A Land Evaluation System Using Artifical Neural Network Based Expert's Knowledge and GIS
Land Suitability Classification using the Trained ANNs
The trained neural networks were used to classify the land suitability for a desired land use type of an area in Quirino province, Philippines. Figure 3 shows the summary for the classification process. The user interface involves the determination of summary of the classification process. The user interface involves the determination of the land use type and the selection of the appropriate trained neural networks in the knowledge base. The physico-chemical soil parameters were automatically extracted from the GIS as shown in Figure 2 for the land suitability class determination. Through these processes, the different land suitability class maps corresponding to the different land use type were generated. These maps were then used to determine the land use type which has highest suitability class maps corresponding to the different land use types were generated. These maps were then used to determine the land use type which has the highest suitability class (using overlay method) which will then be selected as the recommended type of land use.
Figure 3. Automated land evaluation using the trained neural networks.
Result and Discussion
The land evaluation knowledge base in the form of trained neural network was successfully generated. This involved the use of the different crop requirement tables (following the FAO land suitability classification framework) of the different land use types for the different land use types for the training of the ANNs. The trained ANNs were successfully used for the land suitability classification of the different land use types of the study area in the northern Philippines. This is clearly shown in the resulting maps in Figure 4. Indeed, it is very apparent that the differences between figures 4a and 4b, showing the land suitability maps for com cultivation obtained using the rule based method and the ANN, respectively, are almost indistinguishable. The same is true in figures 4c and 4d, showing the land suitability maps for rice cultivation using the two methods. Indeed, using the neural network data base, the system was able to give the best possible land use type of the different land units in the study area.
Figure 4. Land suitability maps for rainfed corn cultivation obtained (a) using the rule based method
and (b) the ANN land suitability maps for rainfed rice cultivation obtained using the (c) rule
based method and (d) the ANN.
Conclusion
A methodology for the transformation of a land evaluation expertise form a rule base to trained ANNs was successfully developed. This includes the design of the ANN architecture and training scheme. The use of these trained ANNs during the actual land evaluation was successful. A GIS linked automated land evaluation system was successfully developed with knowledge base composed of the trained ANNs.
Reference
-
Alexander, I. and H. Mortion. 1990. An Introduction to Neural Computing. Chapman and Hall, London. 237 pp.
- Bandibas, J.C. 1986. The automated land evaluation using artificial neural network based expert's knowledge, GIS and remotely sensed data. Spatial and spectral signatures for satellite image classification. PhD. Thesis. Geologisch Institute, University Gent, Belgium. 201 pp.
- Rumelhart, D.E., G.E. Hintron and R.J. Williams. 1986. Learning Internal Representations by Error Propagation in Parallel Distributed Processing. MIT Press, Massachusetts. Vol. 1 and 2.
