Forest Cover Change in Boe region of Guinea-Bissau, West Africa

The study aimed to increase the understanding of a priority protection site for West African Chimpanzee in Boé Sector, Gabú, Guinea-Bissau, West Africa. Online, free datasets were analyzed with the focuses on forest cover change and burned area through GIS applications.

In 2003, World Conservation Union (IUCN) issued Status Survey and Conservation Action Plan – West African Chimpanzees listed the Boé Sector in Gabú, Guinea-Bissau a priority site for chimpanzee conservation (Kormos et al., 2003). As reviewed by IUCN in March 2016, Pan troglodytes ssp. verus, commonly known as Western Chimpanzee or West African Chimpanzee was listed as the “Critically Endangered”, The major threats are habitat loss and fragmentation, due to slash-and-burn agriculture, which is commonly called shifting cultivation and swidden cultivation in the literature (Mertz et al., 2009), and commercial agriculture, cashew plantations for instance (Humle et al., 2016). Vasconcelos et al. (2015) provide evidence that cashew plantations degrade biodiversity in Guinea-Bissau. Species abundance and richness have also proven to be significantly affected in the case.

The Foundation Chimbo, a Dutch NGO established in 2007 for the conservation and restoration of the chimpanzee population in West Africa hopes to take actions in conduction of survey and monitoring program in the Boé area in line with the “Regional Action Plan for the Conservation of Chimpanzees in West Africa”.

Foundation Chimbo is looking for more information regarding the general loss of the forest cover. Furthermore, the data of fire-events aids the early-fire program Foundation Chimbo has been advocating, where fires are to be set from October to December in the slash-and-burn farming. Fires on humid grass prevent uncontrollable damage to the forest area. (Foundation Chimbo, 2016)


Map of the project area locating at Guinea-Bissau, African west coast


Forest Cover

The rates of deforestation vary from one percent to three percent depending on the different definition of forest and locations. Relatively, forest with 10%-30% canopy closure suffered a smaller degree of deforestation.

The map tiles of the forest cover, loss and gain were downloaded with top-left corner at 10N, 20W where the 10×10 degree granule covers the project area, the Boé Sector, Guinea Bissau. All the raster datasets were first clipped to fit the project area and then converted into polygon. Fifteen layers of for-est loss were selected and separated by year from 2000 to 2014. All the polygon layers were projected to “WGS_1984_Web_Mercator_Auxiliary_Sphere”, the same as the original projection coordinate system as the Boé boundaries shapefile. The coordinate system outputs the shape length and shape area in meter instead of a degree in the original coordinate system.

As the polygons in the forest cover layer are coded with the percentage of canopy closure, which is often considered a synonym of canopy cover (Paletto & Tosi, 2009), selection of forest area with a minimum of 10% canopy closure and 0.5 hectares was performed (5-meter threshold of trees was automatically considered in the datasets), according to the definition of “forest” provided by Food and Agriculture Organization of the United Nation in the Global Forest Resource Assessment (2001). Additionally, a minimum of 30% canopy closure which is also commonly used for forest analysis was considered to provide insight (Owens & Lund, 2009). Forest loss layers are erased from the forest cover baseline layers to produce the final forest covers in 2014.



Forest Cover Loss. This figure illustrates the forest cover loss by year


The analysis revealed a constant and low rate of deforestation since 2001, with a spike in the year of 2013.


Map illustration of the forest loss.


Slash and Burn

Burned area analysis showed a worrying trend where the field preparations were still performed in late-dry season, against what the Foundation Chimbo has been promoting.

The Burned Area Products have a lower spatial resolution because of the nature of the MODIS where the satellites of the collection obtain images in resolutions 500m (Roy et al., 2005). The MCD64A1 products are available in 500m resolution while the actual cell size of a “500m” MODIS grid cell is 463.313m (Boschetti et al., 2015). The shapefiles containing the polygons of burned area coded with “BurnDate” in Julian Day were first clipped with the boundaries of the Boé.

The data were then merged by year, and outputted into Microsoft Excel for table-view of data. Furthermore, data dated by seasons were also produced for a better understanding of early and late field fire preparations in the area.

Dry seasons are classified “early” and “late”, where the “early” is from October to December, while the “late” dry seasons are January to May. Those processes were all done by models in ArcMap and the data, including raw data from server and processed data are all saved in geodatabases for future uses and analyzes.

The daily image collection of MODIS compromises the spatial resolution, producing raster datasets with a pixel size of 463m. Each cell coded with a burn date only marks the occurrence of fire events with a cell. The number of fire events and the percentage of the burned area remains unknown. As a result, only the counts of pixel cells could be concluded. The result showed a distinct wet and dry season when field preparations of slash-and-burn were done. It starts mid-October and continues until late May, as the rain seasons generally stretch from June until mid-October. The early-dry season field preparation promoted by Foundation Chimbo resulted in a worrying trend as the ratios of the number of incidents in early-dry seasons versus late-dry seasons have not been significantly decreasing.



The ratio of the pixel count of early-dry season fire and late-dry season fire. This figure illustrates the ratio between the fire event frequency in early-dry seasons and late-dry seasons.


Most of the burning events and larger areas of burning took place during the late-dry seasons as shown below in the animated map. Red pixels represent the burning of the late-dry seasons which Foundation Chimbo advocates against and blue pixels represent the early-dry seasons.

The affected area in km2 is unavailable in this stage because of the sinusoidal projection used is not commonly included in ArcGIS for processing. Further algorithm development or raster-processing is needed but precise counts of fire events are still not possible with MODIS satellites.


 

Animated burning event from 2008 to 2016. This figure illustrates the occurrence of the fire event frequency in early-dry seasons and late-dry seasons.


Discussion

It was agreed by the Foundation Chimbo during meetings that the size and numbers of cashew plantations have been expanding, due to the rapid-growing price of cashew (Chau, 2016). This might lead to an increasing rate of deforestation in the area. Although the Hansen’s dataset gives information regarding the extent of deforestation, the algorithms developed are not tailor-made for the project area. As cashew trees height varies from 5 to 14 meters, it is very possible that the cashew plantations are included in the forest dataset. The natural forest may suffer from different degrees of disturbance, where many more trees have been removed for plantations. The extent of deforestation remains unclear.

Furthermore, the counts of fire events by pixel do not accurately reflect the truth on the ground, where there were many more fire events on-site. Various fires could have been set within the same pixel cell without being detected by the MODIS satellite. Overall, the MODIS satellite is incapable of fire detection at a local scale (Tang & Shao, 2015). Allison et al. concluded for frequent satellite operation, regrettably, the spatial resolution cannot provide a precise image collection that reflects the occurrence of small and local fire events. The review listed the possibilities of unmanned aerial vehicles (UAVs) and manned aircraft which promise “flexibility of spatial and temporal resolution”. Martinez-de Dios et al. (2011) and Merino et al. (2012) have demonstrated the UAV systems that can provide long-term, low cost and real-time fire detection on a more localized scale. Fire detection in Boé Sector should move towards the utility of drones for a more accurate estimation.


from qgis.core import *
from qgis.gui import *
import calendar

from PyQt5.QtCore import QDate

###This function can be use in field calculator and takes in 2 variables, the year and the 3-digit Julain Day.
###The output can be customized for individual use.

@qgsfunction(args='auto', group='Custom')
def JDcal(yr, jd, feature, parent):
month = 1
day = 0
while jd - calendar.monthrange(yr,month)[1] > 0 and month <= 12:
jd = jd - calendar.monthrange(yr,month)[1]
month = month + 1
fulldate = (str(yr) + '/' + str(month) + '/' + str(jd))
return QDate.fromString(fulldate, 'yyyy/MM/dd')

 

Above is one piece of code I have used in processing the data. The MODIS satellite data are coded with Julian Day which is uncommon and difficult for us to interpret. This code coverts the year and the Julian Day to a yyyy/MM/dd format (or whatever format you fancy.) It’s fun to see such a short code works like a charm.


Full Script of the Project

Leave a Reply

Your email address will not be published. Required fields are marked *