Development of Land Suitability Evaluation System for Coastal Aquaculture Using Artificial Neural Network and Geographical Information Systems (Case Study: Mahakam Delta, East Kalimantan)
Sutarga, I Ketut
MetadataShow full item record
Land suitability is the aptitude of given type of land to support a certain use. Land suitability is one of key factors for the sustainability of aquaculture system. It can be determined by using matching methods between land suitability criteria and land characteristics, such as topography, hydrography, climate, water characteristic, risk, land cover, and spatial plan aspects. Evaluation of land for aquaculture suitability can be reached using a parametric approach, which is implemented by using the distinguish land characteristics and combination of it. Parametric approach uses the numeric value, which classifies the land based on individual characteristics. The objectives of the research are to develop a land evaluation system on aquaculture suitability by using artificial neural network (ANN) and geographical information system (GIS) and to evaluate the performance of the new developed system as compared to GIS vector spatial analysis method. Four types of suitability were defined, i.e. highly suitable (S1), moderately suitable (S2), marginally suitable (S3) and not suitable (N). The new developed GIS model uses the raster format and ANN classifier. The performance of GIS model was then evaluated, and compared to the vector map overlay that used GIS spatial analysis features, including suitability distribution, covered area, processing speed and validity. The multilayer feedforward of the ANN with backpropagation learning algorithm was chosen to do the classification. GIS overlaid layer in form of spatial database was then converted into training dataset, validation dataset and prediction dataset. Training and validation dataset contains a pair of input and output, while the prediction dataset contains input data only. To compare the two methods, the Mahakam Delta-East Kalimantan were selected as a case study. Ten thematic map layers of land characteristic were involved in both methods, i.e.: water salinity, dissolved oxygen, water pH, soil texture, soil drainage, distance to rivers, pollution risk, land cover type, land use plan and annual rainfall intensity. Each feature of thematic map layers were classified and scored using 1 to 4 ranges from worst to best, respectively. The performance of ANN together with GIS was reliable, it includes the accuracy on training dataset of 97%, and validation of 96%. Distributions of each suitability of both methods are located on the same location in the study area. Area covered in the study area was 188,666.19 ha, consisting of 21,202.20 ha (11.24%) of 3 S1, 87,190.83 ha (46.21%) of S2, 6,958.26 ha (3.69%) of S3 and 61,251.12 ha of N. Only 0.19% of total area represents the “Unclassified Area” and 6.20% is “No Data”. Unclassified areas may be caused by less representative training dataset. “No Data” class represents the area that was not included in the data processing by raster data model. The research also proved that ANN together with GIS raster do the classification faster than vector spatial analysis.