今日分享马普所一篇nc期刊
Widespread increasing vegetation sensitivity to soil moisture
https://doi.org/10.1038/s41467-022-31667-9
全球植被的空间分布对全球土壤水的敏感性
explainable machine learning
observation-based leaf area index
hydro-climate anomaly data
在半干旱和干旱地区,LAI对土壤水的敏感性快速增加
LAI的敏感性趋势与多变量的水文气候及气象数据有关
可解释神经网络
Molnar, C. Interpretable Machine Learning: A Guide for Making Black Box Models Explainable (2021). github 免费看.
应该使用的random forest方法,然后可解释性主要体验在用了the Shapley Additive Explanations (SHAP) method
我们的结果是通过可解释机器学习得出的,它可以从本质上隔离土壤湿度对LAI的影响与其他相关驱动因素的影响,从而超越纯粹的基于相关性的分析。
敏感性分析
the Shapley Additive Explanations (SHAP) method
import os
import numpy as np
import sys
# Drawing plots is not supported in sensi.yml environmental
# import matplotlib
# import matplotlib.pyplot as plt
# import statsmodels
# import cartopy.crs as ccrs
import statsmodels.api as sm
from sklearn.ensemble import RandomForestRegressor
import shap
from pathos.multiprocessing import ProcessingPool as Pool
from sklearn.linear_model import LinearRegression, TheilSenRegressor
import regressors
import regressors.stats as regressors_stats
import time
import warnings
warnings.filterwarnings('ignore')
# limit number of threads
os.environ["OMP_NUM_THREADS"] = "1"
os.environ["MKL_NUM_THREAD"] = "1"
os.environ["NUMEXPR_NUM_THREADS"] = "1"
# plt.rcParams.update({'font.size': 6})
overall sensitivity 全局敏感
the overall and the 3-year block sensitivities
out-of-bag
five long-term satellite-derived LAI datasets and ERA5-Land soil moisture reanalysis24, as well as modeled data from offline simulations from 9 TRENDY LSMs.
归因分析
Attribution Analysis
数据集
extrapolation of in-situ measurements (SoMo.ml)
O, S. & Orth, R. Global soil moisture data derived through machine learning trained with in-situ measurements. Sci. Data 8, 170 (2021)