GISdevelopment.net ---> Application --- > Urban Planning --- > Overview

Accuracy of Land Suitability Modeling using Spectral Characterization

Sutat Dansagoonpon
STAR program, Asian Institute of Technology (AIT)
P.O.Box 4 Klong Luang, Pathumthani 12120, Thailand
Email: st027123@ait.ac.th

Nitin K Tripathi*, and Roberto S. Clemente**
* STAR program, Asian Institute of Technology (AIT)
** WEM program, Asian Institute of Technology (AIT)
Email: nitinkt@ait.ac.th, clemente@ait.ac.th



1. Introduction
Rapid increase in the demand of Natural Rubber (NR) caused expansion of the rubber plantation in Thailand at a very fast rate to non-traditional and unsuitable areas with limiting conditions. These plantations produced lower yield despite more attention. This resulted in higher cost of production. At the same time, there is a rapid increase in domestic consumption of palm oil. Therefore, it is not surprising to see many rubber plantations switch to invest in oil palms now. Now it is feared that these may also expand to unsuitable areas. The objective of the study is to develop a methodology for the decision maker to replace non-suitable crop by the suitable crop. In this study the case of rubber and oil palm is considered. This was carried out using GIS and multi-factor evaluation. Analytical Hierarchical Processing (AHP) and Pairwise Comparison Method were used for factors weighting. A land potential for rubber and palm oil production are evaluate based on the crop requirements for rubber and oil palm, and climatic and physical-chemical soil properties, which will allow the prediction of yields and crop substitution between rubber and oil palm. In order to get a better agricultural production.

2. Study area and data used

Generalities
Krabi province study area is situated between latitude 7 ° 22 ¢ to 8° 41¢ north and longitude 98° 21¢ to 99° 19¢ east, about 990 km from Bangkok by road. The total area is approximately 4708.51 km2. Climatic type of Krabi province is tropical monsoon climate “Am” according to Köppens’ (1931) classification described by Steele et al (1972) or “B2A’ra” (Hermid megathemal with season of litter or no water deficiency and a temperature efficiency regime normal to full megathermal) according to Thornth Waite’s classification. Annual rainfall is around 2,379.9 mm. Mean temperature is 27.4oC. Relative humidity is 80.0 %. Soil classification according to USDA soil taxonomy was recognized 5 orders namely: Spodosols, Ultisols, Alfisols, Inceptisols and Entisols.

Data used
Modified climatic and land characteristics requirements for rubber and oil palm production according to Somyot (1992), Nakorn et al. (1998) and Sutat et al. (1999). The 7 important factor such as; soil depth, soil texture, ground water table, soil drainage, organic carbon, slope and growing period or water deficit were classified into 3 classes.

GIS-database of study area including: 1. Digital soil series map, which are consist of 98-land unit and its profile description according to USDA format. Physical and chemical soil analysis of all land unit report. 2. The attribute data consist of digital slope, elevation, aspect, contour, soil data, geology, stream, transportation, factory, climatic data shape file etc.

3. The Analytic Hierarchy Process (AHP)
The purpose of weighting is to express the importance or preference of each factor relative to other factor affect on crop yield and growth rate. Since the land physical characteristics such as slope, soil texture, soil depth, ground water table depth, and soil drainage, etc., are uncorrected factors. These very severe factors should be considered as the first priority. Based on crops requirement, questionnaire with soil scientist senior researcher and experience, the multi-factor priority model was structured.

To avoid and reduce the individual biases of factor weighting, the weights in the study were determined by using a pairwise comparison method as developed by Saaty (1980) in the context of the analytical hierarchy process (AHP). Pairwise comparisons are based on forming judgments between two particular elements rather than attempting to prioritize an entire list of elements. A matrix is constructed, where each factor is compared with the other factors, relative to its importance, on a scale from 1 to 9. Then, a weight estimate is calculated and used to derive a consistency ratio (CR) of the pairwise comparisons. If the CR > 0.10, then some pairwise values needs to be reconsidered and the process is repeated till the desired value of CR < 0.10 is reached.

The pairwise comparison method showed that the consistency ratios (CR) of Rubber and Oil Palm were less than 0.1 - Rubber 0.06 and Oil Palm 0.02 (Table 1). This indicates that the comparisons of each factor were perfectly consistent, and the relative weights were suitable for use in the GIS multi-factor evaluation.

Table 1 Relative weights of factors affecting crop yield and growth rate
Rubber Oil Palm
Factor Weight Factor Weight
Soil Depth 0.3 Ground Water Table Depth 0.21
Ground Water Table Depth 0.3 Soil Texture 0.21
Slope 0.17 Water Drainage 0.21
Water Drainage 0.07 Slope 0.21
Soil Texture 0.07 Soil Depth 0.09
Growing period for Tree 0.07 Water deficit of Area 0.05
OC (Organic Carbon) 0.03 OC (Organic Carbon) 0.03
CR = 0.06 CR = 0.02

4. Spatial analysis for land evaluation
In this step, the 7 important factors were mapped and classified into 3 classes (R1, R2 and R3 for rubber and P1, P2 and P3 for oil palm) namely; growing period map or water deficit map, soil depth map, water table map, slope map, soil drainage map, soil texture map, and organic carbon map. Ranging of these R1, R2, R3, P1, P2 and P3 of each factors were done according to Somyot (1992), Nakorn et al. (1998) and Sutat et al. (1999). The relative weight of factor obtained from Table 4 was used for maps (factors) weighting. Then rubber suitability map and oil palm suitability map were generated. (See Figure 1 and 2)


Fig. 1 Rubber suitability map after weighted evaluation factor


Fig. 2 Oil Palm suitability map after weighted evaluation factor

5. Classification accuracy assessment
This step the satellite image, NDVI image and soil map of the study area are linked together with same coordinate point (x,y) or same latitude – longitude by computer software. This satellite image is the false color composite of Landsat 5 TM image using band 4 red color, band 5 green color and band 3 blue color respectively. The NDVI* image was calculated by using landsat 5 TM satellite image band 3 and band 4. Soil map of the study area in digital format and soil profile description of soil types in this study area refer to soil survey of Krabi province that was done by soil survey staff, department of soil development Thailand, reported in year 1986. These 2 images and 1 map are taken into account for sampling regions of interest (ROIs) at existing crop and transferred to new image (ground truth image) by digitizing at the same coordinate. Now all information that need for accuracy assessment is linked together.

The classification accuracy assessment was done by compare those rubber and oil palm suitability classified image and their ground truth image. Table 2 the ground truth (percent) shows the suitability classes for rubber distribution in percent for each ground truth class. The producer accuracy indicated that in R1, R2 and R3 classes the percent classified correctly are 91.57, 100 and 50.0 % respectively. Mean while the user accuracy indicated that the R1, R2 and R3 classes, pixel labeled by classifier are classified correctly 96.2, 74.36 and 100 % respectively. The overall accuracy is 89.5161 %. Kappa Coefficient with value 0.7883 is interpreted that a classified image achieves an accuracy that is 78.83 percent better than the chance assignment of pixels to categories.

The ground truth (percent) table 3 shows the suitability classes for oil palm distribution in percent for each ground truth class. The producer accuracy indicated that in P1, P2 and P3 classes the percent classified correctly are 95.4, 67.8 and 60.0 % respectively. Mean while the user accuracy indicated that the P1, P2 and P3 classes, pixel labeled by classifier are classified correctly 80.5, 93.02 and 60 % respectively. The overall accuracy is 83.4437 %. Kappa Coefficient with value 0.6653 is interpreted that a classified image achieves an accuracy that is 66.53 percent better than the chance assignment of pixels to categories.

Due to ground truth classification of land suitability was based on existing crops. The planting of both rubber and oil palm on unsuitable condition in this study area are rare. A very few number of pixel of unsuitable R3 and P3 can be labeled, this may cause the producer accuracy of R3 and P3 seem rather low. However, if consider the overall accuracy, this new methodology can be accepted.

Table 2 Rubber suitability classification accuracy
Ground Truth (%)
Class R3 R2 R1 Prod. Acc (%) User Acc (%)
R3R2R1 50.025.025.0 01000 08.4391.57 50.010091.57 10074.3696.2

Overall Accuracy = (111/124) 89.5161 % , Kappa Coefficient = 0.7883

Table 3 Oil Palm suitability classification accuracy
Class P3 P2 P1 Prod. Acc (%) User Acc (%)
P3P2P1 60.0040.0 1.6967.830.51 1.153.4595.40 60.067.895.4 60.093.0280.58

Overall Accuracy = (126/151) 83.4437 %, Kappa Coefficient = 0.6653

6. Calculating Vegetation Indices
Use the NDVI (Vegetation Index) to transform multispectral data into a single image band representing vegetation distribution. The NDVI (Normalized Difference Vegetation Index) values indicate the amount of green vegetation present in the pixel—higher NDVI values indicate more green vegetation. Calculating NDVI uses the standard algorithm: (Valid results fall between -1 and +1).

The calculation of NDVI by randomly points sampling inside actual rubber plantation at the end of February. The results shown that NDVI of R1 are range between 0.67-0.736, R2 are range between 0.574-0.673 and R3 0.551-0.64 respectively. (See Figure 4) The calculation of NDVI by randomly points sampling inside actual oil palm plantation at the end of February. The results shown that NDVI of P1 are range between 0.719-0.777, P2 are range between 0.638-0.714 and P3 0.627-0.666 respectively. (See Figure 5)


Fig 4 The relationship between actual rubber plantation NDVI and rubber land suitability classe


Fig 5 The relationship between actual oil palm plantation NDVI and oil palm land suitability classes

7. Crop suitability modeling.
Finally the rubber suitability and oil palm suitability maps were superimposed, then all polygon were union together to get a model for the area selection of those two crops. The resulting crop suitability model (Fig. 6) shows 9 land suitability classes resulting from the combination of highly (R1, P1), moderately (R2, P2) and not suitable (R3, P3) areas for rubber and oil palm planting.

Conclusion
The results obtained from this study indicate that our new method gives better result than the older methods. The application of GIS and multi-factor evaluation using AHP could provide a superior database and guide map for decision makers considering crop substitution between rubber and oil palm in order to achieve better agricultural production.

One somewhat notable limitation of this new method is that it requires a knowledgeable assessment of the relative importance of all factors. This can be carried out to some extent using the pairwise comparison method, but to rate the relative preferences level for two factors using the scale for pairwise comparison still needs priority-setting experience for the specific plant from senior researchers. To set the correct relative preference level without personal bias in this study required not only reliable soil theory information and the writer’s experience, but also the experience of a number of senior researchers.


Fig 6 Crops substitution model for Krabi, Thailand

References
  • Nakorn Sarakoon, Somyot Sinthurahat, and Sutat Dansagoonpon. 1998. “Agro-ecological Zoning Analysis for Oil Palm in The South of Thailand ”. Horticulture Research Institute and Rubber Research Institute, Dept. of Agriculture. 266 p.
  • Saaty, T.L.1980, The Analytic Hierarchy Process, McGraw-Hill, New York.
  • Somyot Sinthurahat. 1992. “Elaboration of Land Evaluation Model of Rubber Cultivation in Peninsular Thailand”. PhD Thesis, ITC, State University of Ghent, Belgium. 261 p.
  • Sutat Dansagoonpon, and Somyot Sinthurahat. 1999. “Agro-ecological Zoning for Rubber (Hevea Brasiliensis) Plantation in Southern Thailand Using Land Evaluation, Remote Sensing and GIS”. Rubber Research Institute, Dept. of Agriculture. 242 p.
  • Sys, C., Van Ranst, E., and Debaveye, J. 1992. “Land Evaluation. Part I (Principles in Land Evaluation and Crop Production Calculation)”. International Training Center (ITC) for Post-Graduate Soil Scientists. State University of Ghent, Belgium, 274 p.
© GISdevelopment.net. All rights reserved.