Many flood warning systems do not have proper communication links to the flood affected communities. Similarly, flood hazard maps do not exist for many flood-prone areas. Flood hazard maps are required to plan land use and develop early warning procedures to evacuate people living in these areas. Flood forecasting and warning systems, flood hazard mapping, and evacuation planning need to be closely linked for the management of entire watershed systems. The primary objective of the Flood Management in Selected River Basins Project is to support the GoI’s transition to an integrated flood management approach and improve the preparedness of the government to manage and mitigate negative impacts of flooding in selected river basins. The goal of the ADB support is to transition from the existing project-oriented flood control centred on structural mitigation measures, to a process-oriented integrated flood risk management (IFRM) system that incorporates a mix of non-structural interventions, institutional and capacity building, and structural measures to mitigate the impacts of floods. 

​

The objective of the EO support project was to develop EO-based information products that could be used to prepare flood risk management plans as well as flood forecasting and early-warning systems in support of IFRM. The selected river basins were the Ciujung, Cidurian, and Cidanau River Basin Territory (3Cs RBT) in Banten Province, the island of Ambon, Maluku Province, and Central Seram, Maluku Province. Basin-wide land use and land cover (LULC) and digital elevation models (DEM) were the primary EO products delivered.

 

The need for geospatial information.

 

In formulating and updating river basin flood management plans, specific geospatial data sets are required to support 1) flood hazard mapping to determine the effects of flooding, and 2) two-dimensional floodplain hydrodynamic modelling. Indonesia’s growing population have increased the demand for human settlements and expansion of urban areas resulting in increasing land conversion from forested to non-forested land. Furthermore, poverty and economic development result in improper land cultivation and the conversion of upload forest and wetlands to agricultural land. These incremental changes in land cover have led to soil erosion, watershed degradation, and the loss of valuable resources. 

​

Identification of these critical lands and assessing the condition of the land cover within the selected river basins requires up-to-date LULC maps. However, completeness and vintage of LULC data varies greatly across Indonesia and up-to-date maps are not readily available for many areas. 

​

Topographic data such as DEMs are also fundamental sources of geospatial information required for hydrodynamic modelling, accurate catchment delineation, and drainage mapping. Topographic data also help determine areas of erosion potential and areas prone to flooding by defining low lying areas, topographic depressions, and gradients. Collecting detailed and accurate elevation data for large areas using methods such as LiDAR or stereo-aerial photography is expensive. As a result, other sources of topographic information are used, such as digitised historical topographic maps or freely available global DEMs (e.g., Shuttle Radar Topographic Mission, SRTM, or Aster GDEM). These global data sets require extensive post-processing to convert Digital Surface Models (DSM) to Digital Terrain Models (DTM) and have insufficient detail to be used in hydrodynamic modelling. Other sources of DEMs may be too old to accurately reflect current topography. The EO support project identified a clear need for a more cost-effective source of up-to-date topographic data, with accuracy that is “fit for purpose” for flood hazard mapping and two-dimensional floodplain hydrodynamic modelling for large basins. 

​

Basin-wide land use and land cover 

​

Cost-effective high/resolution optical imagery (RapidEye and Pléiades) was used to produce the basin-wide and urban LULC products. Image compositing techniques to mitigate persistent cloud cover using additional imagery was successfully demonstrated. As a result of image compositing, the project was able to obtain an additional 1,000 km2 (16%) of usable data across for all three study areas. However, due to specific restrictions on image collection dates and the availability of cloud-free images in the archives, complete removal of cloud covered areas was not possible. 

​

Basin-wide LULC classification was based on RapidEye data (5 m resolution) using an object-based methodology. All images were atmospherically corrected. During correction and removal of atmospheric affects, automated cloud and cloud shadow masks were generated allowing for more efficient image compositing. Additional datasets from the Sentinel-1 mission were used and co-registered to the RapidEye data. Given the distinctive radar response from agricultural land in radar imagery such as Sentinel-1, multi-temporal composite images were created to support the classification process. Classification rules incorporated additional information such as elevation, road networks, etc. Thematic information followed GoI’s Bakosurtanal standard for land cover categories (Standar Nasional Indonesia SNI 7645:2010).

​

Urban density and land use 

​

For the urban centre of Ambon Island (i.e. Kota Ambon), urban density and urban land use classification was based on 0.5 m Pléiades data, again using an object-based methodology. All images were processed in a similar fashion as the RapidEye data. Initial urban density was determined by a pixel-based assignment of artificial and natural surfaces within the study area. Subsequently, urban density was derived based on the local area proportion of artificial and natural surface within a given area (50×50 m). Density classes were then defined based on the categories presented in the European Commission’s GMES Urban Atlas project. Additional urban land use classes were also identified via image segmentation and classification, e.g. settlements, natural vegetation, urban greenery and agricultural areas. Urban density and urban land use classes were then merged to produce the final product. 

​

Regional and precision DEMs 

​

High-resolution DEMs were produced for all three project areas using the PRISM sensor onboard the ALOS satellite, with a spatial resolution of 2.5 m. Elevation values were extracted using stereo pairs collected between 2008 and 2011 and the resulting DEM was resampled to 10 m. Elevation values were referenced to above mean sea level (AMSL) and advanced manual editing and filtering to remove artefacts related to the coastline and water surfaces was performed. For areas of dense cloud cover and haze, 30 m Shuttle Radar Topography Mission (SRTM) data was used to interpolate the voids. The production of the DEM required the use of 116 ground control points (GCP) with accurate horizontal and vertical positioning. The Badan Informasi Geospatial (BIG) geodetic control network data (Ina-CORS) was used as reference during the field survey. Forested and mountainous areas of central Seram were inaccessible and primary GCP points were not collected in these areas. 

​

To improve the utility of the EO-based DEM products future enhancements were identified: 1) integrating existing river network data during the DEM creation process to ensure surface drainage flows match the observed river network; 2) producing more cross-sections along river networks to increase the confidence in the accuracy of the product (e.g., over 100 existing cross sections for multiple rivers in Banten Province), and 3) combining river network cross section data with DEM values to estimate water flows. Furthermore, additional validation of the DEM products should be completed for flatter floodplain areas as well as mountainous areas, to quantitatively determine the bias and limitations of the DEM products. 

​

Challenges and limitations 

​

The main challenges in developing the EO-based products was the lack of ancillary data (e.g., road network and cadastral data), persistent cloud cover and presence of haze, and the lack of sufficient ground validation data. Limitations resulting from insufficient ground validation data could have been minimised with the use of alternative validation data sources (e.g. aerial photography captured by Unmanned Aerial Systems – UAS). Persistent cloud cover could have been overcome through a more comprehensive use of the image archives to select cloud-free scenes. Based on end-user feedback, the vintage of the satellite image would have little effect on the final DEM product. However, this would be more important for the creation of the LULC product. The availability of ancillary land use data would also have addressed information gaps where spectral information alone was not sufficient to extract the necessary LULC categories (e.g., industrial areas, primary and secondary forests, etc.). 

​

As persistent cloud cover and limited access to remote areas are common challenges in Indonesia and other parts of the tropics, use of alternative technologies and data sources such as radar imagery which is less susceptible to cloud cover is an attractive option. Radar technology can be a reliable source of remote sensing data that can support LULC mapping and monitoring. Freely available options such as Sentinel-1 could be used to determine baseline land cover categories such as forest and non-forest and flooded paddy fields that are able to assist in studying water run-off or erosion potentials across landscapes. 

​

Impact and benefits 

​

Incremental changes in land cover in the river basins across Indonesia have led to watershed degradation and impacts on valuable ecosystem services. Assessing the LULC and topography within watersheds can identify important areas for protection and restoration to reduce extreme runoff, soil loss, and sediment loads entering river systems. However, completeness and currency of LULC data varies greatly across Indonesia, and up-to-date land cover information is not-readily available for remote areas. 

​

Overall, the products developed under the EO support project fulfilled the important information needs related to flood risk management planning and hydrological modelling at accuracies suited to the needs of the DGWR. The LULC product provided new and improved information. Previously, supplemental information, such as ground based surveys, of land use and land cover was required to help estimate flood impacts to the surrounding settlement areas.

 

Increased detail provided by the urban LULC product for Kota Ambon will enable DGWR and ADB to conduct new tasks and analyses that previously were not possible, such as modelling flooding impact on residential areas. Furthermore, the information gained from higher resolution DEMs will help the DGWR and ADB reduce uncertainty and risks within flood modelling studies. 

​

The LULC dataset derived from RapidEye provides an up-to-date baseline to support assessments of watershed condition that is more detailed compared to existing datasets. Comparing the LULC products against the 2013 national land cover data sets obtained from the Ministry of Environment and Forestry, the RapidEye LULC product derived in the EO support project is more detailed, especially in the delineation of urban settlements. 

​

Topographic data such as DEMs are fundamental sources of geospatial information required for hydrological analyses, such as hydrodynamic modelling and accurate catchment delineation. Topographic data also supports the study of erosion potential, defining low-lying areas, topographic depressions, and slope gradients. SRTM and ASTER GDEM provide a free source of topographic and elevation data. However, without further post-processing and editing, incorrect drainage information, watershed boundaries and gradient profiles are obtained. The horizontal and vertical accuracy of SRTM data is not suitable for floodplain analysis since drainage patterns and simulated flood extent cannot be captured with sufficient detail, especially in low-lying areas. The ASTER GDEM dataset improves on SRTM accuracies. However, since the dataset is a DSM, accurate drainage and elevation profiles can only be exextracted after extensive correction of elevation values to remove peaks and sinks associated to surface features. 

​

The PRISM-derived DEM products fulfil an important information need by providing a much-improved and more detailed representation of topography allowing for more accurate catchment delineations, definition of drainage patterns, and calculation of upstream areas. The greater level of detail in surface topography obtained by the PRISM-based DEM products allows for more accurate watershed boundary delineation. For example, the figure on the left illustrates the boundaries for the 3Cs RBT study area derived from ALOS PRISM, SRTM 30 m, and ASTER GDEM. These boundaries are also compared to the basin boundary obtained from the GIS dataset supplied by DGWR. Generally both the PRISM-derived DEM and SRTM 30 m data set have similar boundaries while the ASTER GDEM and DGWR data set produce significantly different boundaries. Differences in upstream drainage can have significant impacts on estimated water flow and flood volume calculations given these important metrics are a function of drainage area, slope, soil sealing among other parameters. 

​

Based on a review meeting with the ADB and DGWR, future EO-derived LULC products will support two-dimensional floodplain hydrodynamic modelling and assessment of flood hazard and risk while reducing uncertainty and project risk. Flood management activities will be enhanced and improved on through the use of the EO-based products. Specifically, the basin-wide LULC product for all three study areas extracted from high resolution satellite imagery will 1) provide a more detailed description of river features including river flooding zones, allowing for better prediction of flooding effects in critical areas; and 2) support levee monitoring and identification of critical levies within the basin. 

​

Service sustainability 

​

The LULC and DEM products in the present project are particularly relevant to disaster risk management, hazard monitoring, environmental assessment and monitoring, and route and facilities siting and planning. Their incorporation into ADB project workflows and activities would provide more confidence in subsequent analysis conducted by these projects and typically reduce costs compared to other methods of data collection. 

​

The products can also be used to derive other geospatial information such as erosion potential and runoff potential. An approach to sustainability is to demonstrate their multiple uses and work with EO product suppliers to standardise products and services. Some EO services have been standardised, such as elevation products from some vendors. However, the standardisation in these cases tends to be targeted to the source image products from each vendor, and not necessarily to the specific needs of individual end users.