Land monitoring in Nigeria using geo-spatial data mining approach.
Land occupies a unique place in the development process of any individual or society. The supply of useable land is, however, limited. No society therefore, exists without a regulation of some kind peculiar to it to rationalize the mode of ownership and the use of land (Laarakker, 2013)
According to Obaseki (1988) Land includes land of any tenure, buildings or parts of buildings (whether the division is horizontal, vertical, or made in any other way), and other corporeal hereditaments. In Nigeria's traditional setting, for instance, land was not viewed as a mere economic tool, rather it had religious and other social functions. Thus, native rule was rightly seen to depend upon the native land system. In the strict native rule and custom, land was believed to belong to the living, the dead and the unborn. Land, therefore, had metaphysical content and was viewed as an inherent part of social relations process between people, society and gods (Carlisle, 1999).
The promulgation of the Land Use Decree in 1978 was an exercise to redirect the general philosophies of pre-existing land tenure systems in our society through the application of a uniform statutory regulation of ownership and control of land rights. Hence, sustainable land use implies activities that are ecologically sound, socio-culturally acceptable, economically viable as well as equitable in terms of access to land resources, benefits and decision-making process (Schwabe, 2001, Atilola, 2013).
Government is faced with the aforementioned problems of making crucial decisions on deforestation, agriculture, and food security amongst others. These problems have constituted a stumbling block against the millennium's development goal and proper land use monitoring is a key for food security. Craig (2005) reported the major challenge in Africa to be that many countries on the continent do not have the foundational geo-information needed to create the spatial layers of information for use in the implementation and monitoring of national and regional development strategies such as the New Partnership for Africa's Development (NEPAD).
II Land Use Availability and Usage
In many countries, projected population increases superimposed on existing land holding pattern will result in an incredible increase in poverty as a result of accelerated erosion, deforestation and desertification along with continued loss of the genetic resources needed to provide a steady stream of new seed varieties (USAID, 2000). Land use coupled with the effort of small farmers is the key instruments for achieving sustainable increases in yield and productivity. However, insecurity of tenure, especially among small-scale farmers, has been known to act as a disincentive to the conservation of resources, including reforestation and soil conservation projects.
In Nigeria, land use problems that result into land pollution and are accorded highest priority range from the many causes of deforestation, soil erosion and dumping or disposal of both industrial and domestic wastes that are hazardous or harmful and consequently render land unproductive or degrading and unsustainable (Akpomrere and Nyorere, 2012)
An important resource on land which has been grossly abused and unsustainably used is our forests. Forests provide human beings with a wealth of benefits including contribution of about 19% of the energy supply of lower income countries through fuel wood resource, provision of resource base for agriculture, tourism, recreation, religion, culture, music, etc. Despite these functions/benefits our forests have been degraded through unsustainable logging, shifting agricultural practices, fuel wood gathering, bush burning and overgrazing of land (Ladan, 2007). For centuries, shifting cultivation and trans-human pastoralism systems allowed people to derive their livelihood in a sustainable manner from nature. When soil fertility declines or pasture vegetation disappears, people move to new lands and allow natural regeneration of used lands to its original state. The fallow period could be between 10-20 years. With increase in population, farming and pastoral land has become scarce and new forest lands have been opened up for both traditional and mechanized farming, even in urban centres land use poses a serious challenge (Ibrahim and Kwankur, 2012), An example can be found in Nigeria where forests were reserved for palm and rubber plantation development (FEPA, 1998). By 1994, about 83,672 hectares of such area were de-reserved. Fallow periods have been shortened while in several communities repeated farming on the same piece of land is carried out using the same traditional systems that are suitable only for shifting cultivation. The effect is non-restorable soil fertility, low crop yield and farmers' migration to marginal land that encroaches into forests.
A Federal Government study showed that firewood is the source of energy for 80% of the rural population. With increasing cost of fossil fuel energy, it is likely that there will be increased dependence on firewood for energy supply. The increasing demand for firewood accelerated the rate of woodland destruction, soil degradation, river siltation, desertification and general environmental degradation (David, 2011). Other factors are bush burning, overgrazing of land as a result of cattle rearing.
A Geo-Spatial Data Mining
Geo-spatial data mining is a branch of data mining, but differs from the general spatial data mining and data mining associated with business databases. Geo-spatial data mining has large space dimensions vis-a-vis the general data mining. Spatial data mining extracts existing knowledge, the space relation or other meaningful features of database space. Shekhar et al (2003) described spatial data mining as the process of discovering interesting and previously unknown, but potentially useful patterns from large spatial data sets. Useful patterns extraction from spatial datasets is more complex than the corresponding patterns from traditional numeric and categorical data.
III Conceptual Land Use Monitoring Model and Model Analysis
[FIGURE 1 OMITTED]
The pre-processing of geo-spatial data is stored in the database by removing noise and smoothened using the Gaussian filter.
Feature extraction of features is essential for accurate classification (Hermosillo, 2010). It concerns finding shapes in computer images. In feature extraction, the invariance is calculated.
A The naive Bayes probabilistic model
The probability model for a classifier is a conditional model
over a dependent class variable C with a small number of outcomes or classes, conditional on several feature variables [F.sub.1] through [F.sub.n]. The problem is that if the number of features n is large or when a feature can take on a large number of values, then basing such a model on probability tables is infeasible. We therefore re-write the model to make it more tractable.
Using Bayes' theorem, we write
P(C|[F.sub.1],.....,Fn)= F(c)[beta](F1, ..., Fn)|c)/p ([F1], ..., Fn) 1
The above equation can be written as
Posterior = prtor x likelihood/evidence 2
In practice, only the numerator of that fraction is of interest, since the denominator does not depend on C and the values of the features [F.sub.i] are given, so that the denominator is effectively constant. The numerator is equivalent to the joint probability model
which can be re-written as follows, using repeated applications of the definition of conditional probability: P(C.[F.sub.1] ..., [f.sub.n])
= P(C)p([F.sub.1],.....,[F.sub.n]) 4
= P(C)p([F.sub.1]|C)p([F.sub.2], ..., [f.sub.n]|[C.sub.1][F.sub.1]) 5
= P(C)p([F.sub.1]|C)p([F.sub.2]|C, [F.sub.1])p([f.sub.3], ..., [f.sub.n]|[C.sub.1][F.sub.1], [F.sub.2]) (6)
= P(C)p([F.sub.1]|C)p([F.sub.2]|C,[F.sub.1])p([f.sub.3]||C, [F.sub.1], [F.sub.2])p([F.sub.4], (7) ....., [F.sub.n]|C, [F.sub.1], [F.sub.2], [F.sub.3]) 7
= P(C)p([F.sub.1]|C)p([F.sub.2]|C, [F.sub.1])p(/3 |C, [F.sub.1], [F.sub.2]).......p([F.sub.n]c, [F.sub.1], [F.sub.2], [F.sub.3], ..., [F.sub.n-1]) 8
The "naive" conditional independence assumptions come into play: assume that each feature (F) i is conditionally independent of every other feature (F) j for
j [not equal to] i. This means that p([F.sub.i]|C,[F.sub.j])= p([F.sub.i]|C) 9
for i [not equal to] j and so the joint model can be expressed as
P(C, [F.sub.1],...., [F.sub.n]) = p(C) p([F.sub.1]|C)p([F.sub.2]|C)p([F.sub.3] |C)...... 10
P(C)[[PI].sup.n.sub.i=1] F([F.sub.1]|C) 11
This means that under the above independence assumptions, the conditional distribution over the class variable C can be expressed like this:
P(C|[F.sub.1],...., [F.sub.n]) = 1/z p(C)[[PI].sup.n.sub.i=1] P([F.sub.I]|C) 12
where Z (the evidence) is a scaling factor dependent only on [F.sub.1], ... [F.sub.n], i.e., a constant if the values of the feature variables are known. Models of this form are much more manageable, since they factor into a so-called class prior p(C) and independent probability distributions p([F.sub.1]|C). If there are k classes and if a model for eachp([F.sub.1]|C = c). can be expressed in terms of r parameters, then the corresponding naive Bayes model has (k--1) + n r k parameters. In practice, often k = 2 (binary classification) and r = 1 (Bernoulli variables as features) are common, and so the total number of parameters of the naive Bayes model is 2n + 1, where n is the number of binary features used for classification and prediction
IV SYSTEM IMPLEMENTATION
The proposed model used for this project makes use of the supervised method of classification which classifies the extracted attributes from aerial images into different categories based on the features (attributes).
A Implementation of Naive-Bayes approach
Let [X.sub.I],..., [X.sub.m] denote our features (attributes), Y is the class number, and C is the number of classes. The problem consists of classifying the case ([x.sub.l], ..., [x.sub.m]) to the class c maximizing P(Y=c| [X.sub.1] = [x.sub.1], ..., [X.sub.m]=[x.sub.m]) over c=1, ..., C. Applying Bayes' rule gives
P(Y = c| X1= x1, ..., Xm = xm) = P(X1
= x1, ..., Xm
= xm |y = c)P(Y = c) P (x1
= x1, ..., Xm = xm) 13
P(Y|X) = P(X[intersection]Y)/P(x)ifP(X)>0 14
where X[intersection]Y is the intersection of the two events, X and Y. This indicates that if the events, Y, and X, can both occur, then the probability of such an occurrence is equal to the probability of the X occurrence multiplied by the probability of Y, given that X occurs,
P(X[intersection]Y) = P(X)P(Y/X). 15
P([Y.sub.j]|X) P([Y.sub.j][intersection]X)/[[SIGMA].sup.K.sub.i=1]P([Y.sub.1][intersection]X) 16
= P([Y.sub.j])P(X|[Y.sub.j])/[[SIGMA].sup.K.sub.i=1]P([Y.sub.1])P(X|[Y.sub.j]) 17
The denominator is invariant across classes, and therefore can be ignored. Under the NB's assumption of conditional independence, P([X.sub.1] = [x.sub.1], ..., [X.sub.m]=[x.sub.m] | Y=c) is replaced by m [PI] P([X.sub.i]=[x.sub.i] | Y=c) and the NB classification reduces the original problem to that of finding i=l (that is, the group 1 class)
[??] = argmax [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] 18
All discrete (categorical) features estimating the probabilities in (1) can be done using frequency counts. In other words, P([X.sub.i]=[x.sub.i] | Y=c) is estimated as #([X.sub.i]=[x.sub.i] & Y=c) / # (Y=c), where #() denotes the number of cases in the training data set satisfying the condition in parenthesis.
B The pre-processing
In this stage, the geo-spatial data undergoes filtering so as to achieve enhanced data. The images to be filtered are gotten from Google maps (Akure as a case study). The image shows the land use.
[FIGURE 2 OMITTED]
C The Feature Extraction phase
The filtered image is shown below, in which desired area is extracted. This was input into the program Terra View which has geoDMA. Terra View is used for land analysis and other functions pertaining to land while geoDMA which means geographical data mining analyst does a lot of things ranging from segmentation to classification. Figure 3 shows the aerial image on left and the processed image on right
[FIGURE 3 OMITTED]
Extraction, normalization, training, visualization and classification were done. Segmentation was initially carried out, and then features were extracted.
Features with prefix p_ are spatial (use polygons), and with prefix c_ are cell-based (use cells.). Features with prefix r are spectral, combined with polygons rp_ or with cells rc_. Normalized features in the next tap have the _n suffix
The attributes are shown below.
[FIGURE 5 OMITTED]
The classification is done using GeNIe which makes use of the Bayesian network classifier embedded in it to classify extracted data. The network is graphically displayed in the Graph View. This is used for building graphical decision-theoretic models. Geospatial data extracted is stored in the GeNIe. The mean, variance, standard deviation, minimum, maximum and count are computed and displayed in a relational database mode and also the correlation matrix is gotten. Figure 5 shows the statistics interface displaying the mean, variance, standard deviation, minimum and maximum and also the count of data mined from the aerial image.
E The Network Bayesian MACRO
The text of the macro for Naive-Bayes classification is shown below. It has five parameters: training data set (contains classified cases); score--data set containing cases to be classified; nclass--# classes (C); target--name of the variable in the 'train' data set that has the class number (Y); 'target' is assumed to be a numeric variable with values 1,2,.. for classes 1, 2, and so on; if it is not, it has to be recoded before running the macro; inputs--the list of features' names ([X.sub.1], ..., [X.sub.m]).
Table 2.0 illustrates the classification results. The training data includes 3,377 observations of which 92.7% ((153151+39061)/3377) were correctly classified by the NBC, and the testing data includes 3377 observations of which 92.7% ((76521+19565)/3377) were correctly classified.
A 3-phased conceptual model based on Bayesian classifier for a given geo-spatial data is presented. It comprises geospatial image processing, features extraction and classification using Bayesian classifiers. The result shows 92.7, classification. Consequently, if adopted, land use monitoring will become easier for the government. Thereby enhancing decisions making, planning and policies formulation. The multiplier impact will enhance the citizens' livelihoods as well as achieving sustainable development in Nigeria with Vision 2020:20.
Akpomrere, O. R. and Nyorere, O. (2012), "Land Use Patterns and Economic Development of Ikeja in Lagos State, Nigeria: The Geographic Information System Approach", International Journal of Economic Development Research and Investment Vol 3 No. 3, Dec. 2012.
Atilola Olusola (2013), Systematic Land Titling and Registration in Nigeria: Geoinformation Challenges, FIG Working Week 2013, Abuja, Nigeria, 6-10 May 2013. Carlisle, E.F., (1999). "Postcolonial cultural studies: Africa and the Caribbean", Centre for Interdisciplinary Studies, Viginia Tech, http://athena.english.vt.edu/~carlisle/Postcolonial/Colonialism_Africa.html
Craig, A.S., (2005), "Geo-information and NEPAD, Human Sciences Research Council (HSRC), NEPAD Report
David Stephen (2011), "Climate Change Mitigation and Adaptation in Nigeria", http://trpns.com/wpcontent/uploads/2011/09/ Climate-Change-Mitigation-andAdaptation-in-Nigeria.pdf
FEPA (1998) Federal Environmental Protection Agency Report
Hermosillo Gerardo, (2010) "Variational Methods for Multimodal Image Matching", tel-00457459, version 1-17 Feb 2010. http://tel.archives ouvertes.fr/docs/00/45/74/59/PDF/Hermosillo.pdf Accessed Oct 5, 2011.
Ibrahim Usman Jibril and Kwankur Tatte Garba (2012), "The Challenges of Housing Development and Needs in Abuja", FIG Working Week, Rome, Italy, 6-10 May 2012
Ladan, M.T. (2007), "Environmental Law and Sustainable Land Use in Nigeria" in Nathalie J. Chalifour, Patricia Kameri-Mbote, Lin Heng Lye and John R. Nolon, (eds.), Land Use Law for Sustainable Development", IUCN Academy of Environmental Law Research Studies/Cambridge University Press, UK, 2007, pp. 240252.
Laarakker Peter, (2013), Surveying (in) Nigeria 1978-1980, http://www.fig.net/pub/fig2013/ppt/ts04e/TS04E_laarakker_6691_ppt.pdf
Obaseki, Andrew, (1988) "The Judicial Impression of the Nigerian Law of Property: Any Need for Reform?" in Proceedings of the 26th Annual Conference of the National Association of Law Teachers (Falculty of Law, Rivers State University of Science and Technology, Port Harcourt, Nigeria), pp17,18
Schwabe, C.A., (2001), "African Renaissance: towards the development of a geo-information system for socioeconomic development in Africa". Nairobi, Kenya: 5th AfricaGIS Conference.
Shekhar, S. Zhang, P., Huang, Y. and Vatsavai, R. (2003) "Trends in spatial data mining," in H. Kargupta, A. Joshi, K. Sivakumar, and Y. Yesha (eds.), Data Mining: Next Generation: Challenges and Future Directions, AAAI/MIT Press.
USAID, (2000). U.S. International Food Assistance, United States Agency for International Development (USAID) Report.
A F Thompso, A O Adetunmbi and B K Alese
Computer Science Department, Federal University of Technology, Akure, Nigeria
firstname.lastname@example.org. email@example.com and firstname.lastname@example.org
Table 2.0 Results of the Naive Bayesian classifier Dataset Increase Normal Train 153,151 1,321 (Training) 13,819 39,061 Test 76,521 645 (Testing) 6944 19565
|Printer friendly Cite/link Email Feedback|
|Author:||Thompso, A.F.; Adetunmbi, A.O.; Alese, B.K.|
|Publication:||Computing and Information Systems|
|Date:||Feb 1, 2013|
|Previous Article:||ICT for development by Nigerian NGOs: an initial conceptual paper.|
|Next Article:||Social media for knowledge sharing in African development institutions: a viewpoint paper.|