*Appendix A.2. Three Climate Prediction Center Morphing Technique Datasets (CMORPH RAW, CMORPH CRT, and CMORPH BLD)*

CMORPH records precipitation from 1998 to the present with a 0.07◦ spatial and 0.5-hourly temporal resolution [66]. CMORPH blends passive microwave-based rainfall estimates from low-orbit satellite sensors (TMI, SSM/I, AMSR-E, AMSU-B) using spatial propagation information obtained from the IR data of geostationary satellites. In this study, three different CMORPH datasets are compared: the raw satellite data (CMORPH RAW), CMORPH CRT, and CMORPH BLD. The CRT dataset is produced by blending the RAW dataset with the CPC gauge dataset and the GPCP over lands and ocean surfaces via the probability density function matching the bias correction method. An optimal interpolation method is used to combine the CRT dataset with gauge analysis to produce the BLD dataset [69].

## *Appendix A.3. Precipitation Estimation from Remotely Sensed Information Using Artificial Neural Networks Dataset (PERSIANN) and PERSIANN-Climate Data Record Dataset (PERSIANN CDR)*

PERSIANN and PERSIANN CDR record precipitation from 2000 to the present with a spatial coverage that covers 50◦S to 50◦N latitude. PERSIANN utilizes infrared brightness temperature images from geostationary satellites to compute the rainfall rate estimation by the neural network function classification method [67]. Low-orbit satellite sensors (e.g., TMI, AMSR-E, AMSU-B, and SSM/I) provide passive microwave observations to update the ANN parameters. This method could narrow down the uncertainties of the relationship between cloud-top brightness temperature and precipitation estimated according to cloud properties and atmospheric conditions. Whenever independent estimates of precipitation are available, an adaptive training feature facilitates updating the network parameters [67]. Differing from PERSIANN, which is attainable in near-real time and only based on satellite measurements, PERSIANN CDR, initially designed for long-term climatic event research, is a pre-trained NCEP hourly precipitation dataset that merges the GPCP gauged dataset with its rainfall estimates [70].

## *Appendix A.4. Climate Hazards Group Infrared Precipitation with Station Dataset (CHIRPS)*

With its period from 1981 to the present, CHIRPS is particularly designed for drought monitoring studies [102]. The latest version 2.0 dataset, issued in February 2015 is used in this study. CHIRPS is mainly produced by four steps. (1) TRMM 3B42 is used to calibrate the Cold Cloud Duration (CCD) thermal infrared data to get the 5-day CCD, which is further transformed to the fractions of the long-term mean rainfall estimation. (2) The fractions are multiplied with the monthly precipitation climatology created using rain gauge stations from the Food and Agriculture Organization (FAO) and the Global Historical Climatology Network-Monthly (GHCN) datasets in order to remove the systematic bias and produce the CHIRP dataset. (3) A modified IDW method is used to combine the CHIRP dataset with the rain gauge datasets from many archives, including the Global Telecommunication System (GTS), to obtain the CHIRPS. (4) Since the above three steps are carried out at the 5-day timescale, daily CCD data and the NOAA Climate Forecast System (CFS) atmospheric model rainfall field data are used to disaggregate the CHIRPS to daily rainfall estimates by a simple redistribution method [72].
