Mapping and characterizing the vegetation types of the Democratic Republic of Congo using SPOT VEGETATION time series

doi:10.1016/j.jag.2008.08.001

International Journal of Applied Earth Observation and Geoinformation

Volume 11, Issue 1, February 2009, Pages 62-76

https://doi.org/10.1016/j.jag.2008.08.001 Get rights and content

Abstract

The need for quantitative and accurate information to characterize the state and evolution of vegetation types at a national scale is widely recognized. This type of information is crucial for the Democratic Republic of Congo, which contains the majority of the tropical forest cover of Central Africa and a large diversity of habitats. In spite of recent progress in earth observation capabilities, vegetation mapping and seasonality analysis in equatorial areas still represent an outstanding challenge owing to high cloud coverage and the extent and limited accessibility of the territory. On one hand, the use of coarse-resolution optical data is constrained by performance in the presence of cloud screening and by noise arising from the compositing process, which limits the spatial consistency of the composite and the temporal resolution. On the other hand, the use of high-resolution data suffers from heterogeneity of acquisition dates, images and interpretation from one scene to another. The objective of the present study was to propose and demonstrate a semi-automatic processing method for vegetation mapping and seasonality characterization based on temporal and spectral information from SPOT VEGETATION time series. A land cover map with 18 vegetation classes was produced using the proposed method that was fed by ecological knowledge gathered from botanists and reference documents. The floristic composition and physiognomy of each vegetation type are described using the Land Cover Classification System developed by the FAO. Moreover, the seasonality of each class is characterized on a monthly basis and the variation in different vegetation indicators is discussed from a phenological point of view. This mapping exercise delivers the first area estimates of seven different forest types, five different savannas characterized by specific seasonality behavior and two aquatic vegetation types. Finally, the result is compared to two recent land cover maps derived from coarse-resolution (GLC2000) and high-resolution imagery (Africover).

Introduction

The need for quantitative and accurate information to characterize the state and evolution of vegetation types at a national scale is widely recognized, particularly to analyze the diversity of landscapes and land cover dynamics, and to assess the degradation of habitats and better manage natural resources. This type of information is crucial for the Democratic Republic of Congo, which contains the majority of tropical forest cover of Central Africa and has a large diversity of habitats.

In spite of recent progress in earth observation capabilities, vegetation mapping and seasonality analysis in equatorial areas still represent an outstanding challenge owing to high cloud coverage and the extent and limited accessibility of the territory.

The first regional and continental land cover maps over Central Africa were derived from National Oceanic and Atmospheric Administration (NOAA) Advanced Very-High-Resolution Radiometer (AVHRR) data (Tucker, 1985, Mayaux et al., 1997, Laporte et al., 1998, Loveland et al., 1999, Hansen et al., 2000). However, the AVHRR data, originally designed for meteorological applications, have poor geometric accuracy and limited radiometric calibration (Meyer et al., 1995, Cihlar et al., 1997) that reduces the spatial resolution of the imagery and introduces spatial and temporal inconsistencies in the syntheses. Therefore, the classification process is often based on the single use of a vegetation index that is less disturbed by the noise induced by preprocessing. For instance, The International Geosphere Biosphere Programme (IGBP) land cover map (Loveland et al., 1999) was based on a multi-temporal unsupervised classification of 12-month NDVI composites. However, this approach prevents efficient discrimination between land cover classes. More dedicated sensors appeared with enhanced spatial and spectral characteristics, such as SPOT-VEGETATION (VGT) and the MODerate resolution Imaging Spectroradiometer (MODIS) on board the Terra platform, even though the spatial resolution of MODIS varies much more than that of VGT. Recent studies confirm the potential of VGT (Fritz et al., 2003, Mayaux et al., 2004) and MODIS data for producing more detailed land cover and vegetation maps (Borak et al., 2000, Zhan et al., 2000, Hansen et al., 2002, Townshend and Justice, 2002) at global and regional scales. However, the use of coarse-resolution optical data is constrained by their performance in the presence of cloud screening and by noise arising due to the compositing process, which limits the spatial consistency of the composite and the temporal resolution. Advanced BRDF correction exhibited limited performance for syntheses using MODIS (Huete et al., 2002) and VGT data (Duchemin et al., 2002, Hagolle et al., 2004, Swinnen, 2004, Vancutsem et al., 2007b). Performance in the presence of cloud screening is also still a constraint in producing cloud-free and consistent syntheses, particularly in equatorial areas (Mayaux et al., 2004, Gond et al., 2005).

Another way to produce land cover map involves the use of high-resolution data. In the framework of the Food and Agriculture Organization of the United Nations (FAO) Africover program (Latham, 2001), high-resolution data have been used to realize land cover maps of several countries in Africa, such as the DRC. These products, based on visual interpretation of Landsat images by several experts, present a great level of detail, but suffer from large inconsistencies because of heterogeneity in acquisition dates, images and interpretation from one scene to another. Moreover, this approach hardly takes into account the seasonal variation and phenologic behavior of different vegetation types.

The objective of this research was to propose and demonstrate a semi-automatic processing method for vegetation mapping and seasonality characterization based on temporal and spectral information from SPOT VGT time series. More specifically, three main issues are addressed: (i) the production of spatially very consistent syntheses for all available reflectance bands using a new compositing methodology; (ii) stratification of the country to deal with various seasonal behaviors; and (iii) the use of ecological knowledge from botanists and reference documents to improve the interpretation step and to characterize the seasonality of various vegetation types. This approach is applied to the DRC, which presents particularly difficult conditions in term of cloud coverage and accessibility.

The classification result has been assessed by comparing it with two recent land cover maps derived from coarse-resolution (GLC2000) and high-resolution imagery (Africover).

Section snippets

Data

The data set used to produce the land cover map consists of 366 daily (S1 products) SPOT VGT images of the year 2000 at a spatial resolution of 1 km. The VGT instrument was launched on board SPOT-4 in March 1998 and delivers measurements that are specifically tailored to monitor land surface parameters on a global basis (http://www.spot-vegetation.com/). The four spectral bands of the sensor allow characterization of the main features of the plant canopy: (i) the red band centered on the

Methodology

The land cover mapping methodology involves the four steps described in this section: (i) data preprocessing; (ii) stratification; (iii) classification; and (iv) evaluation. The three first steps are illustrated in Fig. 1.

Compositing

The annual composite (Fig. 2) is remarkable for its spatial consistency in the red, NIR and SWIR spectral bands over the entire central African region and is absent of clouds and haze. The major visible features of the dense moist forest biome are the presence of swamp forest along the rivers (in dark green in Fig. 2a) and the ribbons of secondary forest formations along the road network (in light green in Fig. 2a and d). It is also possible to distinguish rural complex areas surrounded by

Conclusions

A new land cover map of the DRC (available at http://www.uclouvain.be/enge-cartesRDC) has been produced based on data of high temporal resolution. This product provides a synoptic and consistent view of this huge territory. The resulting 18 classes and their detailed descriptions imply a high information content of the map, which is useful for national development and environmental programs.

This study demonstrates that the MC strategy provides great spatial consistency for composites and

Acknowledgements

We gratefully acknowledge Paul Bamps, Michel Schaijes and Luc Pauwels for their comments on the land cover product and for the field documents provided. We are also grateful to the Joint Research Center for providing SPOT VGT daily data within the framework of the Global Land Cover 2000 program. Finally, we would like to acknowledge the financial support of the FRIA, and Baudouin Desclee, Gregory Duveiller, Julien Radoux, Marie-Aline Wibrin, Carlos de Wasseige, and Philippe Mayaux for their

References (51)

S. Boles et al.
Land cover characterization of Temperate East Asia using multi-temporal image data of VEGETATION sensor
Remote Sensing of Environment
(2004)
P. Ceccato et al.
Designing a spectral index to estimate vegetation water content from remote sensing data (Part 1: theoretical approach)
Remote Sensing of Environment
(2002)
J. Cihlar et al.
Multitemporal, multichannel AVHRR data sets for land biosphere studies-artefacts and corrections
Remote Sensing of Environment
(1997)
B. Duchemin et al.
Normalisation of directional effects in 10-day global syntheses derived from VEGETATION/SPOT. II. Validation of an operational method on actual data sets
Remote Sensing of Environment
(2002)
B.-C. Gao
NDWI-a normalized difference water index for remote sensing of vegetation liquid water from space
Remote Sensing of Environment
(1996)
M. Hansen et al.
Towards an operational MODIS continuous field of percent tree cover algorithm: examples using AVHRR and MODIS data
Remote Sensing of Environment
(2002)
A. Huete et al.
Overview of the radiometric and biophysical performance of the MODIS vegetation indices
Remote Sensing of Environment
(2002)
J.R.G. Townshend et al.
Towards operational monitoring of terrestrial systems by moderate-resolution remote sensing
Remote Sensing of Environment
(2002)
C. Tucker
Red and photographic infrared linear combinations for monitoring vegetation
Remote Sensing of the Environment
(1979)
X. Xiao et al.
Sensitivity of vegetation indices to atmospheric aerosols: continental-scale observations in Northern Asia
Remote Sensing of the Environment
(2003)

J. Borak et al.

The use of temporal metrics for land cover change detection at course spatial scales

International Journal of Remote Sensing

(2000)

Compère, P., 1970. La carte des sols et de la vegetation du Congo belge et du Ruanda-Urundi: 25. Bas-Congo. Notice...

Bruxelles Devred, R., 1958a. La carte des sols et de la vegetation du Congo belge et du Ruanda-Urundi:10. Kwango....

Devred, R. (1958b). La vegetation foresti‘ere du Congo Belge et du Ruanda-Urundi, Bulletin de la Societe Royale...

Doumenge, C., 1990. La Conservation des Eco-Syst‘emes Forestiers du Zaïre, Gland, Switzerland: The world consevation...

ESC, 2004. Landsat GeoCover 2000ETM+ and 1990TM Edition Mosaics, US Geological Survey, Sioux Falls, South...

Evrard, C., 1960. La carte des sols et de la vegetation du Congo belge et du Ruanda-Urundi: 11. Oubangui. Notice...

Evrard, C., 1968. Recherches ecologiques sur le peuplement forestier des sols hydromorphes de la Cuvette centrale...

FAO, 1993. Forest Resources Assessment 1990: Tropical Countries. Food and Agriculture Organisation of the United...

Fritz, S., Bartholomé, E., Belward, A., 2003. Harmonisation, mosaicing and production of the Global Land Cover 2000...

Germain, R., 1965. Les biotopes alluvionnaires herbeux et les savanes intercalaires du Congo équatorial, Acad. Roy. Sc....

Gond, V., Bernard, J., Brognoli, C., Brunaux, O., Coppel, A., Demenois, J., Engel, J., Galarraga, D., Gaucher, P.,...

Gregorio, A.D., Jansen, L., 2000. Land Cover Classification System, Classification Concepts and User Manual. Food and...

O. Hagolle et al.

Quality assessment and improvement of temporally composited products of remotely sensed imagery by combination of VEGETATION 1 and 2

International Journal of Remote Sensing

(2004)

M. Hansen et al.

Global land cover classification at 1 km spatial resolution using a classification tree approach

International Journal of Remote Sensing

(2000)

Cited by (43)

Integrating multiple land cover maps through a multi-criteria analysis to improve agricultural monitoring in Africa
2020, International Journal of Applied Earth Observation and Geoinformation
Monitoring agricultural land cover is highly relevant for global early warning systems such as ASAP (Anomaly hot Spots of Agricultural Production), because it represents the basis for detecting production deficits in food security assessment. Given the significant inconsistencies among existing land cover datasets, there is a need to obtain a more accurate representation of the spatial distribution and extent of agricultural area in Africa. In this research, we explore a fusion approach that combines the strength of individual datasets and minimises their limitations. Specifically, a semi-automatic method is developed, relying on multi-criteria analysis (MCA) complemented with manual fine-tuning using the best-rated datasets, to generate two hybrid and static agricultural masks – one for cropland and another for grassland. Following a comprehensive selection of land cover maps, each dataset is evaluated at country level according to five criteria: timeliness, spatial resolution, comparison with FAO statistics, accuracy assessment and expert evaluation. A sensitivity analysis is performed, based on an evaluation of the impact of weight settings on the resulting land cover. The proposed methodology is capable of improving agricultural characterisation in Africa. As a result, two static masks at 250 m spatial resolution for the nominal year 2016 are provided.
Integrating age in the detection and mapping of incongruous patches in coffee (Coffea arabica) plantations using multi-temporal Landsat 8 NDVI anomalies
2017, International Journal of Applied Earth Observation and Geoinformation
Citation Excerpt :
Consequently, attempts to apply course-resolution time series NDVI anomalies to map perennial tree crops, such as coffee have not been as widespread and as successful as in annual crops. For example, Vancutsem et al. (2009) using monthly SPOT Vegetation composite data in mapping perennial croplands in tropical Africa, observed incongruent results with those from reference datasets. In a separate study, Stibig et al. (2007), used the same data to map expanding coffee plantations in southern Asia and concluded that results were not consistent as coffee areas were rather mapped as shrubland.
The development of cost-effective, reliable and easy to implement crop condition monitoring methods is urgently required for perennial tree crops such as coffee (Coffea arabica), as they are grown over large areas and represent long term and higher levels of investment. These monitoring methods are useful in identifying farm areas that experience poor crop growth, pest infestation, diseases outbreaks and/or to monitor response to management interventions. This study compares field level coffee mean NDVI and LSWI anomalies and age-adjusted coffee mean NDVI and LSWI anomalies in identifying and mapping incongruous patches across perennial coffee plantations. To achieve this objective, we first derived deviation of coffee pixels from the global coffee mean NDVI and LSWI values of nine sequential Landsat 8 OLI image scenes. We then evaluated the influence of coffee age class (young, mature and old) on Landsat-scale NDVI and LSWI values using a one-way ANOVA and since results showed significant differences, we adjusted NDVI and LSWI anomalies for age-class. We then used the cumulative inverse distribution function (α ≤ 0.05) to identify fields and within field areas with excessive deviation of NDVI and LSWI from the global and the age-expected mean for each of the Landsat 8 OLI scene dates spanning three seasons. Results from accuracy assessment indicated that it was possible to separate incongruous and healthy patches using these anomalies and that using NDVI performed better than using LSWI for both global and age-adjusted mean anomalies. Using the age-adjusted anomalies performed better in separating incongruous and healthy patches than using the global mean for both NDVI (Overall accuracy = 80.9% and 68.1% respectively) and for LSWI (Overall accuracy = 68.1% and 48.9% respectively). When applied to other Landsat 8 OLI scenes, the results showed that the proportions of coffee fields that were modelled incongruent decreased with time for the young age category and while it increased for the mature and old age classes with time. We concluded that the method could be useful for the identification of anomalous patches using Landsat scale time series data to monitor large coffee plantations and provide an indication of areas requiring particular field attention.
Kamala tree as an indicator of the presence of Asian elephants during the dry season in the Shivalik landscape of northwestern India
2016, Ecological Indicators
The availability of forage resources during the dry season is often a critical factor in determining the distribution and movement of large herbivores. It has long been suspected that the Kamala tree (Mallotus philippinensis) can serve as an indicator of the distribution for Asian elephants during the dry season in northwestern India. However, there is little direct evidence in support of this speculation, especially at a large landscape scale. Here, we predicted the distribution of Kamala trees in the Shivalik landscape of northwestern India based on topographic and bioclimatic variables, as well as satellite-derived vegetation indices and forest canopy height data using a presence-only ecological niche model. We used the area under the receiver operating characteristic curve (AUC) and true skill statistic (TSS) to validate the model. We then examined the relationship between the occurrence probability of Kamala trees and the presence of Asian elephants with data collected during dry seasons between 2010 and 2014 using logistic regression models. Our results showed that the probability of occurrence of Kamala trees was predicted with good accuracy (AUC = 0.88 and TSS = 0.51). The logistic regression models showed that the presence of Asian elephants can be adequately predicted by the occurrence probability of Kamala trees. This result suggests that the distribution of Kamala trees is a good indicator of the presence of Asian elephants during the dry season in the Shivalik landscape. These findings may have major implications for the conservation of Asian elephants, especially in designing wildlife corridors and mitigating human-elephant conflicts.
Decrease of L-band SAR backscatter with biomass of dense forests
2015, Remote Sensing of Environment
Synthetic aperture radar (SAR) is one of the most promising remote sensors to map forest carbon. The unique spaceborne and long-wavelength SAR data currently available are L-band data, but their relationship with forest biomass is still controversial, particularly for high biomass values. While many studies assume a complete loss of sensitivity above a saturation point, typically around 100 t.ha^− 1, others assume a continuous positive correlation between SAR backscatter and biomass. The objective of this paper is to revisit the relationship between L-band SAR backscatter and dense tropical forest biomass for a large range of biomass values using both theoretical and experimental approaches. Both approaches revealed that after reaching a maximum value, SAR backscatter correlates negatively with forest biomass. This phenomenon is interpreted as a signal attenuation from the forest canopy as the canopy becomes denser with increasing biomass. This result has strong implications for L-band vegetation mapping because it can lead to a greater-than-expected under-estimation of biomass. The consequences for L-band biomass mapping are illustrated, and solutions are proposed.
Satellite-derived vegetation indices contribute significantly to the prediction of epiphyllous liverworts
2014, Ecological Indicators
Citation Excerpt :
The Normalized Difference Vegetation Index [NDVI, equal to (λNIR − λRED)/(λNIR + λRED)] is often used as an indicator reflecting the vigor of green land cover (Carlson and Ripley, 1997; Gamon et al., 1995). The temporal behavior of NDVI has proven to be valuable for mapping different forest types, due to its ability to capture seasonal differences in phenology and greenness by vegetation type (Chen and Pan, 2002; Vancutsem et al., 2009; Xiao et al., 2002). In this study, atmospherically corrected SPOT4 and SPOT5 Vegetation (VGT) Sensor 10-day composite NDVI-images at a 1 km resolution over the three year period from 2008 to 2010 were obtained from www.vgt.vito.be (108 images, three images per month).
Epiphyllous liverworts form a special group of bryophytes that primarily grow on leaves of understory vascular plants, occurring in constantly moist and warm evergreen forest in tropical and subtropical regions. They are very sensitive to climate change and environmental pollution. Previous studies have focused largely on microhabitat preferences of epiphyllous liverworts and demonstrated the importance of climate factors such as humidity, temperature and light. However, little is known about the relationship between distribution of epiphyllous liverworts and macro-habitat factors at broad spatial scales. Here, we predicated the distribution of epiphyllous liverworts in China based on topographic and bioclimatic variables, as well as satellite-derived vegetation indices at a 1 km spatial resolution using presence-only ecological niche models. We used the Area Under the receiver operating characteristic Curve (AUC) and True Skill Statistic (TSS) to validate the models, and then used the Wilcoxon paired test to compare model performances. Furthermore, we applied the jackknife test to identify the important factors affecting predictions. Our results showed that the highest accuracy (i.e., AUC = 0.98 and TSS = 0.93) in predicting epiphyllous liverworts was achieved by the model that combined climatic and remotely sensed vegetation variables. The satellite-derived annual mean and minimum Normalized Difference Vegetation Index (NDVI) as well as the annual mean and minimum Normalized Difference Water Index (NDWI) emerged as the most important predictors of distribution patterns of epiphyllous liverworts, while climatic variables such as precipitation in the wettest quarter and temperature of the coldest quarter were of ancillary importance. The significant contributions of NDVI and NDWI in defining the distribution range and spatial patterns of epiphyllous liverworts, and the strong relationship between this species and evergreen forest implies that epiphyllous liverworts may be a useful indicator for forest degradation or integrity at broad spatial scales.
A near real-time water surface detection method based on HSV transformation of MODIS multi-Spectral time series data
2014, Remote Sensing of Environment
Citation Excerpt :
This compositing method offers two advantages: (i) to reduce both the bi-directional reflectance distribution function (BRDF) effects and the possible remaining perturbation after atmospheric correction and cloud removal, and (ii) to reduce the computing time. The method has been used in various land cover mapping projects such as the Global Land Cover 2000 (GLC2000) (Bartholomé & Belward, 2005; Mayaux, Bartholomé, Fritz, & Belward, 2004) and the GlobCover project (Arino, Kalogirou, Perez Jose, & Pinnock, 2011; Defourny et al., 2009), for land cover monitoring (Pekel et al., 2011; Vancutsem et al., 2009) as well as to derive seasonal indicators for spatial prediction of food vulnerability (Marinho & Gerard, 2008). The daily SPOT VGT surface reflectance images are composited using the same compositing methodology over the same “10-day periods” as used for processing MODIS time series.
In the face of global population growth and the uneven distribution of water supply, a better knowledge of the spatial and temporal distribution of surface water resources is critical. Remote sensing provides a synoptic view of ongoing processes, which addresses the intricate nature of water surfaces and allows an assessment of the pressures placed on aquatic ecosystems. However, the main challenge in identifying water surfaces from remotely sensed data is the high variability of spectral signatures, both in space and time. In the last 10 years only a few operational methods have been proposed to map or monitor surface water at continental or global scale, and each of them show limitations. The objectives of this study are to develop, demonstrate and validate the adequacy of a generic multi-temporal and multi-spectral image analysis method to detect water surfaces automatically, and to monitor them in near real-time at the African continental scale as a first step towards global scale coverage. The proposed approach, based on a transformation of the RGB color space into HSV, provides dynamic information at the continental scale. Two different validations were done at the continental scale over Africa: i) The algorithm validation checked the ability of the proposed algorithm to perform as effectively as human interpretation of the image: it showed an accuracy of 96.6% and no commission errors. ii) The product validation was carried out by using an independent dataset derived from high resolution imagery: the continental permanent water surface product showed an accuracy of 91.5% and few commission errors. Potential applications of the proposed method have been identified and discussed. The methodology that has been developed is generic: it can be applied to sensors with similar bands with good reliability, and minimal effort. Moreover, this experiment at the African continental scale showed that the methodology is efficient for a large range of environmental conditions. Additional preliminary tests over other continents indicate that the proposed methodology could also be applied at the global scale without too many difficulties.

View all citing articles on Scopus

View full text

Mapping and characterizing the vegetation types of the Democratic Republic of Congo using SPOT VEGETATION time series

Abstract

Introduction

Section snippets

Data

Methodology

Compositing

Conclusions

Acknowledgements

Remote Sensing of Environment

Remote Sensing of Environment

Remote Sensing of Environment

Remote Sensing of Environment

Remote Sensing of Environment

Remote Sensing of Environment

Remote Sensing of Environment

Remote Sensing of Environment

Remote Sensing of the Environment

Remote Sensing of the Environment

The use of temporal metrics for land cover change detection at course spatial scales

International Journal of Remote Sensing

Quality assessment and improvement of temporally composited products of remotely sensed imagery by combination of VEGETATION 1 and 2

International Journal of Remote Sensing

Global land cover classification at 1 km spatial resolution using a classification tree approach

International Journal of Remote Sensing