Study Of The Zaïrian Tropical Forest: Mapping Of The Vegetation Types And Understanding Of The Local Factors Of Change
|
|
Study Of The Zaïrian Tropical Forest: Mapping Of The Vegetation Types And Understanding Of The Local Factors Of Change |
M. Massart, S. Eloy, M. Sintzoff and J. Wilmet, Remote Sensing Laboratory
University Of Louvain , Place L. Pasteur, 3-B-1348 Louvain-la-Neuve, tel: 32(0)10-472870/472873, fax: 32(0)10-472877, telex: 59037 UCLB.
ABSTRACT
The main remote sensing contribution to the study of Zaïrian humid tropical forest was up to now the global forest map carried out by NASA with AVHRR LAC data.
In order to improve the global results already achieved, the present study was carried out at a local scale. Apart from the comparison between the local and the global results, the particular objectives of this study were to evaluate the evolution of the forest cover and to assess the factors responsible for the changes. To achieve these goals, four test sites located at the southern border of the zaïrian humid tropical forest were studied with Landsat MSS data and aerial photographs of the fifties.
The local land cover was mapped at two or three selected periods for each test site. The land cover evolution for periods of more than 30 years and the explanatory factors were thus highlighted. The mean annual deforestation rates were calculated for each test site. Finally, the relation between the roads, population densities and forest degradation were analyzed for some of the test sites.
The relation between the AVHRR forest map and the Landsat classifications was analyzed in order to calibrate the former and thus to improve the accuracy of the low resolution forest results. This calibration was studied at the scene scale and for a segment sampling in order to derive a regression estimator. Fragmentation of the forest cover was recognized as having a main impact on the AVHRR forest estimates. In this context, various tools were used to characterize and to define the limitations of the AVHRR product. Attention focused also on the definition of an aggregation level at which the AVHRR map can be used.
1. INTRODUCTION This study is in keeping with the general patterns of the international scientific efforts to protect the forest environment and understand the global factors of climatic changes. In this context, different programs have been implemented to map the tropical forest at a global scale.
In Africa, most of the humid tropical forest covers the Republic of Zaïre. A map of this forest was drawn up by NASA from low resolution satellite data (AVHRR-LAC data). It shows the extension of the humid tropical forest in Zaïre since 1990, as well as the degraded areas, savanna and mixed areas. But research also needs to be conducted on a regional or local scale in order to evaluate the evolution of land cover and to assess the factors responsible for change. An integrated study over a long period is necessary to assess and understand them. Aerial photos taken in the fifties and remotely sensed data from 1973 to 1989 were therefore interpreted in order to map the land cover.
The study zone is located in the Guinean region, at the southern limit of the Zaïrian tropical forest. Four test sites (Landsat MSS Scene) have been selected : Luisa, Kananga, Dimbelenge in Kasaï region and Kongolo in Shaba region.
For each test site, the objectives of the study were at first to map the vegetation types for the three or four selected dates, to compare these results to the global low resolution map achieved by NASA, to map and to measure the development of the principal vegetation types from the fifties to now and at last to seek for explanatory factors for this evolution.
2. CLASSIFICATION Classifications were performed on georeferenced (in lat/long, using the IGCB planimetric maps at 1 : 200 000 scale) Landsat MSS data for the mapping of land cover. In order to synthesize the information and destripe the Landsat MSS data, three bands were used for the classification : the first and the second principal components calculated on the raw bands and the NDVI, all three standardized. The classifications are based on an unsupervised method and use an ISODATA algorithm (sequential clustering). This method allows an "a posteriori" labeling of the classes (computer aided photointerpretation) and a continuous control of the classification process and results.(click here for pictures) The definition of the three resulting vegetation classes was based on the interpretation of the photomosaïcs and on the previous classifications of Zaïrian satellite data. These three classes are : "closed forest", "degraded forest/agricultural domain complex" and "savanna". A measurement of accuracy of the classification results is not relevant without any ground survey or without auxiliary vegetation data. The results seem only fairly good but will undergo another control through the consistency study of the change detection results.
Classification results for the four test sites are summarized in Tables 1 to 4.
1954 1975 1984 Closed Forest 1271 (39 %) 1170 (36 %) 1099 (34 %) No Forest 1997 (61 %) 2099 (64 %) 2170 (66 %)
1954 1975 1984 1989 Closed Forest 14234 (49 %) 12250 (42 %) 12144 (42 %) 11408 (40 %) No Forest 14522 (51 %) 16506 (58 %) 16612 (58 %) 17348 (60 %)
1975 1986 Closed Forest 8451 (34 %) 8369 (34 %) No Forest 16455 (66 %) 16537 (66 %)
Tab 1 to 4. Extent of the closed forest for the sites of Luisa, Kananga, Dimbelenge and Kongolo, in Km2 and % total area.
1957 1973 1984 1989 Closed Forest 1896 (24 %) 2216 (28 %) 1712 (22 %) 1264 (16 %) No Forest 6104 (76 %) 5784 (72 %) 6208 (78 %) 6736 (84 %)
3. CHANGE DETECTION A post classification method was performed, based on a two by two comparison for the two or three time periods over the last twenty years.(click here for pictures) Before producing the results, a logical analysis allows us to eliminate the unjustified change classes (ex. a savanna evolving into a closed forest in nine years). For the changes between the fifties (analogical photos) and the seventies, digital analysis was used for the test site of Kongolo (photointerpretation and digitizing) (click here for pictures) and a sampling method was performed for the other test sites in order to reduce processing time and the digitizing problems in a high fragmented forest landscape.
The mean annual rates of changes of the closed forest class (forest shrinkage) for the four test sites are presented in Table 5.
Tab 5. Mean Annual Rate of closed forest recession for the 4 test sites (%/year)
Mean Annual Rate Luisa (1954 - 1984) - 0.48 Kananga (1954 - 1989) - 0.63 Dimbelenge (1975 - 1986) - 0.08 Kongolo (1957 - 1989) - 1.26
4. FACTORS OF CHANGE The relation between roads and forest degradation was analyzed for the different test sites and for the selected time periods.
For "Luisa" and "Kananga", we observe that the patterns of closed forest degradation are located near the roads.(click here for pictures) The deforestation seems to relate to shifting cultivation activities and firewood exploitation, in connection with population density. SPOT data has been used to estimate the cultivated areas but is unfortunately difficult. This discrimination is closely related to the data acquisition period. The composite radiance values of the fields (cultivated species in association), the lack of clear limits for the fields and the trees left in the fields for agronomical reasons are other factors responsible of the poor discrimination of the fields. The relation between deforestation rates and population density increase for the sites of "Kananga" and "Luisa" are presented in Table 6. The mean annual rates of closed forest change are similar for both periods and both test sites although the initial conditions (population density and amount of forest areas) are quite different between the test sites.
For "Dimbelenge", very low deforestation rates were observed, compared to the population growth. This can be explained by the high pedological quality of the soils and the diamond exploitation of the region (Mbuji-Mayi). High population growth is not a farming population growth.
For "Kongolo", the deforestation areas are located along roads, close to the distribution centers of an agricultural development project. Deforestation results from population growth but also from the increase of cultivated areas per household, just the opposite of the situation at the Dimbelenge test site.
From the above figures, it can be seen that the deforestation rates differ from one site to another and from one period to another. These rates are related to various factors. Population growth is one of the main factors but local and initial environmental and human conditions can be of particular importance. If we just consider the location of the deforestation areas, the proximity of the roads is again the classical prevailing factor.
Tab 6. Mean annual rates of Population growth
Mean Annual Rate (Per Year) Luisa 2.57 Kananga 2.06 Dimbelenge 4.45 Kongolo 2.58
5. GLOBAL SCALE STUDY Validation and calibration studies of the AVHRR Forest map illustrate through various tools the quality and the limitation of AVHRR data for forest mapping and estimation. Attention has been focused on the definition of an aggregation level at which an AVHRR Map can be used.
Calibration of the AVHRR Map with the Landsat MSS classification was carried out for different test sites. In the calibration procedure, the importance of a good stratification based on a fragmentation measure has been recognized. In this context, the MSS fragmentation measure represents the best solution. Unfortunately, for stratification, this measure is only available at the sampling level. An operational approach would be the use of the AVHRR fragmentation measure considering that both are related. Their relation is unfortunately not linear. A modeling approach seems necessary to perfectly understand their common variations.
A general summary of the work performed under this section is given in Figure 1.
6. CLOSING REMARKS A thorough study of explanatory variables of forest change cannot be undertaken due to the lack of local information. Indeed, at a local scale (Landsat MSS scene size for Zaïre), a combination of human (social behavior, population growth, proximity to roads, ...) and physical factors (relief, soil fertility, ...) explains the patterns of change. Many data are not available at this local scale. Especially, one of the most important data (social behavior) required for a relevant analysis of forest development is not available. This information is often neglected or taken into account with difficulties but it seems that it plays a prominent part in the deforestation process.
A ground survey would enable more information to be gathered about the factors of forest change. Such a survey is also necessary to assess the classification method applied in this research. The photomosaïcs provided us with a framework for the satellite data interpretation. However, it would be very interesting to have an interpretation of the most recent satellite images based on a ground survey in order to compare it with the one based on photomosaïcs. A more favourable political context in the Republic of Zaïre will, without any doubt, enable us to evaluate this research and complete the study of the explanatory factors of forest change.
Fig. 1. Summary of the global scale comparison study. (a) Composition of a mean AVHRR forest pixel in each test site ; (b) Relationship between the AVHRR forest estimation and the Landsat MSS forest estimation based on a sampling of 25 segments for the test sites of Dimbelenge and Kananga ; (c) Study of the fragmentation evolution with different resolution cells and the fragmentation effects on the forest estimation for different forest fragmentation patterns ; (d) Sensitivity of the regression coefficient with the size of the sampling units in the different test sites ; (e) Effects of the aggregation level on the forest estimation errors for the AVHRR forest map.
REFERENCES Beguin, H. (1960). La mise en valeur agricole du Sud-Est du Kasaï, INEAC, Bruxelles, 289 p.
Belward A. (1992). Spatial Attributes of AVHRR imagery for environmental monitoring. Int J. Remote Sensing, Vol 13, no 2, pp 193-208.
Belward A. and Lambin E. (1990). Limitations to the identification of spatial structures from AVHRR data. Int. J. Remote Sensing, Vol 11, no 5, pp 921-927.
BSP (1993). Central Africa : Globate climate change and development - Technical report. Biodiversity Support Program (WWF, WRI and the Nature Conservancy), Washington, 182 p.
Castiaux, N., M. Massart and J. Wilmet (1991). Environmental study of tropical african urban areas by multitemporal satellite imageries (Lubumbashi in Zaïire, Central Africa), Proceedings of the International Symposium on Remote Sensing of Environment, Rio de Janeiro, Brazil.
Chambon, R. and A. Leruth (1954). Monographie des Bena Muhona. Territoire de Kongolo - District du Tanganika. Ministère des Colonies du Royaume de Belgique, Bruxelles, 80 p.
Cochran, W.G.,(1977). Sampling techniques. New-York, John Wiley & sons.
Dale, V. H., R. V. O'Neill, M. Pedlowski and F. Southworth (1993). Causes and effects of land use change in Central Rondônia, Brazil. Photogrammetric Engineering and Remote Sensing 59 (6) : 997-1005.
Devred, R. (1958). Carte de la végétation du Congo belge et du Ruanda-Urundi. Bull. de la Soc. Roy. For. de Belg. no 6.
Dwivedi, R. S. and T. Ravi sankar (1991. Monitoring shifting cultivation using space-borne multispectral and multitemporal data. International Journal of Remote Sensing 12 (3) : 427-433.
Gilruth, P. T. and C. F. Hutchinson (1990). Assessing deforestation in the Guinea highlands of West Africa using Remote Sensing. Photogrammetric Engineering and Remote Sensing56 (10) : 1375 - 1382.
Goossens, R., T. Ongena, E. D'Haluin and G. Larnoe (1993). The use of remote sensing (SPOT) for the survey of ecological patterns, applied to two different ecosystems in Belgium and Zaïre in Landscape ecology and GIS, edited by Haines-Young, Green and Cousins, London-New-York-Philadelphia, pp147-159.
ICIV, Scot Conseil and CIRAD (1994). Etude Méthodologique : Typologie de la fragmentation du domaine forestier tropical. Projet TREES, contrat 5396-93-07 ED ISPF, 44 p + annexes.
INS-Zaïre (1991). Zaïre, recensement scientifique de la population-Juillet 1984. Totaux définitifs. Ministère du Plan et Aménagement du Territoire, République du Zaïre, Kinshasa, 90 p.
Jeune Afrique (1978). Les atlas Jeune Afrique. République du Zaïre. Jeune Afrique, Paris, 72 p.
Kleinn C., Dees M. and Petz D. (1993). Sampling aspect in the TREES project - Global inventory of tropical forest -. TREES Final report. Universität Freiburg, Germany, 36 p + annexes.
Lambin E. (1990). Potentialités des données AVHRR pour l'étude des paysages agraires. Internal Report, Ispra.
Lee, J. K., K. P. Lulla and P. W. Mausel (1989). Data structure characterization of multispectral data using principal component and principal factor analysis. Geocarto International (2) : 43-47.
Massart, M., M. Pétillon and E. Wolff (1992). Impact of an agricultural development project on tropical forest environment - the case of Shaba (Zaïre) Photogrammetric Enginneering & Remote Sensing (in press).
Mausel, P., Y. Wu , Y. Li, E. Moran and E. S. Brondizio (1993). Spectral identification of successional stages following deforestation in the Amazon. Geocarto International(4) : 61 - 71.
McGuire, K. C. (1992). Analyst variability in labeling of unsupervised classifications, Photogrammetric Enginneering & Remote Sensing 58 (12) : 1673-1677.
Monat, D., G. Mahin and J. Lancaster (1993). Remote sensing techniques in the analysis of change detection. Geocarto International (2) : 39-50.
Moody A. and Woodcock C. (1994). Scale dependent errors in the estimation of Land-Cover proportions : Implications for Global Land-Cover Datasets. Photogrammetric Engineering & Remote Sensing, Vol 60, no 5, pp 585-594.
Nelson R. and Holben B. (1986). Identifying deforestation in Brazil using multiresolution satellite data. Int. J. Remote Sensing, Vol 7, no 3, pp 429-448.
Nelson, R., N. Horning and T. A. Stone (1987). Determining the rate of forest conversion in Mato Grosso, Brazil, using Landsat MSS and AVHRR data. International Journal of Remotte Sensing 8 (12) : 1767-1784.
Pu, R. and Y. Fang (1992). Application of remote sensing techniques to forest site survey. Geocarto International (3) : 19-24.
Richards, J. A. and D.J. Kelly (1984). On the concept of spectral class. International Journal of Remote Sensing 5 (6) : 987-991.
Sader, S., T. A. Stone and A. T. Joyce (1990). Remote sensing of tropical forests : an overview of research and applications using non-photographic sensors. Photogrammetric Engineering and Remote Sensing 56 (10) : 1343-1351.
Singh, A. (1989). Review article. Digital change detection techniques using remotely-sensed data, International Journal of Remote Sensing, 10 (6) : 989-1003.
Singh, A. (1990). Design of a global tropical forest resources assessment. Photogrammetric Engineering and Remote Sensing 56 (10) : 1353-1354.
Singh, A., (1987). Spectral separability of tropical forest cover classes. International Journal of Remote Sensing 8 (7) : 971-979.
Smit, G. S. (1978). Shifting cultivation in tropical rainforests detected from aerial photographs. ITC journal(4) : 603 - 632.
SPPS (1981). Télédétection-Zaïre. Prem ière phase. Rapport final. Contrat SPPS noTZ / A002, Laboratoire de Télédétection , Louvain-la-Neuve, 148 p.
Stibig H-J. (1993). Global Tropical Deforestation Using Satellite Observation Techniques. Report contract 4677 92 03 ED ISP D, Goettingen, Germany, 46 p + annexes.
Thomas I. L., N. P. Ching, V. M. Benning and J. A. D'Aguanno (1987). Review article : A review of multi-channel indices of class separability. International Journal of Remote Sensing 8 (3) : 331-350.
Westman, W. E., L. L. Strong and B. A. Wilcox (1989). Tropical deforestation and species endangerment : the role of remote sensing. Landscape Ecology 3 (2) : 97-109.
White, F. (1986). La végétation en Afrique. Orstom - Unesco, Paris, 384 p + cartes.
Wilmet, J. (1958). Essai d'une écologie humaine au territoire de Luisa, Kasaï, Congo belge. Bulletin de la Société Belge d'Etudes Géographiques XXVII (2) : 308-363.
Wilmet, J., M. Massart, E. Wolff et M. Pétillon (1992). Analyse des systèmes agro-pastoraux dans les régions confrontées à la dégradation de l'environnement. Deuxième phase. Rapport final. SPPS - Laboratoire de Télédétection et d'Analyse Régionale , Louvain-la-Neuve, 164 p.
 |
WHY BIODIVERSITY FIELD STORIES PARTICIPATE RESULTS MARKETSPACE ABOUT THE NETWORK WHAT'S NEW LEARNING MATERIALS SEARCH LINKS SITE MAP HOME |