像元尺度上不确定性对空间景观直观模型模拟的影响

The effect of cell-level uncertainty on spatially explicit landscape model simulation

L ANDIS模型是模拟自然和人为干扰下森林景观变化的空间直观景观模型。模型把景观概念化为由相同大小的像元或样地组成的格网。在每一个像元上 ,模型要求输入物种和年龄组信息。但是 ,由于研究区一般由成千上百万个像元构成 ,不可能通过实际调查获取每一个像元上的物种和年龄组信息。因此 ,采用了一种基于小班的随机赋值法从森林调查数据中获取每一个像元的物种和年龄组信息。该方法是一种基于概率的方法 ,会在 L ANDIS模型模拟的物种和年龄组信息的输入中引入不确定性。为了评价由基于小班的随机赋值法所引入像元尺度上的不确定性对模型模拟结果的影响 ,用蒙特卡罗模拟法进行不确定性分析。对 L ANDIS模型模拟的每一个物种 ,用众数年龄组发生频率来定量化单个像元上年龄组信息的不确定性 ,用所有像元上的众数年龄组平均发生频率来定量化年龄组信息在像元尺度上总的不确定性。平均发生频率越高 ,不确定性越低。为了评价基于小班的随机赋值法对景观尺度上模型模拟结果的影响 ,计算了每一个物种在整个研究区内的面积百分比和聚集度指数。变异系数越大 ,不确定性越高。对所有物种 ,年龄组信息不确定性在模型模拟的初期是比较低的 (平均发生频率大于 10 )。种子传播、建群、死亡和火干扰使模型结果的不确定?

LANDIS is a cell-based spatially explicit forest model designed to explore the succession dynamics under the natural and anthropogenic disturbances. At each cell, species and age cohort information is required and providing such information for a landscape comprising millions of cells is challenging. In this study, a stand-based assignation (SBA) approach is developed to stochastically assign species and age information to each cell based on the forest inventory data. The algorithm assumes that each cell in a stand has a probability of being assigned with a species, which is determined by the relative occurrence of the species within the stand (0~1). As a probability-based approach, SBA will introduce errors in LANDIS input. In order to assess the effect of errors produced by SBA on LANDIS results, we conducted 20 Monte Carlo simulations to assess the uncertainties associated with model outputs at cell level and landscape level. For each species simulated in LANDIS, the recurrence frequency (RF) of the majority age cohort (the most frequently occurring species age cohort) from 20 Monte Carlo simulations are used to quantify the uncertainty in the age cohort information for each individual cell. Average recurrence frequency (ARF) of the majority age cohorts is used to quantify the overall cell-level uncertainty for each species age cohorts . Higher RF and ARF values indicate lower uncertainty. In order to examine effects of uncertainty at the cell level on the simulation results at the landscape level, we also calculated percent area (PA) and aggregation index (AI) for each species from the species distribution map in the LANDIS output. PA is the percent of the area occupied by a certain species in the study area and AI is a class specific landscape index used to quantify the spatial aggregation of classes. For each species, the coefficient of variation (CV) for PA and AI for the 20 Monte Carlo simulations was used to quantify the variability of species abundance and its spatial pattern at landscape level. Both PA and AI were calculated using APACK, software for calculating landscape metrics. Higher CV values indicate higher uncertainty. Results show that uncertainties at cell level is relatively low at the beginning of the simulation for all species (ARF is larger than 10). Seed dispersal, seedling establishment, mortality, and fire disturbance cause uncertainty to increase with simulation year. The uncertainty eventually reaches an equilibrium state where input errors in original species and age information have no effects on the simulation outcomes. At the landscape level, species percent area and their spatial pattern measured by the aggregation index are not significantly affected by uncertainties of species age structures at the cell level. Since the purpose of the LANDIS model simulation is to predict the overall landscape pattern change, not a single event, SBA can be used to parameterize species and age information for each cell.