A Framework for Assessing Error Heteroscedasticity of Satellite Estimates and Extracting Spatiotemporal Variability from Precipitation Data
- Author(s): Liu, Hao
- Advisor(s): Sorooshian, Soroosh
- Gao, Xiaogang
- et al.
Precipitation is an important climate variable influencing human society. This dissertation focuses on developing a framework to assess the Remotely-sensed precipitation error's heteroscedasticity and precipitation's distribution patterns over space and time. The first part of the framework is to establish the joint-probability distribution functions (pdf) between satellite precipitation and ground observations with identical analytic format but adaptive parameters. The adaptability of the proposed model is verified by applying it to three locations (Oklahoma, Montana, and Florida), and by applying it to cold season, warm seasons and the entire year. Then the heteroscedasticities in the errors of satellite precipitations are investigated using my proposed model under those scenarios. The results show the joint-pdfs have the same formulation under these scenarios, whereas their parameter-sets were adaptively adjusted. This parametric model reveals detailed information about the spatial and seasonal variations of the satellite precipitation. I found that the shape of the conditional pdf shifts across the intensity ranges. At the 10~20 mm/d range, the conditional pdf is L-shaped, while in the 40~60 mm/d range, it becomes more bell-shaped. I also conclude that no single satellite precipitation product outperforms others with respect to the different scenarios (i.e., seasons, regions, climates).
The second part of the framework applies Empirical Orthogonal Function (EOF) and Nonlinear Mode Decomposition (NMD) to extract multi-scale spatial patterns and physically-meaningful periodic signals from a space-time array of precipitation measurements. A case study over the southwestern US shows this combined EOF-NMD technique is capable of identifying the teleconnections that the regional precipitation's seasonal cycles are amplitude-modulated (AM) by large-scale climate oscillations: the AM signal from the first PC is correlated (R=0.41) with Pacific Decadal Oscillation index during 1976-2001, the AM signal from the fourth PC is correlated (R=0.56) with Nino 3.4 index during 1985-2000.