基于长序列遥感降水融合数据集的黄河源气象干旱特征研究

Meteorological drought characteristics in the source region of the Yellow River based on long-term fused remote sensing precipitation datasets

摘要:

遥感降水产品可为气象干旱精准监测提供近实时、多时空分辨率的降水数据,但其性能易受复杂地形和极端气候的影响,基于机器学习的多源数据融合方法为提高复杂地形和偏远地区遥感降水产品精度和干旱监测能力提供了新思路。基于此,本研究以资料稀缺的黄河源为研究区,利用随机森林(random forest,简称RF)方法生成了黄河源区长序列(1983—2018年)高精度的格网降水数据集,并基于该数据集利用标准化降水指数(standardized precipitation index,简称SPI)和游程理论方法识别气象干旱事件,阐明融合降水产品数据集对气象干旱事件特征的捕获能力。结果表明:1)基于机器学习的降水融合数据集在站点尺度性能方面优于3套原始遥感降水产品,能较理想地捕获降水的年内和年际波动;2)黄河源降水和4个时间尺度的SPI (SPI1、SPI3、SPI6和SPI12)均呈显著增加趋势(通过0.05信度的显著性检验),表明近36年黄河源区降水增加,气象干旱趋缓;3)降水突变点发生在2006年,2006年前的平均干旱事件历时长、烈度强和极值大,但2006年后干旱事件特征趋缓;空间分布上,源区西北部干旱历时和烈度较高,东南部干旱强度和极值较高。

Abstract:

Remote sensing precipitation products provide near real-time,multi-temporal,and spatially resolved precipitation data,which are essential for accurate meteorological drought monitoring.However,their accuracy is often compromised by complex terrain and extreme climate conditions.Machine learning-based data fusion methods offer a novel solution to for enhancing the precision of remote sensing precipitation products,particularly in challenging environments.This study focuses on the source region of the Yellow River,a data-scarce area with complex topography,to develop a high-resolution gridded precipitation dataset and evaluate its utility in drought monitoring.
Using the Random Forest (RF) model,a long-term (1983—2018) high-accuracy precipitation dataset was generated by fusing multiple remote sensing precipitation products.The fused dataset was applied to identify meteorological drought events using the Standardized Precipitation Index (SPI) and run theory.Temporal and spatial characteristics of drought events were analyzed to assess dataset’s capability to capture drought dynamics.Key findings include:1) The fused precipitation dataset outperformed three individual remote sensing precipitation products (PERSIANN-CDR,MSWEP v2.0,and CHIRPS v2.0) at the station scale,exhibiting higher correlation coefficients (CC),lower root mean square errors (RMSE),reduced relative bias,and improved Kling-Gupta efficiency (KGE).The dataset accurately captured both monthly and inter annual variations,demonstrating its adaptability to the Yellow River source region.2) Precipitation and SPI values across four temporal scales (SPI1,SPI3,SPI6,and SPI12) exhibited statistically significant increasing trends (P<0.05),indicating increased precipitation and a reduction meteorological drought severity over the past 36 years.3) An abrupt change in precipitation occurred in 2006.Prior to this point,the region experienced more frequent and severe droughts with longer durations,higher intensities,and greater extremes.After 2006,drought characteristics became milder.Spatially,the northwest of the source region experienced longer and more severe droughts,while the southeast exhibited higher drought intensity and extremes.
This study provides critical insights into precipitation and drought dynamics in the source region of the Yellow River,supporting efforts in meteorological drought early warning,water resource management,and regional climate adaptation.The observed increasing precipitation trend and alleviation of drought conditions are vital for developing sustainable development strategies and disaster mitigation plans.
The research underscores the potential of integrating remote sensing products with machine learning techniques to improve the accuracy and applicability of climate datasets,especially in regions with limited ground-based observations and complex topography.The fused dataset not only demonstrated enhanced accuracy but also provided a robust foundation for analyzing the spatiotemporal evolution of meteorological drought events.Future work could extend this approach to other regions and incorporate additional hydrometeorological variables for more comprehensive drought assessments.