, H, Tieszen, L L, Ruzycki, T R, Schill, S R and Carrington, C M. (2008). Land cover and forest formation distribution for St. Kitts, Nevis, St. Eustatius, Grenada and Barbados from Decision Trees classification of cloud-cleared satellite imagery. Caribbean Journal of Science, 44, 175–189.