# Spatial Model of Forest and Land Fires Hazard Level

Analysis of stepwise regression of the six variables stated in Table 22.2, yielded three non significant variables (a > 0.1), i.e., distance from river (X2), distance from village centre (X4) and land system (X6) with an R2 value of 40%. Consequently, the composite score model (hazard score) to develop forest and land fires hazard score used the three significant variables namely land cover (X1), distance from road (X3) and peat depth (X5), to estimate the hotspots density. Linear regression analysis of these three variables resulted in R2 of 42.2% with each variable weight shown in Table 22.3.

From the variables weights in Table 22.3, a model equation for forest and land fires hazard score in Kapuas District, Central Kalimantan Province was formulated as follows:

where:

x1 is rescaled score of land cover

x3 is rescaled score of distance from road

x5 is rescaled score of peat depth

From the mathematical equation above, the weights related to human activity and biophysical variables could be calculated. In Kapuas District, human related variables (distance from road and land cover) had a weight of 27.0% while the biophysical variables (peat depth) had a weight of 73.0%. According to the model with three variables, in predicting fire activity, peat played the most important role in forest and land fires, with weighted value of almost 75%. This indicated that peat- land were fire-prone areas in Kapuas District, hence it should become the main focus in controlling forest and land fires. In addition, land cover and distance from road were also important variables in the prevention of forest and land fires. Thoha et al. (2014) have found that peatland in conservation areas and unmanaged lands in Kapuas District were fire-prone.

Other associated human variables in this study, were land cover, settlements, mining, rice fields, agriculture dry lands and peatland as shown in Table 22.1. Samsuri et al. (2010) have found that land cover had a high weight in determining

 Variable Coefficient Weight Land cover 0.001440 0.16118 Distance from 0.000974 0.10902 road Peat depth 0.006520 0.72980

Table 22.3 Coefficient score and weight of composite score of forest and land fires hazard level in Kapuas the level of forest and land fires hazard in Central Kalimantan. Road is very important factor in the occurrence of forest and land fires because it opens people’s access to use the land. While in West and East Kalimantan, roads, rivers and village centre are the dominant variables in determining fire risk in (Jaya et al. 2008). Thus, both human related and biophysical variables were interconnected. Biophysical related variables provided the fuels and the means of fire to occur, while human related variables were significantly determined the occurrence of fire on areas with specific characteristics.

Currently, research on the determination of fire-prone areas has been developed using a wide variety of maps. Jaya et al. (2008) research on the spatial model of forest fires hazard in Central Kalimantan using Composite Mapping Analysis (CMA) method, and found that human related variables such as distance from village and distance from road have contributed as much as 52% as compared to biophysical variables, namely land cover, that contributed only 48%.

Determination of the forest and land fires hazard levels and zones were done using a quantitative approach (empirical) by means of Composite Mapping Analysis method. The model was built based on composite scores, which was composed through statistic equations that described the relationship between the number of hotspots per km2 (y) with composite scores (hazard scores) of each corresponding variable (x), as follows:

where:

Y: Hotspot density X: Composite Score Model

The established forest and land fires vulnerabilities model was a polynomial model with determination coefficient (R2) of 73.8%, meaning that the model resulted from the relationship of composite score model with hotspot density was adequately accurate. The accuracy test of forest and land fires hazard class showed that the classification of five classes into three classes improved the accuracy from 57.7 to 78.8% (Table 22.4). Thus, in developing forest and land fires hazard level, the level should be divided into three classes.

Extent of each fire hazard level, composite score, hotspots density and fire hazard map are presented in Table 22.5 and Fig. 22.9 by dividing the Y score into three classes.

 Number of classes Accuracy 5 classes 57.7 3 classes 78.8

Table 22.4 Coincidence matrix of Z model and hotspot from observation results

Table 22.5 Extent of fire with three levels of hazard

 Hazard level Extent (ha) Composite score Hotspot density Area percentage (%) Low 1133930.4260 10.00-27.91 0.082-0.0291 69.3161 Moderate 412345.0360 27.91-42.10 0.091-0.539 25.2063 High 89607.4460 42.10-100.00 0.539-1.678 5.4776

Fig. 22.9 Hotspot distribution in Kapuas