USING SPOT AND RADAR DATA TO INVENTORY FORESTS IN SARAWAK

Introduction	Linchpin of the local economy
Identifying user needs	Image preprocessing
Network extraction and forest inventory	Estimating timber reserves

Introduction Logging is a major economic activity in the Southeast Malaysian State of Sarawak. Timber companies therefore need to inventory forest resources and estimate timber reserves to ensure they are managed sustainably. Satellite-based earth observation provides timely, accurate and affordable data on the quality and quantity of forest resources. This article highlights the combined advantages of radar and Spot data in establishing a forest inventory and evaluating timber reserves. Top of the page

Linchpin of the local economy

The Malaysian State of Sarawak, in the northern part of the island of Borneo, still has more than 60% of its territory covered by forest. Logging for construction wood and paper is vital to the local economy. The need to develop this activity sustainably is therefore a major priority.

Many studies have shown that satellite-based earth observation imagery can provide the information timber companies need to manage forest resources and evaluate timber reserves. To investigate how multisource satellite imagery could best be adapted to the needs of logging concession holders, Spot Image undertook a pilot study in partnership with its subsidiary Spot Asia in Singapore and Borneo Pulp & Paper (BP&P), a timber company based in Sarawak.

This pilot study, initially geared to BP&P's specific needs, will give Spot Image the opportunity to develop new forestry-oriented geographic information products.

Spot panchromatic and multispectral imagery was used in combination with RADARSAT SAR images acquired from different viewing angles, and an interferometric pair of ERS SAR images. Radar data complemented Spot imagery in zones where cloud cover made acquisition of optical imagery impossible, and were merged with Spot data to yield thematic information that helped discriminate different forest types. RADARSAT SAR stereopairs also served as input for a digital elevation model (DEM) used to georeference thematic products.

Top of the page

Identifying user needs

BP&P already uses a geographic information system (GIS) to manage its forest resources and has experience with satellite-derived geographic information. Specific information needs identified with BP&P were:

• DEM,
• linear features, particularly rivers and logging tracks, in vector form,
• forest inventory,
• production yield estimation map (showing timber volumes in m3/ha).

For the pilot study, these GIS products were generated over a 30 km x 60 km test area in Sarawak projected onto BP&P's own geodetic reference system. The following data sources were used to generate the GIS products: - one XS multispectral Spot scene, - one P panchromatic Spot scene, - two ERS SAR scenes acquired in tandem mode (one day apart), - one RADARSAT stereopair (Standard 2 and 7 beam modes). Multisource satellite imagery is required because cloud cover often precludes total reliance on Spot data, and because radar data alone are insufficient to meet users' needs. Spot Asia, a joint venture of Spot Image and Swedish partner SSC-Satellitbild, distributes Spot, ERS and RADARSAT data in Southeast Asia. For this project, Spot Image France is providing Spot Asia and its customers technical support for radar products and applications, and developing new commercial applications combining Spot and radar data. Top of the page

Image preprocessing

Preprocessing of source imagery has a large bearing on the quality of the final thematic products. For example, SAR imagery undergoes complete radiometric calibration and a Bayesian filter is used to reduce speckle in multi-channel SAR images.

French digital mapping specialist Istar used radargrammetry techniques to generate a DEM from the two RADARSAT SAR standard-beam images acquired at different viewing angles (beams S2 and S7). Elevation accuracy of the DEM estimated from GPS ground readings was about 19 metres. Relief displacement with respect to true geographic location is particularly noticeable in SAR images, and also affects Spot optical imagery to a lesser extent. The DEM is therefore used to orthorectify, geocode and correct radiometry in the source imagery before generating forestry data products.

Top of the page

Network extraction and forest inventory

Forest managers and timber companies need accurate information on the forest hydrologic network and on roads and logging tracks. These features are extracted automatically at the same time from at least two images using pattern recognition techniques. Combining optical and radar data effectively highlights features that might otherwise prove difficult to detect.

For the forest inventory, the cloud-free part of Spot multispectral images (XS) was classified using indices derived from Spot and RADARSAT images to exploit thematic complementarity. These indices, obtained by merging Spot and radar data, are designed to select information, such as forest biomass, relevant to the application.

In the cloudy part of the Spot XS image, classification was performed using SAR images, interferometric coherence between the two ERS images, and the Spot panchromatic image.

These two partial classifications matched closely enough to be merged into a single final image containing six classes (map right): logging tracks, water courses and rice fields (blue); very-high-yield forests (dark green); high and medium-yield cutover (medium green); re-entered cutover and secondary forest (light green); clearcuts and very-low-yield areas (grey); bare soils and crops (orange). Preliminary checks have produced very satisfactory results. Top of the page

Estimating timber reserves

Timber volume reserves are estimated using a new technique developed and validated by Privateers NV, based on SAR/optical data fusion, a forest structure model and results of the classification (forest inventory). Imagery is calibrated using field evaluations of stand timber volumes. For this study, calibration was performed using only general mean reference values, which led to some inaccuracies in timber yield estimation. Nevertheless, a field trip showed that timber volumes were never overestimated by more than 20%.

The pilot study offered Spot Image and Spot Asia the opportunity to develop new products and services to better meet the needs forest managers. The prototype products generated by combining Spot and radar data yielded more complete and more accurate data sets. As a result, Spot Asia helped BP&P to obtain value-added products tailored to their needs and which they can integrate directly into their GIS. BP&P is currently evaluating these products in the field, providing us with valuable feedback on their accuracy and suitability to users' needs. The lessons learned here and with other timber companies will help us to define a set of requirements for a new line of products dedicated to forestry applications, which could then be adapted to other regions of the world.

Top of the page

The pilot project is being coordinated by Spot Image and Spot Asia, in partnership with:

• Privateers N. V., Netherlands, responsible for the majority of image processing and generation of thematic products.
• Istar, France, responsible for producing DEMs from RADARSAT stereopairs using radargrammetry.
• SSC-Satellitbild, Sweden, responsible for orthorectification of multisource data.
• Gamma Remote Sensing, Switzerland, responsible for interferometric processing of SAR/ERS pairs acquired in tandem mode, used to generate a coherence image.