Land-use changes in an agro-pastoral area (Djougou, Benin) from Landsat archive images (1984 and 2012): a regressive approach combining remote sensing and field surveys
Dynamiques de l'occupation du sol en milieu agro-pastoral dans la commune de Djougou au Bénin à partir d'images Landsat acquises entre 1984 et 2012 : une approche régressive associant télédétection et enquêtes de terrain
Résumé
Extended abstract:
The vegetation of the district of Djougou is affected by large changes due to tree logging, crop field clearing and grazing by increasing numbers of livestock. The dynamics of the landscape is analyzed with a time series of Landsat satellite images (1984-2012). A regressive approach, exploiting the field observations and semi-structured interviews conducted in 2012-2013 in the Bakou-Wewe territory, documents the evolution of land use and facilitates the analysis of Landsat images. An original procedure targets the building of Region of interest (ROIs) for the historic images for which no available field observations exists.
The study area is located in central Benin (Plate 1). Particular attention is paid to the Bakou-Wewe territory where interviews were conducted in the villages of Alheri and Sew Sewga. A collective interview was first performed in each village with the village chief and the household heads. Then individual semi-structured interviews were conducted in each village as well as outside with Fulani individuals (BarguuBe and MbororooBe) and resource persons. These interviews are analyzed to try and explain the causes of changes in land use, and assess their consequences (Robert and Gangneron, 2015; Gangneron and Robert, 2015). Both villages harbor a large diversity of populations, and the dynamics of their agropastoral systems is representative of the district dynamics. Alheri is a village of sedentary DjugureeBe crop farmers. Sew Sewga is a village of Yom crop farmers who also harbor Ditamari crop farmers. A diversity of Fulani groups are also present (BarguuBe and MbororoBe).
A time series of satellite images is realized from Landsat archive over the period 1984-2012. The outline of several field plots within the Bakou and Wewe territory was recorded by GPS in October 2012 and April 2013. Their land use was coded in seven types: dense forest, woodland, shrub and wooded savanna, fallow land, crop field, and built-up units, and burnt area (Plate 2). A photo-interpretation of the Google Earth image (2012) completes these field observations.
The creation of multi-temporal ROIs includes six steps: 1) the creation of ROIs for the current status of the land units (2012-13), 2) the supervised classification of the 2012 image using these ROIs, 3) a first pass of the supervised classification of historic images based on the spectral definition of the land-use types observed in 2012, 4) the examination of the first pass land-use classification of pixels within each ROI at each historical dates to select ROIs and an allocate them to a land-use type, 5) the random partition of the ROIs reclassified as training areas and tests areas for each date classification, 6) the second pass of the supervised classification done separately for each image based on these reclassified training areas (Plate 3).
The ROIs used to classify separately each of the seven Landsat images display some heterogeneity (Table 1). As expected, the 2012 classification that rely on ROIs established on field observations is performant. The ROIs used for the classifications of historic images are retained for each date depending on the homogeneity of the first pass classification. Three scenarios are possible: 1) if over 75% of the pixels of the ROI belong to the same land-use class as in 2012, the ROI is maintained; 2) if over 75% of the pixels are assigned to another land-use class, the ROI is assigned to this other land-use class for that date after checking if the dynamics between the two land-use is coherent with the time required for this land use shift; 3) else if less than 75% of the pixels are classified in one land-use type, the land use over the ROI is considered heterogeneous and the ROI is no retained for image classification at that date. Table 2 illustrates the homogeneity or heterogeneity for the different ROIs for the 1998 Landsat image and Table 3 shows the results for all ROIs for all dates.
The retained ROIs are then divided into two groups of equal counts: training areas and test areas (used to the confusion matrix). The final classifications, using these training areas, are performed separately for each image.
The tracking over years (1984-2012) of the land-use composition of the ROIs, informs on different land-use dynamic pathways (Figure 1).
The statistics calculated in the confusion matrices indicate satisfactory and stable (across years) quality of classifications (Table 4).
A dual reading of the land use changes is possible. First, the evolution over years of each land-use class provides general information on landscape dynamics (Table 5); and second, a spatial analysis of each date land-use map localizes the dynamic events and highlights local trends.
A synthesis on the one hand of the land-use dynamics of the region between Djougou and Parakou and on the other hand of the agricultural pioneer front (Bakou-Wewe territory) is built from the confrontation of the series of land-use maps and information gathered from the semi-structured interviews. Complementarity between the information from satellite images and interviews is essential to interpret the landscape dynamics especially over the Bakou-Wewe territory. The Plate 4 shows the land-use maps for 1984, 1998, 2003 and 2012.
The dynamics observed can be summarized in three pathways (1-3), an intermediate situation related to the expansion of built-up units (4), and three constants (5-7, Table 6):
1. Woodlands were gradually brought under cultivation.
2. Woodlands or shrub savannas cleared in the 1980’s were left in fallow for a time and returned to shrub savannas in the 2010’s.
3. Cropped fields in the 1980’s left in long fallow evolving into shrub savannas that were set again in crops in the late 2000’s.
4. Initial built-up units remain throughout the study period. They spread around urban centers and along highways.
5. Stable cropland with alternating crop and short fallow fields.
6. Stable dense forest.
7. Stable woodlands.
These dynamics events are mapped (Plate 5) allowing a precise spatial analysis of land use changes over the 1984-2012 period.
The study highlights and estimate a distinct increase in area cropped at the expense of woodlands and savannas, mostly. This is illustrated in particular by the progression of an agricultural pioneer front in the Bakou-Wewe territory following the authorization to crop in the State Forest Reserve buffer zone in 1992, and the field clearing acceleration during the 2000’s. The clearing of forest or savannas and long duration fallows are until now the two forms of soil fertility management in the area.
La commune de Djougou (Bénin) témoigne d’importantes modifications liées à l’exploitation forestière, aux défrichements agricoles et à l’extension de l’élevage. La dynamique paysagère est analysée à partir d'une série temporelle d'images-satellite Landsat (1984-2012). Une approche régressive, exploitant les missions de terrain et les entretiens semi-directifs réalisés en 2012-2013 dans la zone de Bakou-Wéwé, documente l’évolution de l’occupation du sol et aide à analyser par classification supervisée les images anciennes. Une procédure originale est développée pour établir des ROIs en l'absence d’observations de terrain antérieures à 2012.
Cette étude révèle la progression d’un front pionnier agricole dans la zone de Bakou-Wéwé suite à l'autorisation des cultures dans la zone tampon en 1992 et son accélération au cours des années 2000. La carte synthétisant les changements d'occupation du sol de la commune de Djougou (1984-2012) montre une stabilité des forêts denses et une augmentation des espaces agricoles au détriment de la forêt claire et de la savane arbustive.