Drivers of fire in the boreal forests: Data constrained design of a prognostic model of burned area for use in dynamic global vegetation models
Drivers of fire in the boreal forests: Data constrained design of a prognostic model of burned area for use in dynamic global vegetation models
复制标题
DOI：
10.1029/2006jd008372
复制
发表时间：
2007-12-28
影响因子：
4.4
通讯作者：
Pacala, Steve
中科院分区：
文献类型：
--
作者：
Crevoisier, Cyril;Shevliakova, Elena;Pacala, Steve
[1]北方地区是全球碳循环的一个重要组成部分，因为它们拥有大量的地上和地下碳储量。由于北方森林的演变与火灾制度密切相关，气候变化可能会引起生态系统的变化，可能导致碳和其他微量气体大量释放到大气中。因此，预测这种潜在的气候反馈对地球系统的影响是重要的，需要在动态全球植被模型(DGVM)中将火作为气候驱动的过程进行建模。在这里，我们开发了一个新的基于数据的预测模型，用于DGVMS，根据北方森林的四个气候(降水、温度、土壤水分和相对湿度)和一个与人类有关的(道路密度)预测因子来估计每月的燃烧面积。烧毁面积模型是当前气候条件的函数，因此对气候变化作出反应。模型参数是使用马尔可夫链蒙特卡罗方法估计的，该方法应用于加拿大大型火灾数据库的地面观测。该模型根据来自三个北方地区的独立观测数据进行了验证：加拿大、阿拉斯加和西伯利亚。提供了现实的气候预报因子，该模式能够再现烧毁地区的季节性、强度和年际变化，以及火灾事件的地点。特别是，该模型很好地模拟了燃烧事件的时间，三分之二的事件预测为正确的月份，其余几乎所有事件都是在观察到的事件之前或之后一个月预测的。预测的年烧毁面积在目前各种估计的范围内。在一个网格单元中估计的年相对误差(标准差)为12%，这使得该模型适合于定量研究火烧面积随气候的演变。
[1] Boreal regions are an important component of the global carbon cycle because they host large stocks of aboveground and belowground carbon. Since boreal forest evolution is closely related to fire regimes, shifts in climate are likely to induce changes in ecosystems, potentially leading to a large release of carbon and other trace gases to the atmosphere. Prediction of the effect of this potential climate feedback on the Earth system is therefore important and requires the modeling of fire as a climate driven process in dynamic global vegetation models (DGVMs). Here, we develop a new data- based prognostic model, for use in DGVMs, to estimate monthly burned area from four climate (precipitation, temperature, soil water content and relative humidity) and one human-related (road density) predictors for boreal forest. The burned area model is a function of current climatic conditions and is thus responsive to climate change. Model parameters are estimated using a Markov Chain Monte Carlo method applied to on ground observations from the Canadian Large Fire Database. The model is validated against independent observations from three boreal regions: Canada, Alaska and Siberia. Provided realistic climate predictors, the model is able to reproduce the seasonality, intensity and interannual variability of burned area, as well as the location of fire events. In particular, the model simulates well the timing of burning events, with two thirds of the events predicted for the correct month and almost all the rest being predicted 1 month before or after the observed event. The predicted annual burned area is in the range of various current estimates. The estimated annual relative error (standard deviation) is twelve percent in a grid cell, which makes the model suitable to study quantitatively the evolution of burned area with climate.