基于对象化方法的高分辨率降水集合预报释用方法研究

At present, most ensemble forecast interpretation methods are grid pointwise. These pointwise methods are showing weakness in high resolution ensemble forecast for precipitation. For example, ensemble mean forecast often over-estimates rainfall area, and under-estimates local maxima because of the smoothing in pointwise averaging. The pointwise posterior probability methods also show weakness since they cannot correctly evaluate the performance of ensemble forecast in training..Therefore, the current project proposes a new interpretation method by using object-based method in ensemble forecast interpretation to substitute the incompetent pointwise methods. The new method first coverts grid-based quantitative precipitation forecast to object-based forecast, and then verifies the probability distribution of different object features. Based on these probability distribution, the optimal deterministic forecast and its uncertainty information, and probabilistic forecast of precipitation are estimated.

目前，大多数集合预报释用方法都是在单个模式格点上对集合预报进行释用（简称格点方法）。这些方法在高分辨率降水预报中存在很多不足。如：集合平均预报因为格点平均的平滑作用，常常夸大降水落区，丢失降水峰值；格点检验方法常常产生错误的后验信息，导致基于后验信息的集合预报释用模型预报失败。.因此，为了解决高分辨率降水集合预报释用中格点方法不适用问题，本项目拟基于降水对象识别方法，研究一套新的集合预报释用方法。该方法分类分析集合预报中降水对象的各种特征的概率分布，并据此生成特征分离的集合平均预报和概率预报。

为了解决高分辨率降水集合预报释用中格点方法不适用问题，本项目发展了一套新的基于降水对象识别方法的集合预报释用方法。该方法设计了格点降水预报对象化、对象特征提取、对象间相似性度量、预报间对象最优匹配、基于对象的预报相似性度量、对象化聚类分析、对象融合集合平均等技术，实现了分类分析集合预报中降水对象的各种特征，如：位置、朝向、面积、强度等，实现了特征分离的集合预报释用，有效的改善了高分辨率降水集合预报中格点方法的各种不足。