Monitoring agroecosystem productivity and phenology at a national scale: A metric assessment framework
Monitoring agroecosystem productivity and phenology at a national scale: A metric assessment framework
复制标题
DOI：
10.1016/j.ecolind.2021.108147
复制
发表时间：
2021-08-27
影响因子：
6.9
通讯作者：
Taylor, Shawn D.
中科院分区：
文献类型：
--
作者：
Browning, Dawn M.;Russell, Eric S.;Taylor, Shawn D.
有效测量生产时间和数量的季节性变化，对于在不断变化的气候中管理空间异质的农业生态系统至关重要。尽管有许多用于这种测量的技术，但它们在大陆范围内的相互关系尚不清楚。使用从长期农业生态系统研究（LTAR）网络和其他网络收集的数据，我们研究了代表农田，牧场和美国大陆作物放牧综合系统中初级生产力，物候和碳通量的关键指标之间的相关性。我们研究的主要指标包括从涡度协方差（EC）塔估计的总初级生产力（GPP）和从Landsat卫星建模的GPP。Landsat NDVI和植被绿度（绿色色度坐标，GCC），来自2017年和2018年的塔式PhenoCams。总的来说，我们的分析比较了从三个独立的地面和远程平台估计的生产动态，使用了34个农业站点的数据，这些数据构成了51个站点-年的共定位时间序列。在所有四个指标上的成对传感器比较显示，季末（EOS）日期之间的相关性更强，均方根误差（RMSE）更低（Pearson R范围为0.6至0.7，RMSE范围为32.5至67.8）比季节开始（SOS）日期（0.46至0.69和40.4至66.2）。总体而言，SOS和EOS指标之间的中度至高度相关性相互补充，除了在一些生产力较低的牧场网站，估计SOS可能具有挑战性。从16天卫星GPP（179.1天）得出的生长季节长度估计值显著长于PhenoCam GCC（70.4天，padj < 0.0001）和EC GPP（79.6天，padj < 0.0001）。景观异质性不能解释SOS和EOS估计值的差异。在年产量超过1000 gC/m（-2）yr（-1）的地点，EC GPP和PhenoCam GCC的年度综合生产力估计值与Landsat GPP和NDVI的估计值不同。基于我们的研究结果，我们开发了一个“度量评估框架”，阐明了来自卫星，涡度协方差和PhenoCams的度量在何处以及如何相互补充，偏离或冗余。该框架旨在优化监测，建模和预测生态系统功能的仪器选择，最终目标是为土地管理者，政策制定者和行业领导者在多个尺度上的决策提供信息。
Effective measurement of seasonal variations in the timing and amount of production is critical to managing spatially heterogeneous agroecosystems in a changing climate. Although numerous technologies for such mea-surements are available, their relationships to one another at a continental extent are unknown. Using data collected from across the Long-Term Agroecosystem Research (LTAR) network and other networks, we investigated correlations among key metrics representing primary production, phenology, and carbon fluxes in croplands, grazing lands, and crop-grazing integrated systems across the continental U.S. Metrics we examined included gross primary productivity (GPP) estimated from eddy covariance (EC) towers and modelled from the Landsat satellite, Landsat NDVI, and vegetation greenness (Green Chromatic Coordinate, GCC) from tower-mounted PhenoCams for 2017 and 2018. Overall, our analysis compared production dynamics estimated from three independent ground and remote platforms using data for 34 agricultural sites constituting 51 site-years of co-located time series.Pairwise sensor comparisons across all four metrics revealed stronger correlation and lower root mean square error (RMSE) between end of season (EOS) dates (Pearson R ranged from 0.6 to 0.7 and RMSE from 32.5 to 67.8) than start of season (SOS) dates (0.46 to 0.69 and 40.4 to 66.2). Overall, moderate to high correlations between SOS and EOS metrics complemented one another except at some lower productivity grazing land sites where estimating SOS can be challenging. Growing season length estimates derived from 16-day satellite GPP (179.1 days) were significantly longer than those from PhenoCam GCC (70.4 days, padj < 0.0001) and EC GPP (79.6 days, padj < 0.0001). Landscape heterogeneity did not explain differences in SOS and EOS estimates. Annual integrated estimates of productivity from EC GPP and PhenoCam GCC diverged from those estimated by Landsat GPP and NDVI at sites where annual production exceeds 1000 gC/m(-2) yr(-1). Based on our results, we developed a "metric assessment framework" that articulates where and how metrics from satellite, eddy covariance and PhenoCams complement, diverge from, or are redundant with one another. The framework was designed to optimize instrumentation selection for monitoring, modeling, and forecasting ecosystem functioning with the ultimate goal of informing decision-making by land managers, policy-makers, and industry leaders working at multiple scales.