Fuzzy Decision Analysis in Land Suitability Evaluation: A Tool For Precision Land Management Interpretation

GISdevelopment.net ---> Application ---> Environment

Fuzzy Decision Analysis in Land Suitability Evaluation: A Tool For Precision Land Management Interpretation

Sumbangan Baja*, Muh. Ramli**, and Muh. Jayadi*
*Department of Soil Science Hasanuddin University Makassar;
ph (+fax) 0411-587076 (corresponding author,

E-mail: sbja02@yahoo.com.au) **Maros Soil Research Station,
Department of Agriculture,
Republic of Indonesia, ph 0411-371572.

Abstract
Continued deterioration of land and water resources is occurring in many parts of the world partly as a result of land management practices not being suitably matched to the suitability and capability of the land resources. To help address this problem, there is a need for information that will allow land managers to identify both the inherent suitability of land to a particular use and the spatial distribution of land areas where present land use types and their suitability are mismatch. This paper describes a land suitability evaluation approach based on fuzzy set methodology in Geographical Information Systems (GIS). The Semantic Import Model of fuzzy set was employed using the available land resource information. A digital elevation model (DEM) was also developed in this study to provide land slope data information. Using a convex combination approach (between internal and external land characteristics), a Land Suitability Index (LSI) was then mapped throughout the study area. Available land use information generated from classification of Landsat ETM+ images was also employed to identify a spatial mismatch between existing and suitable land use. This is useful for identifying favourable and unfavourable land management practices with the aim of better matching land management with their suitability. The framework will lead to the better interpretation of precision land management in the study area.

1. Introduction
Geographic information technology, in synchrony with global positioning systems, has created opportunities for site-specific land management through the collection and synthesis of resource and production data. However, collection of information over space and time has outperformed our ability to interpret and apply the data. Consequently, the major goal of site-specific management, which is to see a continuous improvement of management decisions, has lagged as a result of poor agronomic interpretation and information delivery (Cook and Bramley 2001). Engineers, agronomists, and geo-spatial analysts must together develop frameworks to maximise the flow of information back to farm managers to improve land management.

Land evaluation can assist government, industry and individuals in rational decision-making about agricultural land use. The most commonly used systems of land evaluation are land capability and land suitability assessments, which can be seen in a classical systems such as the USDA method (Klingebeil and Montgomery, 1961) the FAO framework (FAO, 1976). To date these methods have been used for:

Linking the survey of basic land resources to land use planning (Baja et al., 2005)
Quantifying the degree of land degradation risk present and providing a baseline that can be used to assess changes in the landscape over time (see Baja et al., 2001b; 2002b)
Permitting the use of mathematical modeling such as fuzzy decision analysis to be used in refining analytical procedures and output presentation (Baja et al., 2002a; 2002c)
Allowing modern methods of information processing, such as Geographic Information Systems (GIS), to be employed in land use planning (Baja, 2001; 2005).

Although traditional land evaluation methods are useful for the initial assessment of the feasibility or suitability of proposed agricultural land use, further development of land evaluation techniques is required to effectively assess the sustainability of proposed and current land use and more specifically, land management practices (Smith and McNeill, 2001). There is therefore a need for stronger linkages between regional land use planning and on farm land management. Greater attention needs to be given to how sustainable land management decisions can be made at different scales.

The primary aim of this study is to assess land suitability for agriculture in the agricultural development zone using fuzzy decision analysis technique in Geographic Information Systems (GIS). The procedures used entail the following steps: (i) development of soil data base including digital elevation model (DEM); (ii) assessment of land suitability; and (iii) cross-comparison and validation of model developed. In model validation, an assessment was made against maize (local variety), which is very commonly found in the study region. Cross-comparison analysis was undertaken against existing land use to describe the mismatch between the two sets of land information. This will give insights into precision land management interpretations in the area of interest.

2. Methodology

Study area
The area selected for this study includes some parts of the lower Jeneberang River catchment, covering an area of approximately 37.000 ha, located about 30 km Southeast of Makassar, South Sulawesi, Indonesia (Figure 1). Based on existing land use map (Baja dkk., 2004), agriculture is the predominant land use in the study region consisting of paddy field 16,725 ha (45%), followed by shrubs 9,335 ha (25%), mixed farms 5,071 ha (14%), forest 4,087 ha (11%), water body (Bili-Bili Dam ) 1,766 ha (5%), and residential 379 ha (1%).

Figure 1. Location of study area

Data bases and preliminary data processing
Data bases used were obtained from the following sources:

Soil map of lower Jeneberang River Valley (Research Centre, Hasanuddin University, 1999)
Digital topographic map of study area with a scale of 1 : 50,000 from BAKOSURTANAL (National Agency for Survey and Mapping)
Climate data.

Twenty three types of soil units were found in the study area (see Table 1). Soil unit boundaries were digitised in Arc/Info, and attribute data (texture, effective depth, drainage, organic carbon, and pH) were input in the polygon attribute (PAT) files.

Table 1. Soil units, family names and their respective area (Research Centre, Hasanuddin University, 1999)

As soil data sets were stored in an Arc/Info vector format, calculation of a membership function of each soil characteristic was done in the attribute file (PAT). Therefore, a soil family is used as a delineation unit of individual properties in the spatial representation of membership function (MF) and joint membership function (JMF) of soil attributes.

Digital Elevation Model (DEM) was constructed using interpolation procedures from vector-based digital contour map, with an interval of 25 metres. Grid size of raster data were set at 25 metres.

Suitability analysis using fuzzy set methodology

A fuzzy set is most commonly used for classifications of objects or phenomena in continuous values, where the classes do not have sharply defined boundaries. It deals with a class with a continuum of grades of memberships (Zadeh, 1965). A fuzzy set A may be defined as follows:

Where X = {x} is a finite set (or space) of objects or phenomena, mA(x) is a membership function of X for subset A. Therefore, a fuzzy subset is defined by the membership function (MF) that defines the membership grades of fuzzy objects or phenomena in the ordered pairs, consisting of the objects and their membership grades. The MF of a fuzzy subset determines the degree of membership of x in A (Burrough et al., 1992).

There are several ways to generate a fuzzy membership function. For environmental applications, there are two different but complementary approaches to grouping individuals into fuzzy sets or classes. The first is the Similarity Relation model (SR), and the second is based on the Semantic Import model (SI). The fuzzy c-means and its modifications is one of the examples for the first. An SI model (see Burrough et al., 1992) is comparatively simple to use, because it utilizes an a priori membership function (MF) for individual variables under consideration. Examples can be seen in Baja et al. (2002c), Burrough et al. (1992), and Davidson et al. (1994). With this approach, the attribute values considered are converted to common membership grades (from 0 to 1.0), according to the class limits specified by the analysts based on experience or conventionally imposed definitions.

If MF(xi) represents individual MF values for ith land property x, then, the basic SI model function take the following form in the computation process:

As there are n land characteristics to be rated, the MF values of individual land characteristics under consideration are then combined using a convex combination function to produce a join membership function (JMF) of all attributes, Y as follows:

where li is a weighting factor for the ith land property x, and MF(xi) denotes a membership grade for the ith land property x. With this index, the higher the value (approximate to 1.0) the better the quality of land for developing the land use under consideration. In the computation, it is crucial to examine an appropriate fuzzy model parameter to suit each decision criterion. The choice depends on the 'trend of performance' of the respective land attribute in accommodating a favourable condition for a selected land use type (Baja et al., 2002c). Model parameters include LCP (lower crossover point), b (central concept), UCP (upper crossover point), and d (width of transition zone) (see Burrough et al., 1992). Detailed application of this fuzzy set methodology can be seen in Baja et al. (2001a), and Baja et al. (2002c). The readers are also referred to Baja and Dragovich (2006), and Burrough and McDonnell (1998).

Spatial matching
In this study, a spatial analysis was undertaken to identify the spatial matching between land suitability index and existing land use types in the study area. A simple overlay technique was used between existing land use map and LSI map layers. The land use map used as a 'feature definition image,' was generated from supervised classification of Landsat ETM+ images (Baja dkk., 2004) consisting of six major land use classes: forest, shrubs, mixed dryland agriculture, paddy field, water body, and residential. Therefore, the statistics of LSI within each land use can be produced on a spatial basis.

Model validation
Commonly-used approaches of model validation include testing for predictive ability and comparison against performance standards. For land suitability assessment, the second method may be more appropriate to use, and land productivity measures (such as crop yields, costs required for improving biophysical constraints, etc.) are employed as a performance standard.

Validation exercise in this study entailed the following steps: (i) preliminary survey on crops (maize) grown in the study area; (ii) involvement of the study team in harvesting the crops; and (iii) analysis in GIS. The first step was done in February 2005 with the aim of exploring locations of maize-growing area. Thirty one locations were found in the study region. This step also includes interviews and communications with the farmers (owners or tenants) in each location aiming at getting permission for access during crop harvesting in March and April 2005.

In the second step, intensive work was done in the field to harvest the crops. A representative block with 2.5 x 2.5 sq. m was determined in each location. Two types of production measures were obtained: upper biomass (in kg) and seeds (kg) that can be converted in ton/ha. At the same location, a soil sample was taken to analyse some characteristics (such as texture, P, K, N, pH and organic matter). In the third step, spatial correlation analysis was performed in GIS between production measures and Land Suitability Index (LSI) for maize.

3. Results and Discussion

Digital elevation model
The result of digital elevation model (DEM) mapping is presented in Figure 2a. It shows that the study area is dominated by an altitude of below 100 m asl. The maximum height of the area is 525 m in the southeast of the study region. This DEM was used as the basis for generating slope map of the area (Figure 2b). The statistics of the slope map indicate that flat land (slope 0 to 3%) consists of 27,681 ha (74% of the study region), and land with slope 3 to 8% and 8 to 15% consists of 4,280 ha (or 11% of the study area). Land with slope of more than 15% is found mostly in the eastern part of the region comprising around 5,402 ha or 15% of the study area. The topographic characteristics as such influences the level of suitability for developing different kinds of land use, as described below.

Figure 2. Digital Elevation and slope map of study arae

Land Suitability Index
Spatial distribution of LSIs for annual crop (cropping) and the number of cells represented by the index is presented in Figure 3. The figure indicates that the higher development potential for cropping is found in the west segment of the study area with LSIs ranging from 0.80 to 1.0, and lower towards east. The graph in Figure 3 shows the distribution of cells (1 cell equals 25 by 25 metres), for the index in data space. The statistics show that the most land areas have LSIs more than 0.70 (second peak in the graph), while some areas have LSIs ranging from around 0.45 to 0.55 (first peak).

Figure 3. Spatial distribution of LSIs and the histogram of LSI cells (note: 1 cell represents 25 by 25 metres)

Model validation and existing yield patterns
The correlation between LSIs and yields (ton/ha) is given in Figure 4 with the following form:

Where Y denotes existing corn yield in kg/ha, and X represents Land Suitability Index (LSI). The coefficient of correlation (r2) is relatively low (0.61). This is due to the differences in land management in each point of sampling. Land management could give very different crop yields although the land areas have relatively the same level of suitability. This identify the difficulty of validation against crop yields. It is almost impossible to find out different areas or farms of the same crop types with precisely the same land management practices. According to Young and Goldsmith, (1977), the differences in land management may lead to yield differences between farms as much as three- to five-fold in developing countries. For developed countries the differences may range from 30 to 40% between the best and the worst management practices (Dent and Young, 1981).

Figure 4. Graph depicting the correlation between LSIs and yields (ton/ha)

Map of yield pattern across the study region is given in Figure 5. The map was generated from interpolation of the data based on the equation (4). The map shows that with existing land management, the maximum possible production of corn in the region is 3170 kg/ha, and and the averaged figure is about 2500 kg/ha. This is somewhat above the averaged corn production in South Sulawesi Province, which is of 2200 kg/ha.

Figure 5. Map of existing yield pattern for maize in kg/ha

Spatial matching: LSI against existing land use
Spatial matching between Land Suitability Index (LSI) for cropping and existing land use (taken from Baja dkk., 2004; and Baja and Sallatu, 2005) is seen in Figure 6. The figure shows that land areas with existing forest have comparatively low averaged LSIs (around 0.60) for cropping, with high standard deviations. Shrubs have comparatively higher LSIs (0.80). As expected, mixed agriculture and paddy fields have very high LSIs (nearly 0.90), although the former is somewhat lower than that for the latter. This also applies to residential areas, in fact. Overall, the phenomenon seen in this spatial matching is that the higher the LSI the lower the variation of index (as depicted by standard deviation) over the study area. This spatial matching information will be invaluable for targeting key land management programs and identifying the on-site and off-site impacts associated with current and proposed land use.

Figure 6. Land use map derived from Landsat ETM+ images and the histogram depicting spatial matching between land use and LSIs.

Land management interpretation
Land management interpretation requires an accurate set of geo-information derived from precision agriculture tools. Land suitability indices (LSIs) and production map (yield patterns) give insights into:

The quality of land on a cell-by-cell basis
Possibility of tracing prominent biophysical constraints in land management
Yields expected in every cell (of the land) with certain level of land management.

In addition to agricultural production, other aspects that need to be considered in land evaluation are workability and sustained use of the land (FAO, 1976). From these perspectives, a land area with a high land suitability index (LSI) is expected to have not only the capability of producing high crop yields, but it should also be easily opened and managed, with a comparatively low cost, and more importantly the land will not become more vulnerable to environmental degradation when the nominated land use is continuously practiced.

4. Conclusion
The main aim of land suitability evaluation is to generate easily understandable information about the quality of land, allowing a judgment to be made on land use and land management. With a GIS, such information may be made available in a format that is easy to interpret, can be displayed in an interactive manner, and is amenable to many types of data processing, which subsequently enables updating. Furthermore, with a GIS, exclusion factors in the resource allocation problem can be easily dealt with. For instance, land areas where development exists can easily be excluded from the outputs, or land units having a very severe biophysical constraint in terms of management can be masked out with good positional accuracy.

With fuzzy set functions, a membership value is not expressed solely as being unity or zero, or being suitable or not suitable, it is given wide range of possible indices ranging from 0 to 1.0. Here, the characteristic rating procedures place an emphasis on the weights of individual memberships and the degree of closeness of memberships to an a priori determined ideal point to produce land indices with continuous grades. The significance of using a DEM in this model is also obvious. Use of a DEM has enabled better representation of spatial variability of slope limitations particularly in the context of land management for precision farming. This highlights the strength of fuzzy decision-based model compared with categorical methods. In the GIS environment, land indices with fuzzy boundaries can be subject to various database manipulations. They can be 'hardened' in many different ways, or combined with the results from empirical models (see Baja et al., 2002a) to derive a more robust outputs, or they can serve as an input to other resource-based models.

Acknowledgement
The authors would like to thank farmers in Kabupaten of Gowa and Kabupaten of Takalar for giving excellent response to our research team, and especially for allowing us to use some portions of their farm (harvested area) during harvesting season in March and May 2005. Financial support provided by The Ministry of Research and Technology, Republic of Indonesia through a competitive based-RUT XI program is greatly appreciated.

References

Baja, S. 2001. The quality of the land: using GIS for continuous-based land suitability assessment in the Sydney Region. GIS User, 44: 32-33.
Baja, S. 2005. Urgensi Penataan Kawasan dan Pengembangan Sistem Informasi DAS Jeneberang Berbasis Spasial. Buletin Alir, Edisi 4 (Feb 2005), Hal. 9-10.
Baja, S. and Dragovich, D. (To be published in 2006). Chapter 9: Fuzzy land evaluation. In: Odeh, I. O.A. and McBratney, A.B. (Eds). Fuzzy Environmental Sciences, Elsevier Science, New York.
Baja, S. dan Sallatu, A.M. 2005. Analisis Penutup Lahan Daerah Maros Sulawesi Selatan. Berita Inderaja LAPAN, Edisi Juli 2005.
Baja, S., Chapman, D. M., and Dragovich, D. 2001a. A conceptual model for assessing agricultural land suitability at a catchment level using a continuous approach in GIS. Pages 828-841 in: Proceedings of the Geospatial Information and Agriculture Conference, 17-19 July 2001, Sydney. NSW Agriculture, Sydney.
Baja, S., Chapman, D. M., and Dragovich, D. 2001b. Fuzzy modelling of environmental suitability index for rural land use systems: an assessment using a GIS. In: Proceedings of the Sixth International Conference on GeoComputation, 24-26 September 2001, Brisbane. The University of Queensland, Brisbane.
Baja, S., Chapman, D. M., and Dragovich, D. 2002a. A conceptual model for defining and assessing land management units using a fuzzy modelling approach in GIS environment. Environmental Management, Volume 29: 647-661.
Baja, S., Chapman, D. M., and Dragovich, D. 2002b. Using GIS and remote sensing for assessing and mapping the present status of land use and land qualities in the lower Hawkesbury-Nepean Catchment, Australia. Geocarto International, 17: 15-24).
Baja, S., Chapman, D.M. and Dragovich, D. 2002c. Using GIS-based continuous methods for assessing agricultural land use potential in sloping areas. Environment and Planning B: Planning and Design, Vol 29: 3-20.
Baja, S, Dragovich, D. and Chapman, D. 2005. Spatial Based Compromise Programming for Multiple Criteria Decision Making Modeling in Land Use Planning. Environmental Modelling and Assessment (In Review).
Baja, S., Ramli, M., and Jayadi, M. 2004. Laporan Perkembangan Tahun I Riset Unggulan Terpadu. XI. Kementrian Riset dan Teknologi, dan Lembaga Penelitian Unhas, Makassar. 59 halaman.
Burrough, P.A, and McDonnell, R.A. 1998. Principles of Geographical Information Systems. Oxford University Press Inc., New York.
Burrough, P.A., MacMillan, R. A. and van Deursen, W. 1992. Fuzzy classification methods for determining land suitability from soil profile observations and topography." Journal of Soil Science, 43: 193-210.
Carl Smith and Joanne McNeill . 2001. Using GIS to identify the mismatch between agricultural land use and land capability in Victoria. In Proceedings of the Geospatial Information and Agriculture Conference, 17-19 July 2001, Sydney. NSW Agriculture, Sydney.
Cook, S.E., and Bramley, R.G.V. 2001. Is agronomy being left behind by precision agriculture? In Proceedings of the 10th Australian Agronomy Conference, Hobart, 28 January -1 February. (ASA: Hobart.)
Davidson, D.A., Theocharopoulos, S.P. and Bloksma, R.J. 1994. A land evaluation project in Greece using GIS and based on Boolean and fuzzy set methodologies. International Journal of Geographic Information Systems, 8: 369-384.
Dent, D. and Young, A. 1981. Soil Survey and Land Evaluation. George Allen & Unwin, Boston.
FAO. 1976. A Framework for land evaluation. FAO Soils Bulletin no 32. Food and Agriculture Organisation of the United Nations, Rome.
Klingebiel, A.A., and Montmogery, P.H. 1961. Land Capability Classification. USDA Agricultural Handbook 210. US Government Printing Office, Washington D.C.
Research Centre, Hasanuddin University. 1999. Agricultural Soil Survey of Bili-Bili Irrigation Project Area. Research Centre, Hasanuddin University, Ujung Pandang. 89 pp.
Young, A., and Goldsmith, P.F. 1977. Soil survey and land evaluation in developing countries. Geographical Journal, 143: 67-74.
Zadeh, L.A. 1965. Fuzzy sets. Information and Control, 8: 338-353.