题目：The greenscape shapes surfing of resource waves in a large migratory herbivore
作者：Ellen O. Aikens, 1,3 *Matthew J. Kauffman, 2,3 Jerod A. Merkle, 1 Samantha P. H. Dwinnell, 1 Gary L. Fralick 4 and Kevin L. Monteith 1,3,5
目标
本文提出了一个greenscape的概念，也就是绿波green wave在景观中的进程的模式（The way that the green wave progresses across the landscape）生态学就喜欢zhuai名词，毕竟学科没啥内容举个例子例说明，植被快速的返青（green-up）会通过减少高质量草料的可获得时间影响物种量，在部分有蹄类物种中已经证明会减少年轻个体对种群的补充。这种后果会成为物种最大化surf on the green wave的约束。作者提出来Greenscape Hypothesis，也就是迁徙路途中green-up的速度，时长和顺序会影响动物迁徙时冲绿浪的能力。还预测拥有较缓的green-up速度，持续时间长，依次绿化的路径会增强surfing。
方法与结果
作者用了3年（13-15）99只长耳鹿（mule deer）的gps追踪数据，每1-5小时回传数据，Net Squared Displacement (NSD)方法判断迁徙开始和停止时间。
利用Visvalingam line simplification algorithm 的 程序来简化路径（因为每个个体采样点的个数不统一）。
NDVI采样mod09q1地表反照率计算（8d，250m），处理方法：
- 有云值和负值为空,
- 冬天的值置零,
- 中值滤波
- 归一化时间序列
最后对时间序列进行double-logistic拟合，IRG(顺时返青率,Instantaneous Rate of Green-up)也就是拟合曲线的一阶导数。采用95%的点构成最小多边形作为研究区域，并选取IRG达到最大的日期在整个研究区域分布的0.02和0.98分位数作为春季开始和停止的时间。
每年对长耳鹿进行重捕，测量年龄（通过牙骨质轮），测量体脂，测量怀孕个数（超声）。
为了评价不同个体surf greenwave的情况，分别计算迁徙途中GPS点的IRG和距离IRG峰值的天数的绝对值（Days-From-Peak，DFP），求和后除以前文定义的春季持续时间来消除sample bias（因为采样频率1-5小时不等）。
不同的迁徙路径根据在春季持续时间的长短被分为short(< 0.33 quantile)，mid (0.33–0.66quantile)和long(> 0.66 quantile)
On average, deer received IRG values above 0.8 for fewer days in 2013 (28.87 ± 1.13, mean ± SE) than in 2014 (43.96 ± 1.86) and 2015 (57.35 ± 2.70). In 2013, long‐duration migrants had IRG values above 0.8 for a shorter amount of time than short‐ or mid‐duration migrants (short = 31.63 ± 1.64, mid = 31.43 ± 1.64 and long = 21.80 ± 1.85 days; mean ± SE), whereas the opposite was true in 2014 (short = 39.65 ± 2.57, mid = 41.38 ± 3.40 and long = 49.43 ± 3.18 days; mean ± SE) and 2015 (short = 46.33 ± 5.37, mid = 60.77 ± 4.41 and long = 64.06 ± 3.14 days; mean ± SE)
作者还说明了为什么用这两个指标来度量，就是说不同的DFP和IRG对应了不同IRG曲线的形状。
为了刻画所谓的green scape：作者用了：
- green-up rate：先计算了沿迁徙路线上1km的IRG，然后取倒数，再平均，取倒数的原因是避免偏态分布（skewed distributions），这个倒数被作者叫做spring-scale，被解释为到达green-up峰值的时间。
- green-up order：对各个1km的间隔点的动物到达时间和IRG峰值时间分别排序，求秩相关的系数
- duration of green-up：路径开始和结束500m buffer的IRG峰值日期的差值
为了验证绿波假说，对每个利用的patch的占用日期和当地的IRG峰值日期做线性回归，在y=x的95%置信区间的点为perfect surfing。
Across the entire study area, the period of spring green‐up ranged from 14 April to 20 June in 2013 (a drought year), from 2 April to 30 June in 2014 (heavy snowpack year) and from 14 February to 14 June in 2015 (a low snowpack year with a wet spring). The vast majority (98.4%) of individual regressions between the date of peak IRG and the date of deer use were classified as surfing that was better than random, and 31.7% of those regressions were consistent with theoretically perfect surfing during the migration or spring period.
作者还模拟了随机行走的路线，模拟方法：
- 在研究区域内随机生成路线，方式是在研究区域内随机生成点，然后随机连接其中一对点；
- 用观测数据计算每日的rate of displacement，也就是每日行进的路程，用观测路线上相邻gps点距离除以时间求平均
- 用这个rate of displacement来模拟在随机产生的路线上的迁徙
Mule deer generally matched their movements with peaks in green‐up, with surfing that was nearly twice that of simulated migrations based on the daily rate of displacement from winter to summer range along the same elevational gradient .
检验不同个体间影响green-wave surfing的因素，对IRG和DFP作为相应变量，年龄、Greenscape（上文的三个变量）、以及营养状况和胎儿个数。模型不用说一定是glmm，random factor取为年份，修正偏态分布分别对DFG取了log、对green-up order取了arcsine，计算每个模型的AIC。首先对所有变量组合建模（无自相关），参数估计时对表现好的模型（比最好模型AIC差2以内）的用随机斜率重参数化（感觉像是贝叶斯的一套），估计的系数的95%的置信区间不包括0说明该变量显著。
Days‐From‐Peak decreased as spring scale decreased, green‐up order became more consecutive and green‐up duration increased
The Greenscape Hypothesis best described individual variability in green‐wave surfing. Green‐up duration was the strongest predictor of IRG across individuals (ΔAICc of second best supported model > 6; AICc Weight = 0.92). As the duration of green‐up increased, IRG also increased (β = 0.0034; 95% CI = 0.0022–0.0045)
Days‐From‐Peak was best explained by three models, containing a combination of green‐up duration, spring scale (reciprocal of the green‐up rate) and green‐up order (cumulative AICc weight > 0.5).
最后用的到的只包含greenscape变量的模型预测IRG，用置换检验（permutation test）来判断不同个体和不同年份之间是否IRG有差异，p值就是重采样平均出现比观测平均更极端的次数的占比。
（32条路线，3年的话，96个样本，sample 32个求平均 n次做分布为左图，sample 3个求平均n次做分布为右图个人猜测）
The greenscape was most difficult to surf in 2013 (P = 0.0013) and easiest to surf in 2014 (P < 0.0001), according to predicted IRG. Despite annual fluctuations in the greenscape, 4 of 32 routes (12.5%) were consistently of higher quality and were predicted to result in deer receiving 9% higher average IRG than other routes. Two routes were consistently of lower quality and were predicted to result in deer receiving 7.5% less average IRG.
最后把预测的不同路线预测的IRG做了个图。
讨论
## undo
## too long
## 题目是机器翻译的
