Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Time Series Analysis Of Landcover Change from

1986 to 1999 in the Tropical Rainforest of
Southwestern Nigeria
Shehu, Basiru Gwandu Aminu, Hadiza
Agricultural Research Council of Nigeria, Nigeria Hydrological Services Agency,
Agricultural Research House, Foundation Plaza,
Plot 223D Cadastral, Zone B6, Plot 22, Shetima Ali Monguno Crescent,
Mabushi, Abuja-FCT. Utako, Abuja-FCT.

Abstract:- Deforestation is one of the major forests, forest clearing have not reduced(Njungbwen and
environmental problems in Nigeria, drastically reducing Mbakwe, 2013). One of the reasons might be that large part
the amounts of green cover in the country and eroding of the West African rain forests is linked to the lives and
the economic, environmental and social values of the livelihoods of people, providing food, fuels, fibre and other
plant communities. It is pertinent to understand the ecosystem services(Ken et al., 2010).
extent and distribution of this degradation, including the
dynamics in the other related land uses such croplands. While noting the importance of remote sensing in
Remote sensing data, tools and methods are effective in documenting landcover change on both regional and global
determining and documenting these land dynamics. Two scales,Daniel and Ayobami (2007), pointed out that there is
satellite imagery of 1986 and 1999 were analysed to a need for development of basic datasets providing
determine land changes that occurred between these two quantitative and spatial landcover information in Nigeria
periods in an area covering about 500,000 hectares in the and particularly southwest. The study area in this analysis
rainforest area of southwest Nigeria. Thus, the method covers about 500,000 hectares and it is one of the major
applied was image processing using Erdas-imagine timber production centers for decades, especially towards
software. The process involves pre-processing of the the southern part of the area. This study is geared towards
imagery, classification and accuracy assessment. The identifying areas where deforestation had occurred, together
result shows that 17, 472 hectares of forest were removed with change in the other land uses such as croplands. This
during the period under review. The extent of surface was achieved through application of remote sensing
water bodies, settlements and bare ground increased by technology. The temporal scope of the study is between
1,131, 9,073 and 15,565, respectively. Agricultural land 1986 and 1999 – a period of about 13 years.
decreased by 8,296 hectares which could be related to the
urban expansion and other infrastructural development II. MATERIALS AND METHODS
in the area. This analysis was constrained by time,
resources and the techniques. However, it provided some Two satellite imagery were aquired from the United
idea about change in the plant cover in the area from States Geological Survey (USGS). The first image was
1986 to 1999. This work could be scaled to national captured on the 15th of November, 1986 and the second was
scope. captured on the 13th of December, 1999. The first image is a
landsat5 TM image of 30 metres spatial resolution in all the
Keywords:- Land cover change, satellite imagery, image bands except band 6, the resolution of which is 120 metres.
processing, deforestation, tropical rainforest, southwest, Also, the spatial resolution f the second image is 30 metres
Nigeria. in all the bands save band 6 and the image is from Landsat7
ETM+. Both the imagery were downloaded from the
I. INTRODUCTION Landsat archive of the USGS as geotiff and the header file
revealed that they were projected to UTM (WGS84) Zone
Food and Agriculture Organisation (FAO) defined 31, resampled using cubic convolution and are both from 8-
deforestation as the conversion of forest areas to non-forest bit quantized sensors. Due to remoteness of the study area
land use or long-term reduction of tree canopy cover beyond and the limited time frame for the study, google earth/map
10% threshold within an area of at least a half was used as means of selecting training areas and ground
hectare(Tejaswi, 2007). In Nigeria, vegetation cover is validation. The analysis software used was ERDAS-
broadly classified into savannah and forest. One of the forest Imagineto run the classification algorithms and ArcGIS 10.3
subclasses in the country isTropical Rainforest which is the was used in producing the maps.
most specie-rich type of vegetation. This irreplacable
vegetation cover is drastically being reduced (FAO, 2009). A step-wise methodwas adopted in executing the
The reduction happens especially on the northern fringe analysis. The imagery were preprocessed as suggested in
where the encroaching savannah (derived savannah) Zhi-Hua et al., (2011).A classification scheme,as could be
expands deforestation extent, togather with land clearing for found in some published workssuch as Mengistu and
production of staple crops, timber and other tree-related Salami, (2007), Adediji and Ajibade, (2008), Okrie, (2012)
products and infrastructure development (Okorie, 2012). and John, (2014), was adoptedin producing landcover
Despite increased understanding of the importance of (thematic) maps from the two imagery. Following FAO,

IJISRT21NOV167 www.ijisrt.com 398

Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
(2010) and Adedeji and Adeofun, (2014), the area of each data, classification of the imagery and assessing the
land cover was calculated and compared between the two accuracy of the classifications.
imagery to identify the changes in the forest landcover. The
major methodological steps include preprocessing of the raw

III. PRE-PROCESSING

The two imagery were imported into the ERDAS- band 6 was excluded because it differed in spatial resolution.
Imaginesoftware. The two imagery, though from different As mentioned earlier, UTM (WGS84) Zone 31was the
satellites (Landsat5 and Landsat7), have similar bands 1 to projection, while the resampling technique was cubic
7. For both the imagery, band 1 to 5 and 7were stacked; convolution. The spectral bands are shown in table 1.

Band Name Band width (µm) Spatial resolution

1 Blue 0.45 – 0515 30m
2 Green 0.525 – 0.605 30m
3 Red 0.63 – 0.69 30m
4 Near IR 0.75 – 0.90 30m
5 Shortwave IR-1 1.55 – 1.75 30m
6 Thermal IR 10.4 – 12.5 60m/120m
7 Shortwavw IR-2 2.09 – 2.35 30m
8 Panchromatic 0.52 – 0.9 15m

Table 1: Landsat 5 and 7 spectral bands

The capturing satellite path and the row of the imagery cloud cover in both the imagery northern part of the imgery
used in this analysis are 190 and 55, respectively. The was subsetted. Inquire box was used in the process taking
imagery covered an area extent too large than required for XY coordinates of two corners – 635370 (ULX), 890820
this study.Therefore, there was need for subsetting as (ULY), 715050 (LRX) and 828720 (LRY). Figure1 and 2
suggested in Adedeji and Adeofun, (2014). Initially, show the subsets and their imagery.
southern part of the image was intended. However, due to

Fig. 1: Landsat5(1986) and its subset

Fig. 2: Landsat7(1999) and its subsets

After subsetting, the raw digital numbers (DN) would need model was applied using dark object subtraction to correct
to be converted to at-surface reflectance (Zhi-Hua et al., the image to at-surface reflectance. For Landsat5 image, the
2011). For the Landsat7 image, the algorithm in ERDAS for algorithm cannot be applied to TM5. Therefore, model
radiometric conversion was applied using Lmax/Lmin maker was used to convertthe raw DN to at-sensor
technique to correct it to at-sensor reflectance. Then COST reflectance (DN to radiance and radiometric correction).

IJISRT21NOV167 www.ijisrt.com 399

Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Figure 3 shows the conversion model developed in the Erdas COST model applied using dark object subtraction to
modeller. Figure 4 shows the model developed to correct the calibrate the image to surface reflectance.
imagery to at-surface reflectance. The figure shows the

Fig. 3: Raw DN to Radiance and to At-sensor Reflectance

Fig. 4: Cost Model for surface reflectance calibration

The pattern of the spectral classes in the imagery were classes, 20 iterations and 0.970 convergence threshold.
investigated using unsupervised classification. This was Figure 5 shows the spectral classes. Some of these classes
done to examine the possibility of using the spectral are sensible (did not merge different features as one) for
signatures for information classification. For both the exampleclasses 6 and 8 in 1986 image and classes 7 and 8 in
imagery, isodata technique was used with different number the 1999 image. Some of the classes are not sensible for
of classes and other parameters – iterations, convergence example class 1 in the 1986 image and class 3 in the 1999
and so on. Convergence never reach 0.70 even with 40 image.
iteration. Therefore, K-means technique was applied using 8

IJISRT21NOV167 www.ijisrt.com 400

Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig 5: Spectral classifications

IV. INFORMATION CLASSIFICATION

The conclusion from the spectral classification was that created in order to use the same AoI for the two imagery.
the spectral classes cannot be used for supervised For water class, three AoIs were created and merged into
classification due to missclassification. Therefore, Signature one. The AoIs of each class were then used to create
editor was used to create signatures using information from signatures for both the imegery. The classes are light forest,
google map/earth. Different training areas were created for dense forest, mixed wood/grass, arable land, dense urban,
each image and classes were established. The nine classes suburban, bare ground and water(figure 6). Image alarm was
comprised of deep water, shallow water, dense forest, light displayed to evaluate the signatures using pattern
forest, arable land, grassland, dense urban, suburban and recognition. Jefferies Matusita was used to evaluate the
open ground. The classifier was successfully executed for separability between the classes. For Landsat5 the minimum
both the two imagery. However, due to higher errow separability was 1314 and the average was 1392. For
probability in this method of extracting signature, using Landsat7, 1162 was the minimum and 1357 was the
Area of Interesst (AoI) was explored.The two classes of average. Since the minimum required is 1000, the
water (deep and shallow) were merged to one class ‘water’. classification was executed. Maximum likelihood algorithm
For each class, an area of interest (AoI) was separately in the Erdas Imagine software was applied.

Fig. 6: Maximum Likelihood Classification

IJISRT21NOV167 www.ijisrt.com 401

Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
V. ACCURACY ASSESSMENT

For the 1986 image, 256 random pixels were selected by the selected and the overall accuracy showed 72.66%. Also, the
software and the assessment showed the overall accuracy two classes were could not be captured. Thus, 10 random
was 72.66%. However, dense urban and water classes were pixels were added using stratefied sampling within the two
not captured. Therefore, another 140 pixels were added classes (water and dense urban) only. The overall accuracy
using stratefied sampling and the accuracy increased to was thus 72.67%.
75.25%. For the 1999 image, 140 random pixels were

VI. CALCULATING CLASS AREAS

S/N Class Name 1986 Area (ha) 1999 Area (ha) Change in Area
(ha)
1 Light Forest 237919 190228 -47691
2 Dense Forest 18836 102079 +83243
3 Mixed Wood/Grass 106802 53778 -53024
4 Arable Land 49106 40810 -8296
5 Dense Urban 3565 5601 +2036
6 Suburban 54586 61622 +7036
7 Bare Ground 23892 39457 +15565
8 Water 532 1663 +1131
Table 2: Change in land cover area between 1986 and 1999

In both the two imagery, spatial resolution for all the dense in the 1986 map than in the 1999, the calculated areas
bands used is 30metres. Therefore, each pixel has an area of showed increase in dense forest rather than decrease (table
900m2. Since 1 hectare equals 10,000m2, each pixel in this 2) over the period under consideration in this study.
analysis has an area of 0.09 hectares. Thus for each Therefore, a recoding system was applied to get a clear
lancover, 0.09 was multiplied by the histogram (number of perspective of the forest cover change between the two
pixels) to get areas in hectares. A column was added in the images. Light forest, dense forest and mixed wood/grass
attribute table of both the imagery for the values of the were merged and recoded as forests. Dense urban and
calculated areas. A decrease was observed in the light forest, suburban were also merged to form settlements class (figure
mixed wood/grass and arable land. While an increase was 7). Thus, a final land cover map was produced with
observed in the rest of the classes (table 2). fiveclasses comprising of Water, Settlements, Forests,
Bareground and Arable land.The area of each of the classes
There might be a missmatch between spectral was recalculated (table 3).
definition of dense forest in the two image. While it is
clearly obvious from the maps (figure 7) that forest is more

Fig. 7: Recoded Land cover Classes

IJISRT21NOV167 www.ijisrt.com 402

Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
VII. RESULTS AND DISCUSSION at the north central part of the 1999 map was a vegetated
area in the 1986 map. The water displaced the vegetal cover
Figure 7shows that during the period under review, due to the Ede Erinle Dam/Reservoir which was designed in
forest cover and arable lands decreased in the study area, the early 1980s by the old Oyo State Water Corporation.
while settlements, bareground and water surfaces increased. According to Adedeji and Ajibade (2008), the
From table 3, the magnitude of this decrease in forest cover reservoircovers about 14km2 at the normal water level and
(deforestation) is clear,covering 17,472 hectares between about 15km2 at the maximum water level (figure 9). It was
1986 and 1999. It is obviously noticeable in figure 7 that the expanded to improve water supply system to Osogbo (the
strands of forest in the northwestern corner of the 1986 map capital of the current Osun State) and other cities and towns
are almost cleared in the 1999 map. The densely packed around. The GoogleEarth image shows the dam structure
forests around the centre in the 1986 map is less dense in the and the map shows the position of the dam as published by
1999 map. Water bodies expanded by 1,131 hectares in the Adedeji and Ajibade (2008).
area. This change in surface water bodies is also clearly
visible comparing the two maps in figure 7. The water body

S/N Class Name 1986 Area (ha) 1999 Area (ha) Change in Area
(ha)
1 Forests 363557 346085 -17472
2 Arable Land 49106 40810 -8296
3 Settlements 58150 67223 +9073
4 Bare ground 23892 39457 +15565
5 Water 532 1663 +1131

Table 3: Change in land cover area between 1986 and 1999

Fig 8: Ede Erinle Dam/Reservoir, after Adediji and Ajibade (2008)

Increase was observed for Settlements and bareground most prevelent activities for forest exploitation(Ikemeh,
to the tune of 9,073 and 15,565 hectares, respectively. 2013). Other forms of activities, according to him, include
Deforestation due to agricultural expansion, urban growth, collection of Thaumatococcus daniellii (broad leaf used for
industrial expansion and pressure from an increasing wrapping food), Irvingia gabonensis (edible seed), Garcinia
population has reduced the extent, diversity and stability of mannii (chewing stick), and Carpolobia lutea (sanda stick).
the Nigerian forests(Omofonmwan and Osa-Edoh, 2008).
Acccording to Daramola and Ibem (2010), urban sprawl VIII. LIMITATIONS, ERRORS AND
leads to depletion of green areas resulting in the loss of UNCERTAINTIES
biodiversity especially where the development is haphazard
and without proper planning. Typical in developing Google earth/map was the basis forselecting training
countries, rural-urban migration contributes to the rapid areas as mentioned earlier.Thus, the techniques of photo
expansion of cities and abandoning of the farmlands in the interpretation such as shape, pattern, and so on were used to
rural areas. This may be the reason for the decrease in arable identify and differentiate land cover types. Several parts of
land as shown in table 2 and 3 (8,296 hectares).Also, it may the study area is converted to plantation and this study did
partly be the result of cities expansion into the farm lands not take the account of that; not all the forest covers are
around due to construction of roads, buildings and so on. In natural as suggested by Adedeji and Adeofun (2014).
addition, in the whole of the Nigeria’s lowland rainforest, Attempts were made to use imagery of similar dates, but that
logging, farming and hunting were identified as the three was not possibledue to the availability of the satellite
imagery and the problem of cloud cover. However, both the

IJISRT21NOV167 www.ijisrt.com 403

Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
images used were aquired within the short dry season of the sensing (RS) and GIS techniques. Journal of
area (November and December). The assumption from the Geography and Regional Planning. Vol. 1(6), pp. 110-
experience was that one month difference will not cause 115. ISSN 2070-1845. Available at:
considerable seasonal effects. Also the imagery were both <http://www.academicjournals.org/article/article13794
pre-resampled using cubic convolution, so there was no 09769_Ajibade%20and%20Adediji.pdf> [Accessed:
possibility of applying other algorithm like nearest 29/03/2015].
neighbour. The classification accuracy assessment was [4.] Eroarome M. A. 2009. Country Pasture/Forage
based on the computer-generated random pixel samples; the Resource Profile. FAO-Nigeria. Available at:
accuracy indicates only the match or otherwise between the http://www.fao.org/ag/agp/AGPC/doc/Counprof/PDF
thematic map and the input image instead of the actual land %20files/Nigeria.pdf [Accessed: 03/08/2013].
cover on the ground. However, it is believe that the input [5.] Giri Tejaswi, 2007. Manual on deforestation,
imagery provide good representation of the actual land degradation, and Fragmentation using remote sensing
covers. and GIS. FAO, MAR-SFM Working Paper 5, Rome.
Available at: <http://www.fao.org/forestry/18222-
IX. CONCLUSION 045c26b711a976bb9d0d17386ee8f0e37.pdf>
[Accessed: 07/04/2015].
Although there were limitations and uncertainties, it [6.] FAO 2010. Global Forest Resource Assessment. Main
was concluded from this study that huge area of forest have Report, Rome. Available at:
been removed without replacement between 1986 and 1999 <http://www.fao.org/docrep/013/i1757e/i1757e.pdf>
in the study area. Some of the areas were covered by the [Accessed: 22/02/2015].
expanded water reservoirs andexponential urban [7.] John O. 2014. Predicting Changes in Landuse and
expansion.Some of the forests were exploited for fuel, Landcover in Niger Delta Using Post-classification
commercial and other domestic uses. This research could be Analysis. Civil and Environmental Research. Vol. 6(5).
scaled up to cover the whole country in order to get general Pp 110-117. ISBN 2225-0514. Available at:
understanding of how much plant cover changed over the <http://www.iiste.org/Journals/index.php/CER/article/d
decades and where these changes occurred. ownload/14021/14305> [Accessed: 21/02/2015].
X. ACKNOWLEDGEMENT [8.] Ken N., Alex A., Ben C., Jim G., John M., Ben P. and
Any W. 2010. Biodiversity in a Forest-agriculture
This study was originally conducted as a coursework Mosaic – The Changing Face of West African Rain
during the MSc programme of the corresponding author at Forest. Biological Conservation. Vol.143. pp 2341-
the Faculty of Engineering and Science, the University of 2350. Available at: <http://ac.els-
Greenwich, London. Dr. Meredith Williams was the course cdn.com/S0034425711003312/1-s2.0-
instructor with whose guiding critiques the study was a S0034425711003312-main.pdf?_tid=3b75b3f4-b7c7-
success. West Africa Agricultural Productivity Programme 11e4-a2c3
(WAAPP-Nigeria) funded the entire MSc programme. 00000aab0f01&acdnat=1424302885_9e8ad443277ecd
Agricultural Research Council of Nigeria provided the 45226f246a7319ed9b>[Accessed: 07/04/2015].
nomination for the sponsorship. The satellite imagery were [9.] Mangistu D. A. and Salami A. T. 2007. Application of
sourced from the USGS free-of-charge. All other sources of Remote Sensing and GIS in Landuse/Landcover
literature are also hereby earnestly acknowledged. The work Mapping and Change Detection in a Part of
has now been developed into publishable article together Southwestern Nigeria. African Journal of
with the co-author. Environmental Science and Technology. Vol. 1(5).
099-109. ISBN 1996-0786. Available at:
REFERENCES <http://www.academicjournals.org/article/article13801
10623_Mengistu%20and%20Salami.pdf> [Accessed:
[1.] Adedeji Daramola & Eziyi O. Ibem, 2010. Urban 24/02/2015].
Environmental Problems in Nigeria: Implications for [10.] Njungbwen E. and Mbakwe R. 2013. A GIS and
Sustainable Development. Journal of Sustainable Remote Sensing Approach to Assessment of
Development in Africa.Volume 12(1). ISSN: 1520- Deforestation in Uyo, Akwa Ibom State, Nigeria.
5509. Available at: Ethiopian Journal of Environmental Studies and
<http://core.ac.uk/download/pdf/12356053.pdf> Management. Vol. 6(4). Pp 348-357. Available at:
[Accessed: 07/04/2015]. <http://www.ajol.info/index.php/ejesm/article/viewFile
[2.] Adedeji, O.H. and Adeofun, C.O., 2014. Spatial /90834/80264>[Accessed: 09/03/2015].
Pattern of Land Cover Change Using Remotely Sensed [11.] Okorie F. C. 2012. A Spatio-temporal Analysis of
Imagery and GIS: A Case Study of Omo-Shasha- Deforestation in Epe and its Environs (Lagos, Nigeria).
Oluwa Forest Reserve, SW Nigeria (1986-2002). International Journal of Science, Environment and
Journal of Geographic Information System, 6, 375-385. Technology. Vol. 1(5). Pp 548-562. Available at:
Available at: <http://www.ijset.net/journal/492.pdf>[Accessed:
<http://dx.doi.org/10.4236/jgis.2014.64033> 20/02/2015].
[Accessed: 30/03/2014]. [12.] Rachel Ashegbofe Ikemeh, 2013. Sustainable forest
[3.] Adedeji and L.T. Ajibade, 2008. The change detection management in a human dominated landscape and its
of major dams in Osun State, Nigeria using remote implications for biodiversity conservation: A Nigerian

IJISRT21NOV167 www.ijisrt.com 404

Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
lowland forest perspective. Research and Reports in S0034425711003312-main.pdf?_tid=3b75b3f4-b7c7-
Biodiversity Studies. Vol.3 pp 9–23. DovePress, 11e4-a2c3-
Available at: http://www.dovepress.com/sustainable- 00000aab0f01&acdnat=1424302885_9e8ad443277ecd
forest-management-in-a-human-dominated-landscape- 45226f246a7319ed9b>[Accessed: 18/02/2015].
and-its-i-peer-reviewed-article-RRBS [Accessed: [14.] S. I. Omofonmwan and G. I. Osa-Edoh, 2008. The
18/02/2015]. Challenges of Environmental Problems in Nigeria.
[13.] Shi Z., Li L., Yin W., Ai L., Fang N. and Song Y. Journal of Human Ecology. 23(1) 53-57. Available at
2011. Use of Multi-temporal Landsat Images for <http://www.krepublishers.com/02-Journals/JHE/JHE-
Analysing Forest Transition in Relation to 23-0-000-000-2008-Web/JHE-23-1-000-000-2008-
Socioeconomic Factors and the Environment. Abst-PDF/JHE-23-1-053-08-1650-Omofonmwan-S-
International Journal of Applied Earth Observation and I/JHE-23-1-053-08-1650-Omofonmwan-S-I-Tt.pdf>
Geoinformation. Vol.13 Pp468-476. Available at: Accessed [07/04/2015].
<http://ac.els-cdn.com/S0034425711003312/1-s2.0-

IJISRT21NOV167 www.ijisrt.com 405