Datasets
Standard Dataset
Observational Precipitation Data
- Citation Author(s):
-
Chin-ShyurngFahn
National Taiwan University of Science and Technology - Submitted by:
- Chin-Shyurng Fahn
- Last updated:
- Mon, 07/08/2024 - 15:58
- DOI:
- 10.21227/nnth-sf28
- Data Format:
- The format for each observational precipitation data contains 4 columns. The first column represents which data of the 44, 550 observations, and the serial number is from 1 to 44, 550. The second column represents longitude, and the range is from 120.0000°E to 122.0250°E. The third column represents latitude, and the range is from 21.8875°N to 25.3250°N. The fourth column represents the precipitation value with a precision of two decimal places. All observational precipitation values are ranged from 0 to 4921.5 mm/day.
- License:
5 Views- Categories:
- Keywords:
Abstract
The precipitation data can be stored in a two-dimensional array of 275 ×
162, where the x-axis stands for the longitude ranged from 120.0000°E to 122.0250°E and the y-axis stands for the latitude ranged from 21.8875°N to 25.3250°N. That is, the data of each storage cell represents the observational precipitation accumulated in the area of 0.0125°E width by 0.0125°N length. From the original observational precipitation data in the two-dimensional array, we can create the precipitation data with a lower resolution; for example, 0.0625° (the coverage of an observation about 6.5 km in one dimension) and 0.25° (the coverage of an observation about 26 km in one dimension). In these lower resolutions, we can take an average of the original observational precipitation data in the storage cells as the data stored in the corresponding larger cells. This conversion is designed to acquire the input precipitation data for generating the output data with a desired higher resolution by use of a deep learning model. It is guaranteed that the input-output pair of the precipitation data has the same rainfall values.
The following describes how to create the input precipitation data used for generating the output data with a resolution of 0.0625°. We first crop the original observational precipitation data of 275 × 162 to a precipitation data of 275 × 160. Then we set a square area of size 5 × 5 and the rainfall values in the area are averaged and stored to the corresponding cell of a two-dimensional array of 55 × 32 that constitutes the input precipitation data to generate the output data with a resolution of 0.0625° through a deep learning model. The relationship of the precipitation data with the resolutions between 0.25° and 0.0625° is graphically shown in Fig. 2(a).
Next, in the similar way as described above, we present how to create the input precipitation data used for generating the output data with a resolution of 0.0125°. First, the original observational precipitation data of 275 × 162 are cropped into a precipitation data of 260 ×
160. Subsequently, a square area of size 20 ×
20 is set, whose contains rainfall values are averaged and then stored to the corresponding cell of a two-dimensional array of 13 ×
8 that acts as the input precipitation data to generate the output data with a resolution of 0.0125°. Such data are guaranteed to be generated by a deep learning model. Figure 2(b) illustrates the relationship of the precipitation data with the resolutions between 0.25° and 0.0125°.
