基于神经网络的太赫兹冰云探测反演算法研究

太赫兹频段在冰云探测上具有独特优势,但是目前的太赫兹冰云反演算法将不同种类冰相粒子（主要是冰和霰）视为冰粒子统一计算。本文根据冰云太赫兹辐射特性实现了一种预分类的神经网络算法,能够从太赫兹亮温中分别反演得到冰、霰两种粒子的统计参数和廓线分布。首先,基于WRF数值模式和ATMS载荷真实观测的冰云霰数据构建了包含冰、霰粒子密度廓线的混合冰云数据库,然后,使用DOTLRT辐射传输模式模拟183 GHz、243 GHz、325 GHz、448 GHz、664 GHz和874 GHz这6个频段的星载太赫兹冰云探测亮温,最后,开展冰云参数探测仿真试验,验证反演算法性能。仿真试验中反演得到的冰和霰的路径总量均方根误差分别为8.97 g/m²和10.90 g/m²,等效粒径均方根误差分别为7.54 μm和25.38 μm,反演的冰、霰密度廓线也具有较高的精度。研究结果表明本文算法能够以较好的精度从多频太赫兹冰云探测亮温数据中分别反演得到冰、霰两种粒子的路径总量、等效粒径、等效云高和密度廓线,突破现有研究仅仅计算单一冰粒子的局限,更加符合冰云真实情况。     

梯度应力作用下模型试件的岩爆破坏细观分析

岩爆是深部矿山开采亟待解决的安全问题,而应力梯度是影响岩爆的重要因素。为探究不同应力梯度的岩爆对围岩内部细观结构的影响,利用可实现三向六面加载及顶部梯度加载的气液复合型岩爆试验装置,对类岩体进行不同应力梯度条件下的岩爆物理模型加卸载试验,并借助扫描电镜对试件破坏面进行细观形貌特征分析。研究结果表明：不同应力梯度环境下试件岩爆的破坏现象与特征存在明显差异;试件所受的应力梯度越大,岩爆后的细观图像显示的晶体间孔隙越小,晶体的密实程度越高;不同应力梯度下试件产生剪切和劈裂破坏比例不同,应力梯度越大,剪切破坏比例越大;岩爆碎屑晶体轮廓具有分形特征,且随着试件所受的应力梯度的增加,其分形维数相对越大。     

基于RegCM3和DLBRM的耦合系统CRCHMS的构建及其在黑河上游的应用

以高寒山区—黑河流域上游为研究区，确定区域气候模式RegCM3的模拟方案，率定分布式水文模型（DLBRM），并开发了RegCM3和DLBRM模型接口，从而构建了区域气候水文耦合模拟系统CRCHMS。结果表明，以RegCM3作为气象驱动数据的CRCHMS系统模拟性能优于以观测站点作为气象驱动数据的DLBRM模型，对莺落峡径流量的模拟值与实测值的相关系数在校准期和验证期分别为0.47和0.62，均方根误差分别为0.045和0.044 cm/d，相对误差分别为-0.4%和6%，纳什系数在率定期和验证期分别为0.22和0.36。     

FSS-based Evaluation on Monsoon Precipitation Forecasts in South China from Regional Models with Different Resolution

This study evaluated the forecast skill of CMA-GD 3 km and CMA-GD 1 km with hourly Rapid Update Cycle (RUC) for five monsoon precipitation events in South China from 2018 to 2020, using the fraction skill score (FSS) of the neighborhood spatial verification method. The results revealed that, among the 24-lead-hour forecasts in CMA-GD 3 km, the FSS for the 0.1 mm precipitation threshold increased linearly with the lead time from 3 to 1 hour, while there was no significant improvement in other lead times. For the 5 mm precipitation threshold, the forecast skill was highest for the latest 1-hour lead time, while the FSS showed slight variation between lead times of 24 hours and 8 hours. The FSS for 10 mm and 20 mm precipitation thresholds were similar to that of 5 mm, with the difference that the best score occurred at the 2-hour lead time. Among the 6-lead-hour forecasts in CMA-GD 1 km, the forecasts of the latest 1-hour lead time were the best choices for four precipitation thresholds. When comparing CMA-GD 3 km and CMA-GD 1 km, it was found that CMA-GD 3 km had better skill for forecasts of 0.1 mm and 5 mm precipitation at 2-hour and 1-hour lead times, while CMA-GD 1 km had better skill for all other forecasts, including the forecast of 20 mm precipitation nearly all lead hours (including 3- to 6-hour, and 1-hour lead times). The results suggest that the increased resolution of the model may be beneficial for precipitation forecasts in South China, especially for short-duration heavy precipitation over a longer lead hours. However, the limited sample size of this study calls for further evaluation using more cases to validate the results′ generality.