Exploring the synergy between environmental geo-referenced field data and remote sensing spatial information in Central Africa

by Pascale Janvier, Philippe Mayaux

How to combine the CyberTracker with the TREES information system.

The TREES project

A European initiative for pan tropical deforestation monitoring of the Joint Research Centre

A large-scale project initiated in the early 1990s, the TREES (Tropical Ecosystem Environment observations by Satellite) project is dedicated to the development of techniques for global tropical forest inventory using satellite imagery. The TREES II project finalized the development of a prototype information system for monitoring tropical forest cover at a pan-tropical scale. The work was being progressively adapted to meet the forest related information needs of the services of the European Commission. In particular efforts were made with respect to the European Development Fund (EDF) regional ECOFAC project in Central Africa.

The project released new tropical forest maps derived from coarse spatial resolution satellite images (Fig. 1). After the publication of the Central Africa map in 1998, the Vegetation map of South America at 1:5.000.000 scale was published in September 1999. These maps are derived from NOAA-AVHRR 1km satellite data acquired in 1992/1993. Those continental maps are currently updated using SPOT-4 VEGETATION data in the framework of the Global Land Cover 2000 project, a component of the Millenium Ecosystem Assessment.

The objective of the maps is to document the dense humid forest extent (both lowland and upland) at a continental scale. The satellite data were used to map three broad classes of vegetation - dense humid forest, fragmented forest and non-forest. In areas of rapid change the TREES map gives a relatively up-to-date view of the forest cover, exhibiting a higher spatial accuracy and precision than conventional maps. A stratification of "deforestation hot spot areas" was performed based on individual expertise. Forest cover change during the period 1990-1997 was then estimated from 100 sampled sites using high-resolution optical remote sensing satellite imagery (Fig. 2).

Fig 1
Fig. 1- Mosaic of satellite images over Central Africa

Fig 2
Fig. 2- High resolution image

The TREES project uses also coarse resolution satellite imagery for the day-to-day monitoring of forest condition. The project used thousands of scenes acquired from the ATSR sensor (on board ESA's ERS-2 satellite) and from the VEGETATION sensor (on board SPOT-4 satellite). The evaluation of these data shows the potential for the updating of existing vegetation maps at regional scale and for identifying active deforestation areas. These methods could now be integrated into an operational analysis and monitoring system.

As well as paper maps, the TREES project provides access to the information in digital form via a bespoke GIS, the Tropical Forest Information System (TFIS).

The TFIS is designed to be:

An archive of the TREES project products and results; including satellite classifications, GIS layers and non-spatial informationA tool for query, browsing and visualization of data in the archiveAn analysis tool bringing together diverse datasetsA basis for modelling deforestation processes.

It is comprised of four components - a metadata database (Fig. 3), a spatial database, an a-spatial database, and GIS software. The system is implemented using standard software products in MS Windows NT environment with GIS ArcInfo to process vector data sets, Erdas Imagine for the satellite image, MS Access to design the different databases (TFIS, ECOFAC, IMAGES) and metadata catalogue and ArcView to query, browse and visualize data sets.

Fig 03
Fig. 3 - TFIS Metadata Browser

TFIS allows the users to analyze the tropical forest maps in conjunction with other spatial data sets such as elevation, other forest maps (e.g. IUCN, WCMC) and non-spatial data. A dedicated TFIS for the Central African ECOFAC project was installed in Libreville. The project supported an on-site TFIS within the Directorate General for Environment in Brussels. A strategy was developed to provide an interface available to a wider numbers of partners: An Internet-TFIS prototype was developed.

The collaboration TREES-ECOFAC

Since 1997, the TREES project is collaborates with the ECOFAC programme, also funded by the European Union, working on similar areas, but at different scales and with different techniques. In other words, ECOFAC had accumulated an enormous amount of field knowledge which was missing to TREES, and TREES had a expertise in the set-up spatial databases which was important for ECOFAC. Moreover, the TREES project produced information about the forest cover in a smaller scale but in a quite exhaustive manner on the 6 continental countries of Central Africa.The respective scales of the TREES and ECOFAC projects are complementary. The collaboration is aiming at the improvement of the management of the spatial data in order to optimise the field patrols and to increase the understanding of the animal migrations. The field observations, collected by the CyberTracker systemare transferred into a relational database and integrated in a Geographical Information System for the spatial analysis. The TREES Project is in charge of the set-up of the database and the Geographical Information System, the vegetation mapping and the archiving of spatial data generated by ECOFAC. On the other hand, ECOFAC contributes to the validation of the interpretation of satellite datasets and a better understanding of the biodiversity of the region.

Collecting field data for validation of map derived from satellite imagery:

Reliable analysis of the satellite radar images comes up against of the lack of appropriate field observations. Effectively, the useful parameters for radar signal modelling are quite specific and are not available in the ecologic literature. It has been decided to collect series of parameters about the different types of forest such as open forest, closed forest, or swamp forest and also the description of the environment on the Odzala site. (Odazla National Park – Congo) These observations will allow to verify the assumption announced during the analysis. An exhaustive list of different types of vegetation has been integrated within the CyberTracker.

Using the CyberTracker in the TREES project

Integration of the field data to the GIS

The main aims of this study described below are to demonstrate the benefits of integrating the results of the CyberTracker into a Relational Data Base and into a GIS, to explore the possibilities for spatial analysis (Fig. 4). The potentials provided by the GIS for selecting and displaying the results according to the location, time or other data attributes are known to be highly useful. The methodology developed is summarised below:

Step1: Exporting from CyberTracker

Creation of tables with attributes containing acquisition time and lat/long position and export in *.dbf or *.xls format readable in MS Access.Step 2: Storage in Relational Data Base

In MS Access: Sort out, data harmonization and queries for pre-analysis and to permit linkages with GISStep 3:Spatial Analysis using GIS

In ArcView: SQL Connect, Queries, New tables, Mapping, Overlaying and Spatial analysis.

Fig 4

Data cleaning and harmonization

The initial table called "cyber Observation" is created from the CyberTracker database and contains the complete attributes of the cyber such as lat/long, observer, displacement, activity, and index. It is imported in a rough manner with all the errors contained in the original database.In this table, the same observation can be described in different manner, with capitals or not, a mix of capitals and minuscule, with accent or not. For the animal species, the description can correspond to one or many fields. For example, an elephant corresponds to only one field, while a "céphalophe à dos jaune" is first qualified as "antilope", then "Cephalophe" and finally "Cephalophe a dos jaune". This is a problem for an accurate analysis (Fig. 5).

Fig 05
Fig. 5 - List fauna table in MS Access

To facilitate the analysis, and to produce a consistent Data Base, a methodology (Fig. 6) has been implemented to harmonize the data by using cleaning functions and queries within MS Access. This step will improve by the quality of the data.

Fig 06
Fig. 6 - Design of the methodology developed in MS Access

In MS Access, a set of tables has been designed, according to the different types of observations such as fauna, human and vegetation. These tables were populated with the cleaned descriptions. These tables are used for the data harmonization processing and could be used, in the future, for the translation from English to French, and for retrieving the scientific name of species. A set of predefined queries has been already implemented regarding the list tables. The cleaning process is carried out using both the lists and queries and is creating a new table called "new cyber observation" (Fig. 7).

Fig 07
Fig. 7- Result of the cleaning function

Development of thematic queries

A set of thematic queries has been developed using MS Access macro commands. Running these macro commands automatically generates new tables which are a selection of records related to some specific phenomena for the spatial analysis. A set of tables has already been created such as the table of the different species, the calculation of distances using Euclidian distance formula, the densities per quadrant, the distance covered per quadrant and by month, etc.For each species a set of queries has been designed (e.g. elephant):lephant_all_observationsElephant_fresh_droppingElephant_fresh_tracksElephant_indice d'abondance (number of fresh droppings per km of patrol)Elephant_indice d'abondance (number of fresh tracks per km of patrol).

Loading tabular data from the databasesTo query, to visualize and for basic spatial analysis, the database is connected to a GIS ArcView (AV). Based on AV's SQL connection feature, a connection has been established to the database server, (MS Access), a SQL query allow retrieving requested records from the database. The retrieved records become a new table in the project. It is possible to develop some other specific queries (Fig. 8) for creating and editing spatial data.

Fig 08
Fig. 8 – SQL Query to create new tables

Visualization and spatial analysis

Our awareness of the complexities of the world increases our desire to understand the nature of the spatial data and spatial pattern. The role of visualization in geographical analysis is not limited to maps and remotely sensed imagery, but extends to numeric and statistical analysis as well. A series of exercises has been done to illustrate these talks. In the following example (Fig. 9), we show, for each quadrant, the frequency of the patrols during one year overlaid with the number of elephant dropping observations over a year period. From such a display, it is possible to establish that the number of dropping observations is spatially correlated with the patrol effort.

Fig 09
Fig. 9 –Spatial analysis

Forest mapping

Accurate estimation of tropical forest cover is arguably the most important scientific technical aspect of TREES. The TREES project aims to estimating forest cover using satellite images, 1km resolution. Because of the coarse spatial resolution of these images the problem of spatial aggregation exists. Improvement of the forest estimation is achieved by using the high images (TM), 30m resolution, (Fig.10) to calibrate the coarse resolution classification. At the regional scale, TM was used by the TREES project as the reference training dataset. ECOFAC was also a beneficiary of this expertise in the interpretation of high spatial resolution satellite images (SPOT, Landsat).

Fig 10
Fig. 10 – High resolution image on Odzala National Park

Fig 11
Fig. 11- Overlaid with the CyberTracker vegetation observation

Indeed, we produced a vegetation map for the national park of Odzala (Congo) derived from Landsat ETM data. The field observations collected by CyberTracker represented a help for image geo-referencing and interpretation and for map validation at a level never achieved before. The two Landsat 7 ETM images were classified into 9 classes (Fig. 12). A part of the Cyber observations was used during the interpretation, the rest for the validation (Fig. 13).

It must be underlined that the satellite technologies participate from two sides to a better management of the protected areas : with the global positioning systems for geo-referencing the field data (CyberTracker) and the Earth observation images for mapping (Landsat).

Fig 12
Fig. 12 – Vegetation map

Fig 13
Fig. 13 - Discrepancy between the TREES map and the CyberTracker observations


In the future, we envisage to improve the following points:

The classifications can be refined using new sensors, as Ikonos, 1m spatial resolution, (Fig. 14) and integrate automatically the CyberTracker observations in the process of classificationA characterization of the park appropriate to the management and to the ecological analysis (percentage of different type of vegetation per quadrant, distance to villages...)A better definition of the required accuracyA more rigorous calculation of the distance (Euclidian distance formula)An optimisation of the frequency of the data analysis. Monthly, yearlyA stability in the structure of the CyberTracker (need for the relational database)

Fig 14
Fig. 14 - Ikonos Image.

JRC-TREES project GVM, Unit TP 440 I-21020, Ispra, Italy

This email address is being protected from spambots. You need JavaScript enabled to view it. ,   This email address is being protected from spambots. You need JavaScript enabled to view it.


Achard F., Eva H., Glinni A., Mayaux P., Richards T., Stibig H.J., 1997, Identification of deforestation hot spot areas in the humid Tropics, TREES publication series B, Research report no 4., EUR 18079 EN, European Commission, Luxembourg, 100 p.

Eva, H.D., G., Glinni, A., Janvier, P., and Blair-Myers, C., 1999, Vegetation Map of South America, Scale 1/5M, TREES Publications Series D, N°2, European Commission, Luxembourg, EUR 18658.

Mayaux, P., Janodet E., Blair-Myers, C. and P. Janvier, 1997, Vegetation Map of Central Africa at 1:5,000,000, TREES Series D: Thematic output No 1, EUR 17322 EN.

Mayaux, P., De Grandi, G.F., and Malingreau, J.P., 2000, Central Africa forest cover revisited: a new approach based on a multi-satellite analysis, Remote Sensing of Environment, 71:183-196.

Richards T.S., J. Gallego, and F. Achard, 2000, Sampling for forest cover change assessment at the pan-tropical scale, Int. J. Remote Sensing, 21 (6&7):1473-1490

Stibig H.-J., F. Achard , H. Eva , P. Mayaux, P. Janvier and T. Richards, 2000, Monitoring of Tropical Forest Cover at global and regional scale using Earth Observation Satellite Data, XXI IUFRO World Congress Kuala Lumpur, Malaysia 7-12 August 2000.