2.5 Prediction of LULC change
Based on the classified maps, this study also attempts to predict LULC for 2023 and 2027. Three steps are involved in the prediction process. They are estimation of transition probabilities, creation of transition suitability maps and finally predicting LULC. A combination of Markov chain with cellular automata (CA) method is employed as former technique is unable to provide spatial dimension of a phenomenon. To simulate future land covers, actual data of 2017 2018 are used to predict 2019 LULC which is then compared with observed data of 2019 to check the effectiveness of model.
Transition probability matrix is derived through markov module as a first step. LULC thematic maps of different periods are inputted to estimate transition probabilities (Pijanowski et al., 2002). A suitability map for each of LULC class defines transformation suitability of a certain class from all other categories (Halmy et al., 2015). Stressor and stimulus parameters are, therefore, required to develop suitability map to account dynamic aspect of land cover change. In this work, forest degradation is based on both stressor and stimulus parameters. Stimulus variable includes number of Rohingya population, stressor parameter comprises high elevation, and constraint is defined by highly protected areas. The stressor, constraint and stimulus variables are determined on the basis of previous studies (e.g., IOM and FAO, 2017; IUCN Bangladesh, 2018), 2018 field works and local knowledge of the sites (Table 3). Since not all LULC classes are subject to change rapidly, six dynamic (Table 3) and one constraint variables are included to isolate suitable locations or forest patches that could be degraded under the influence of refugee occupancy.
As degradation of forest is accelerated by fuelwood collection and illegal logging by the Rohingya communities, distribution of Rohingya population is a key factor for a suitability map. Apart from population variable, four distance variables (Table 3) are also considered. Due to the fact that the Rohingya can travel up to 16 km (IOM and FAO, 2017), and on average, 7 km to collect forest resources, a 7 km buffer is constructed using center of each refugee camps. These buffers are then intersected with population distribution to identify number of people that can conceivably influence forest degradation. In other words, if a forest area is within a distance of 7 km buffer of three camps (C1, C2 and C3) and these camps contain 100, 200 and 150 people, then a particular forest cover has a total of 450 humans. These populations are considered as potentially degraders. The results are subsequently aggregated to a 100x100 m grid based on which a ranking is performed. This helps determining forest covers subject to degradation due to existence of the Rohingya communities. The higher the population in each grid, the greater the likelihood of a forest to be degraded. In the creation of forest degradation suitability maps, maximum weighting (0.5) is assigned to population field whereas other parameters receive rest of the weights (0.5), using a scale of 0-1. A weighted linear combination method is then employed to develop transition suitability maps.
The transition probability or transition suitability maps of 2017 2019 are considered, wherein 2019 LULC is used as base. Since CA Markov provides spatial distribution of LULC change, area of each class to be changed to other classes are determined by transition potential or transition suitable maps (Halmy et al., 2015). These transition areas are divided by the number of time periods in the simulation (1, 4 and 8 in this case). This operation provided areas to be converted to another LULC class. The CA Markov with these principles results predicted LULC data which are then assessed for accuracy by considering kappa index of agreement and disagreement. LULC prediction for the year of 2023 and 2027 are conducted, based on actual data of 2019 (Pontius and Millones 2011).