A novel model to accurately predict continental-scale timing of forest green-up
A novel model to accurately predict continental-scale timing of forest green-up
复制标题
DOI：
10.1016/j.jag.2022.102747
复制
发表时间：
2022-04
期刊：
影响因子：
--
通讯作者：
N. Neupane;M. Peruzzi;A. Arab;S. J. Mayor;J. Withey;L. Ries;A. Finley
中科院分区：
文献类型：
--
作者：
N. Neupane;M. Peruzzi;A. Arab;S. J. Mayor;J. Withey;L. Ries;A. Finley
植被绿度的年度循环是生态系统进程的最重要驱动因素之一。植被恢复和衰老时间的预测模型对于理解生物群落如何应对全球变化至关重要。Greenup的时间与气候密切相关，还跟踪气温的年度变化，这种关系的强度在时空上是不同的。当地的研究有助于理解潜在的机制，但不足以解释更大范围的变化。使用遥感数据的大规模研究有可能利用区域动态，即使潜在的机制尚不清楚，但使用这些方法的预测能力很低。在这里，我们通过一类新的贝叶斯回归模型来预测北美东部的植被物候。我们的建模框架利用中分辨率成像光谱仪(MODIS)的卫星观测，提供了高精度的大陆级峰值绿灯时间预测。除了考虑单个站点的时间结构外，我们的模型还利用了整个研究范围的信息，而不考虑它们的空间邻近程度。模型建立于2000年至2016年，并显示出较高的预测精度(R2>；=95%)。对2017年和2018年的样本外预测在预测峰值的几天内显示出准确性，尽管整个研究地区的年度环保时间可能相差长达30天。在落叶林和混交林类型中，表现非常高。我们的方法可推广到全球的温带森林，并为任何可获得每日温度(无论是直接测量的还是模拟的)的时间段的森林绿化提供了反向预报和预报的基础。
The yearly cycles in vegetation greenness are among the most important drivers of ecosystem processes. Predictive models for the timing of vegetation greenup and senescence are crucial for understanding how biological communities respond to global change. Greenup timing is closely tied to climate and also tracks yearly variability in temperature, and the strength of this relationship varies spatio-temporally. Local studies have been useful in understanding underlying mechanisms but they are insufficient in explaining larger scale variabilities. Large-scale studies using remotely-sensed data have the potential to harness regional dynamics, even if underlying mechanisms remain unknown, Yet predictive power using these approaches is low. Here, we predict vegetation phenology across Eastern North America via a novel class of Bayesian regression model. Our modeling framework provides continental-level peak greenup time predictions with high accuracy using satellite observations from the MODerate resolution Imaging Spectroradiometer (MODIS). In addition to taking into account temporal structure at individual sites, our models make use of information from the entire study extent regardless of their spatial proximity.Models were built from 2000 to 2016 and showed high prediction accuracy (R2> 95%). Out-of-sample predictions for the years 2017 and 2018 showed accuracy within days of the predicted peaks, even though yearly greenup timing can vary by up to 30 days across the study region. Performance was remarkably high across deciduous and mixed forest types. Our method is generalizable to temperate forests across the globe and provides a basis for backcasting and forecasting forest greenup for any time periods where daily temperatures, whether directly measured or modeled, are available.