Calibration And Spatial Downscaling Of TRMM 3B43 Precipitation Product

Precipitation, one of the crucial meteorological elements of climate change, is a major driving force in global climate change, hydrological cycle, and ecological environment, which is of significant importance to agricultural irrigation, disaster prevention, and other production activities. Remote sensing has the advantages of completely scanning the entire region under study, shortly revisiting to the same region, and conveniently accessing to the data, which is the only choice for obtaining precipitation at a regional or global scale. Tropical Rainfall Measuring Mission (TRMM) precipitation data with a relatively high spatial resolution and a relatively large scope of space had been widely used in recent years. Precipitation with high accuracy and high spatio-temporal resolution is intensely essential for hydrological, meteorological, and ecological research in local basins and regions. Therefore, the research on calibration and downscaling of TRMM precipitation has important practical significance.In order to solve the problems of data errors and coarse resolution of the TRMM product, and the limitations of the auxiliary variables and the global regression model, in this paper, the Gansu province, which is a typical arid and semi-arid area in Northwest China, was selected as a case study area in the paper. The measured rain gauges data were used to verify and correct TRMM 3B43 monthly precipitation using the Cumulative Distribution Function (CDF) model. A multivariable geographically weighted regression (GWR) was developed to obtain 1 km annual precipitation. Moreover, the monthly downscale models were also constructed to obtain satellite precipitation data with 1 km resolution. The main research contents and conclusions are as follows:(1) Validity checking and calibration of the TRMM 3B43 precipitation data. The TRMM data at monthly, seasonal and annual scales were validated, and the errors at each rain gauges station were analyzed. The results show that the TRMM precipitation data are highly correlated with the rain gauges data, and the correlation degree becomes higher with the accumulation of time. In addition, the TRMM precipitation significantly overestimates the actual precipitation, and the errors at each rain gauges station are highly correlated with annual precipitation (R2=0.81). The bias is reduced obviously after calibration using the CDF model, indicating the proposed method is of good generalization ability and is effective in the study area.(2) Downscaling of TRMM annual precipitation. A multivariable GWR downscaling method was developed to obtain 1 km annual precipitation. The GWR method was compared with two other downscaling methods (univariate regression, UR; multivariate regression, MR) in terms of the performance of downscaled annual precipitation. Variables selection procedures were proposed for selecting appropriate auxiliary factors in all three downscaling methods. Validation results clearly demonstrated that the GWR method based on local regression, which could avoid over-fitting and scale mismatch errors, performed significantly and consistently better than both UR and MR methods for all three typical years.(3) Extension of TRMM annual precipitation downscaling method to monthly scale. Two monthly downscaling strategies (annual-based fraction disaggregation method and monthly based GWR method) were evaluated. The former method faces the challenge of precipitation spatial heterogeneity and the derived monthly precipitation heavily depends on the annual downscaled results, which could lead to the accumulation of errors. The monthly based GWR method is suitable for downscaling monthly precipitation, but the accuracy of original TRMM 3B43 data would have large influence on downscaling results.The main contribution of this study is that a CDF calibration procedure was proposed, which could well fit the distribution of TRMM monthly precipitation. In addition, a multivariable GWR annual downscaling method and two monthly downscaling strategies were developed to obtain 1 km precipitation. It was demonstrated that the proposed methods were effective for obtaining both annual and monthly TRMM 1 km precipitation with high accuracy.