Ioifrst Campy Iei Lvitv Clnss

Figure 13: Variation in forest canopy density classes extracted from different techniques

>80S «10BOW -WtoSOii 10tu40« < :o% CioD-fDifTt

The accuracies of density maps generated from all the methodologies have been assessed to validate the technique. Comparison of the density stratification accuracies have been carried out with reference to ground control points and different techniques. Accuracy is estimated from the confusion matrix over the training class, in terms of percentage of number of correctly classified category against the total number of classes, viz., class 1 (> 80 %), class 2 (60 - 80 %), class 3 (40 - 60 %), class 4 (20 - 40 %) and class 5 (< 20 %).

It has been observed that overall classification accuracy giving satisfactory results, FCD mapper semi-expert system shows 80.21% accuracy followed by object oriented image analysis of 87.50% and 71.88% respectively. The correlation coefficient value of FCD model with visual interpretation and image segmentation are found to be 0.95 and 0.84 respectively.

Delineation of forest vegetation from the other objects is considered to be very important factor for precise analysis of forest change detection and landuse processes. FCD model shows acceptable degree of delineation of forest vegetation from the other non-forest classes. However, same inputs used for analysis in image segmentation of eCongnition v2.1 and unsupervised cluster analysis of Erdas Imagine v8.5 shows boarden inter mixing of forest vegetation with other classes (Fig. 14).

Figure 14: Comparison of various methods used for forest and non-forest separation. Red circles indicates fraction of the delineation of forest from non-forest

The time factor is also considered in the analysis of the models, there is a big difference in the time taken for forest canopy density mapping in the techniques adapted. Visual interpretation took 4 days where as FCD mapper semi expert system took only half a day to complete the job. However, about half of the times of visual interpretation have been spent on object oriented image analysis. Overall analysis and assessment of the techniques used in the present study indicate that FCD semi-expert shows satisfactory results. It requires less manpower and limited ground checks. Therefore FCD model would be a very useful tool especially for foresters for better monitoring and management of forests for the future. Detailed and accurate maps of forest condition and structure are a necessity for rigorous ecosystem management. Forest cover type map along with density maps are the fundamental source of information for fire behaviour modeling, animal habitat management, prediction and mapping of forest insect infestations, and plant and animal biodiversity assessment.

Was this article helpful?

0 0

Post a comment