冠层四流近似模型的发展历史

1. Kunbelka-Munk theory

This is the earlist model using a two-stream approximation
dIdz=−(k+s)I+sJdJdz=(k+s)J−sI \begin{aligned} &\frac{dI}{dz} = -(k+s)I+sJ\\ &\frac{dJ}{dz} = (k+s)J - sI \end{aligned} dzdI=(k+s)I+sJdzdJ=(k+s)JsI
Here, III and JJJ is downward and upward flux density, and kkk is obsorption coefficient, sss is back scattering coefficient, zzz is the metrical depth.
在这里插入图片描述

Another notation represent the K-M theory by
dE−dz=−aE−+σE+−dE+dz=−aE++σE− \begin{aligned} &\frac{dE^-}{dz} = -aE^-+\sigma E^+\\ &-\frac{dE^+}{dz} = -aE^+ + \sigma E^- \end{aligned} dzdE=aE+σE+dzdE+=aE++σE
Here, a=k+sa=k+sa=k+s is called attenuation coefficient, and σ\sigmaσ is backscattering coefficient.

2. Duntley equations

For considering the specular source like sun, we have Duntley equations.
dEsdz=−kEsdE−dz=s′Es−aE−+σE+dE+dz=−s′Es+aE+−σE− \begin{aligned} &\frac{dE_s}{dz} = -kE_s\\ &\frac{dE^-}{dz} = s'E_s -aE^-+\sigma E^+\\ &\frac{dE^+}{dz} = -s'E_s +aE^+ - \sigma E^- \end{aligned} dzdEs=kEsdzdE=sEsaE+σE+dzdE+=sEs+aE+σE
Here, kkk is extinction coefficient for specular flux density, and s′s's is forward scatter coefficient for specular flux density, and sss​ is backward scatter coefficient for specualar flux density.

To now, these equations are not connected with canopy parameters, such as leaf area index.

3. Suit and SAIL model

Suit model is also Duntley equations, but the coefficients are directly expressed in biophysical parameters of the canopy. The coefficients of suit model only defined for horizontal and vertical leaves, SAIL model improved the Suit and its coefficients can be computed for any leaf inclination.

These two models are actually four-stream model, which is
Es/dz=−kEs,E−/dz=s′Es−aE−+σE+,E+/dz=−sEs−σE−+aE+,πIo+/dz=−wEs−vE−−v′E++KπIo+,πIo−/dz=w′Es+v′E−+vE+−KπIo−. \begin{aligned} & E_s/dz = -kE_s,\\ & E^-/dz=s'E_s-aE^-+\sigma E^+,\\ & E^+/dz=-sE_s-\sigma E^-+aE^+,\\ & \pi I_o^+/dz=-wE_s-vE^--v'E^++K \pi I_o^+,\\ & \pi I_o^-/dz=w'E_s+v'E^-+vE^+-K \pi I_o^-.\\ \end{aligned} Es/dz=kEs,E/dz=sEsaE+σE+,E+/dz=sEsσE+aE+,πIo+/dz=wEsvEvE++KπIo+,πIo/dz=wEs+vE+vE+KπIo.
The parameter are easy to understand and are same to the previous blog.

### 流回归模型的概念 流数据处理涉及连续不断的数据输入,这些数据通常具有时间敏感性和不可重复读取的特点。对于流数据上的机器学习任务而言,算法需要能够在线更新模型参数而无需重新训练整个历史数据集。流回归模型即是在这种背景下发展起来的一种技术。 ### 实现方法概述 为了适应快速变化的数据分布并保持较低延迟,在线梯度下降(Online Gradient Descent, OGD)[^1] 是一种常用的优化策略。OGD允许每次接收到新样本时只调整当前权重向量的一小部分,从而使得模型可以逐步逼近最优解而不必等待全部数据到来后再做一次性大规模计算。 另外,滑动窗口机制也被广泛应用于构建高效的流回归器中。通过维护固定大小的历史观测记录作为局部视图来近似全局统计特性,可以在一定程度上缓解概念漂移带来的影响[^2]。 ### 示例代码展示 下面给出一段简单的基于Python的流回归模型实现: ```python import numpy as np from sklearn.linear_model import SGDRegressor class StreamingRegressionModel: def __init__(self): self.model = SGDRegressor(max_iter=1000, tol=1e-3) def update(self, X_new, y_new): """Update the model with new data points.""" self.model.partial_fit(X_new.reshape(-1, 1), [y_new]) def predict(self, X_test): """Predict using the updated model.""" return self.model.predict(X_test.reshape(-1, 1)) # Example usage if __name__ == "__main__": stream_reg = StreamingRegressionModel() # Simulate streaming data arrival for i in range(10): # Assume we have 10 incoming samples over time x_i = np.random.rand() * 10 # Random input value between 0 and 10 y_i = 2*x_i + 1 + np.random.randn()*0.5 # Linear relationship plus noise print(f"Updating model with ({x_i:.2f}, {y_i:.2f})...") stream_reg.update(x_i, y_i) test_value = 7.5 prediction = stream_reg.predict(np.array([test_value])) print(f"\nPrediction at x={test_value}: {prediction[0]:.2f}") ``` 此段程序展示了如何利用`sklearn`库中的随机梯度下降回归器(SGD Regressor)来进行增量式的参数估计过程。每当有新的观察值到达时调用`update()`函数即可完成即时的学习;而在预测阶段则只需提供待测特征向量给到`predict()`接口便能获取相应的输出结果。
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值