Page 219 - 《高原气象》2022年第6期

高 原 气 象 41 卷
1582
张强，姚玉璧，李耀辉，等，2020. 中国干旱事件成因和变化规律的 张远东，张笑鹤，刘世荣，2011. 西南地区不同植被类型归一化植
研究进展与展望［J］. 气象学报，78（3）：22. DOI：10. 11676/ 被指数与气候因子的相关分析［J］. 应用生态学报，22（2）：
qxxb2020. 032. 323-330. DOI：10. 13287/j. 1001-9332. 2011. 0053.
张强，张良，崔显成，等，2011. 干旱监测与评价技术的发展及其科 赵倩倩，张京朋，赵天保，等，2021. 2000年以来中国区域植被变化
学挑战［J］. 地球科学进展，26（7）：763-778. DOI：10. 11867/j. 及其对气候变化的响应［J］. 高原气象，40（2）：292-301. DOI：
issn. 1001-8166. 2011. 07. 0763. 10. 7522/j. issn. 1000-0534. 2020. 00025.

Remote Sensing Drought Monitoring and Assessment in Southwestern

China based on Machine Learning

1
JIA Hejia ，LI Xiehui ，WANG Lei ，XUE Yuting ，LIN Huiquan 2
1
1
1
（1. Plateau Atmosphere and Environment Key Laboratory of Sichuan Province，School of Atmospheric Sciences，Chengdu
University of Information Technology，Chengdu 610225，Sichuan，China；
2. Daliushu state-owned sand-control forest farm in Naiman county，Tongliao city 028300，Inner Mongolia，China）
Abstract：Due to the complexity of drought and the diversity of influencing factors，the accurate monitoring of
drought still faces many problems，especially the increasing frequency and aggravation of drought in Southwest‐
ern China，and the formation and disaster causing process have certain particularity. However，the traditional
drought monitoring methods cannot meet the requirements of regional drought monitoring，so more scientific
monitoring methods and means are needed. Since machine learning can comprehensively consider a variety of di‐
saster causing factors to establish a comprehensive drought monitoring model，it undoubtedly provides a new
technical means for drought monitoring. Therefore，this paper used multi-source remote sensing data from 2010-
2019 and meteorological station data from 1980-2019 to first construct a random forest monitoring model to re‐
construct and supplement the surface temperature in Southwestern China，and then constructed XGBoost moni‐
toring model to monitor，evaluate and validate the drought in Southwestern China. The results showed that：
（1）The correlation coefficients between the training set and test set of the random forest model and the actual
surface temperature of the stations exceeded 0. 9，which reached a significant correlation. The spatial distribution
of LST reconstruction values was similar to that of remote sensing monitoring values，and the values were close
to the observed values of meteorological stations.（2）The correlation coefficients between the monitoring values
of XGBoost model training set and test set and the SPEI calculated values at the stations were more than 0. 86，
with significant correlation. The overall consistency rate of drought grade between the monitored value and the
calculated SPEI values exceeded 85%.（3）The overall consistent rate between the monitored values of the XG‐
Boost model and the MCI values was above 67. 88%，which was more consistent. The consistent rate of all
months exceeded 58%，with the highest consistent rate of 75. 07% in September and the lowest consistent rate of
58. 26% in February.（4）The drought in each season monitored by the model was basically consistent with the
actual drought，which could better reflect the spatial distribution and drought in Southwestern China.
Key words：Multi-source remote sensing；Random Forest；XGBoost；drought monitoring；Southwestern China