Distribution Change of Conifer and Broad-leaved Tree and Predict Future Distribution at Namsan (Mt.), Sangju

Abstract

Researches about analyzing and predicting distribution of forest are continuously proceeded in aspect of a forest management. To analyze the past distribution of forest, researchers record data about forest distribution and characteristics like forest inventory. However, these data couldnt provide long term data. Therefore, remote sensing was used to construct data. Landsat image which is remotely sensed data was used because Landsat image is appropriate to analyze the forest. Conifer and broad-leaved tree in Namsan, Sangju-si was analyzed for thirty years. As a result, distribution of broad-leaved tree was continuously increasing, on the other hand, conifer was decreased. Forest distribution was highly affected by human activity which is heating in the past. Therefore, broad-leaved tree which is more appropriate for fire wood was distributed in small area. Predicting future distribution was proceeded based on modelling method. Researches that based on niche model predict suitable habitat, and researches based on process-based models and demographic models predict real distribution. However, these methods have limitations in forest management because niche model cannot predict real distribution and process-based model and demographic model predict in large scale because of the scale of input data. This research aims to overcome this limitation by predicting future distribution of conifer and broad-leaved tree. Because conifer and broad-leaved tree are typical species which is in competition and conifer is weak competitor. In this research, present distribution of conifer and broad-leaved tree and replacement probability of conifer by broad-leaved tree was used to predict future distribution. Probability of conifer by broad-leaved tree was modelled based on logistic regression model using forest distributions from past to present. Remote sensing was used to construct data because forest distribution is changing slowly and satellite images provide long periodic data. Furthermore, Landsat images were selected because of fine spatial scale and long temporal extent. Past distribution map was constructed by using classification method. Comparing past distribution of conifer and broad-leaved tree maps for the periods 1984-1995, 1995-2005, and 2005-2014, classes that represented either a conifer to broad-leaved tree or conifer to conifer change were generated. For logistic regression, distribution changed maps were used for dependent variable and distance from broad-leaved forest edge, elevation, slope, topographic wetness index (TWI), annual solar radiation were used for independent variable. Compare the result of distribution changed map with previous researches, distance variable which used in this research seems suitable factor to predict distribution change. Most replacement of conifer by broad-leaved tree was occurred near broad-leaved forest edges and decreases sharply similar to seed dispersal and seed density pattern of other researches. However, distance was calculated based on 30m spatial resolution, therefore, the distance from broad-leaved forest edge has uncertainties. To overcome this uncertainty, Monte Carlo simulation was used. According to simulation result range of distance value was considered. As a result of logistic regression, annual solar radiation and distance from broad-leaved forest edge were turn out to be powerful factor to predict replacement probability of conifer by broad-leaved tree. In other words, replacement probability of conifer by broad-leaved tree was increased where close to broad-leaved forest edge and annual solar radiation is low. It reflects the seed dispersal and density of seed that density of seed is higher near the broad-leaved forest edge. In addition, it reflects the characteristic of conifer and broad-leaved tree that shade tolerance of conifer is weaker than broad-leaved tree. Future distribution of conifer and broad-leaved tree was predicted by using the result of logistic regression model. Distribution of conifer will decrease slowly than before. broad-leaved tree population curve seems similar to sigmoid curve which known as population growth model. It is considered that forest area comes to limited resource, therefore, competition was occurred between conifer and broad-leaved tree. As a result, distance was turn out to be an important variable to predict future distribution. Using the distance from broad-leaved forest edge, it is possible to predict future distribution of conifer and broad-leaved tree. Furthermore, to overcome the uncertainty due to spatial resolution, it is possible to use Monte Carlo simulation.