A variational scheme for the analysis of hourly high-resolution precipitation has been developed over China and tested in extraordinary floods occurrence. The cost function of the proposed scheme simply required scalar and gradient of observations are computed for the input information contained in the new precipitation observations to properly improve the behavior of merged algorithm in short-time at their native points.
Experiment results show variational analysis has a positive impact of merged data over China and performs particularly well at Western and Southwest regions when compared to the sources of inputs.
Validation experiments in which each affected stations is withdrawal at once indicated that the constructed data can clearly seamlessly be beneficial to the analyses and subsequent forecast of heavy rainfall events. In particular, it also confirmed that the merged of precipitation observations has the ability to improve the quality of rainfall and capture the pattern agreements in relatively data-sparse regions.
In this work, variatonal experiments have been performed using manually quality controlled hourly rainfall accumulations that were obtained from National Meteorological Information Center, China Meteorological Administration (NIMC, CMA) and around 29,000 automatic weather hourly stations reports over continental China. One month period (July, 2009) has been selected, for this period was characterized by numerous precipitation events of both convective and stratiform nature. The corresponding precipitation map of the 10km integration has similar overall features to the observations Over the Island Continent there are notable precipitation maxima over the Northern. The gauge-based analysis of hourly rainfall is generated through interpolation based algorithm with consideration of orographic effects [Shen, 2010; Xie et al., 2007]. The temporal evolution of the precipitation field (onset, growth or decay) cannot be represented by extrapolation methods at coarser stations. Fictitious information occurred, especially in spare-gauge western China (Fig.1 top).
The satellite estimates to be merged in this study are CMORPH [Joyce et al., 2004], which take a propagating/morphing estimates of instantaneous precipitation from PMW observations along the cloud system advection vectors computed from consecutive geostationary IR cloud images in 30 min intervals. Even though the original satellite data are available on an 8-km resolution grid, they are projected on horizontal resolution of 10-km covering the whole of mainland China and matched with the time period definition of the gauge observations when being assimilated. (Fig. 1 middle)
Different form Xie et al., , in our approach the inputs errors relate the relative errors to the spatial variability of the precipitation field itself, without directly quantifying specific error sources. The relative error of the hourly precipitation rates is based on empirical models. Practically, error varies between 35% (1:3) over the flat terrain in intensive stations to 65% (2:3) over the Rocky Mountains in isolations. The selected formulation for error is rather arbitrary given the current uncertainties both in gauge observations and in moist physical parameterizations (involved in the observation operator). However, preliminary tests have shown that the results of variational experiments are not very sensitive to the coefficients. The occurrence of the movement of precipitation field when the high surface wind blows in a specific direction is not considered/disregarded in the summer situation study.
3. Experiment and results
Figure 1 (top)(middle) provides an example of the spatial coverage of two types of inputs over China for a single hourly of variational analysis. Fig 1 (top) shows the observed precipitation with peak values over North China and Shandong province with above 50 mm in hour, unrealistic depict rainfall distribution/information can be represented by extrapolation methods in Southwestern regions and Tibet plateau, while Figure 2 (middle) displays the estimated precipitation for CMORPH estimation and shows a much better performance over those data-sparse regions where hardly any precipitation was observed. However it can be seen that the satellite precipitation does not properly estimate the observation precipitation maxima even though satellite bias has been adjusted through/via gauge-based observation. We analysis the diurnal cycle of precipitation gives an impression of the temporal distribution (Fig. 2). Large Bias(Fig.2, top) score show the underestimations of precipitation area in the sparse regions. Interestingly, the precipitation RMSE (Fig2 middle) is dissimilar indicating only large changes in the spatial distribution. The precipitation correlation confirms this behavior. It is very low due to the underestimation of precipitation. The variational analysis shows the best performances. The figure illustrate that variational products have high quantitative quality with a negligible bias and a correction coefficient is above 0.85.
The validation is aim to illustrate that the variational analysis is appropriate for capture character of precipitation extremes. Figure 3 show the area-averaged diurnal cycle of precipitation for July 2009 over the four kinds of domains. While analysis nicely follows the gauge observations higher amounts, still clearly underestimates precipitation near. Variational analysis can present some precipitation structures near Yangtze River basins and South China region, but the satellite estimates systematically underestimated. Except analysis the two others largely missed the strong precipitation around the Yangtze River and South China region, Analysis is similar to the satellite estimations, apparently due to their sparse gauge data in this area, such as Tibet Plateau. In Figure 4, the bias and relative bias for the analysis is the smallest in the three estimates, especially with the rainfall intensity more than 5mm/hour, indicating the variational algorithm effective. Higher precipitation amount exhibit much lower relative errors. The heavy precipitation bands over the Yangtze and Songhua River Basins were depicted very well in analysis with the minimum bias between±1.2685 and -1.5289~1.148 mm/hr, respectively. The study indicates that high-resolution at short-time assimilation product is better to study for investigating precipitation climatologic variability and extreme events.