Detection of Atmospheric Rivers in the Northern Hemisphere based on ERA5 reanalysis data and the IPART algorithm, 1979-2020

# 1. Overview

This is a catalogue of atmospheric river (AR) detections over the Northern Hemisphere, based on 6-hourly ERA5 reanalysis dataset and the Image-Processing based Atmospheric River Tracking (IPART) algorithm.

Time domain of the data:

From 1979-Jan-01 to 2020-Dec-31

Temporal resolution is 6-hourly

Spatial domain of the data:

Northern Hemisphere, land and ocean

Spatial resolution is 0.25 * 0.25 degrees latitude/longitude

Input data from ERA5 include:

Vertical integral of northward water vapour flux, in kg/(m s).

Vertical integral of eastward water vapour flux, in kg/(m s).

Data in the Northern Hemisphere domain (0 - 90 N), at 0.25 * 0.25 degrees latitude/longitude resolution are obtained from https://www.ecmwf.int/en/forecasts/datasets/reanalysis-datasets/era5.

Version v3.0.8 of the IPART Python module used for detection and tracking of atmospheric rivers is preserved at 10.5281/zenodo.4164826, available via Creative Commons Attribution 4.0 International license and developed openly at the Github repository https://github.com/ihesp/IPART.

# 2. File naming convention

The data files are named using the following convention:

ar_YYYYMM.nc

where:

YYYY: 4-digit year number

MM: 2-digit month number

E.g. `ar_199902.nc` means detections in Feb of 1999.

Months are calendar months, including Feb-29th in leap-years.

# 3. Data format

Data are saved in netCDF format.

Each data file contains one 3-dimensional array, of a shape `(t, 360, 1440)`, where:

`t`: length of the time dimension. Since data are 6-hourly, t equals 4 * num_of_days_in_month.

`360`: latitude dimension, from 0 - 90N, with a 0.25-degree step.

`1440`: longitude dimension, from 80 - 440 E (shifted eastward by 80 degrees to put both the Pacific and Atlantic oceans within the domain), with a 0.25-degree step.

Each time slice of the data contains maps of the Northern Hemisphere, with integer values in grid cells. Possible values are:

0: meaning no AR is detected in the grid cell.

1, 2, ... ,n: integer labels, each corresponding to the region of an AR entity.

# 4. Important parameters in the IPART algorithm

Here are the most important parameters used when detecting ARs from ERA5 data using the IPART python module:

    THR filtering kernel: `[16, 13, 13]`. `16` means 16 time slices, or equivalently 4 days given 6-hourly input data. `13` means 13 grid cells, or equivalently ~325 km, given 0.25 degrees latitude/longitude input data. Note that both of these temporal and spacial lengths are half of the sizes of the filtering kernel.

    minimum area: `50 * 1e4`, in km^2, minimum size of AR region candidates.

    maximum area: `1800 * 1e4`, in km^2, maximum size of AR region candidates.

    minimum L/W: `2.0`, minimum length/width ratio of AR region candiates.

    minimum length: `2000`, in km, minimum length of AR region candidates.

    minimum latitude: `20`, minimum latitude of the geometrical centroid of an AR region candidate.

    maximum latitude: `80`, maximum latitude of the geometrical centroid of an AR region candidate.

For more details regarding these parameters, as well as the IPART algorithm, please refer to our published works:

Xu, G., Ma, X., Chang, P., and Wang, L.: Image-processing-based atmospheric river tracking method version 1 (IPART-1), Geosci. Model Dev., 13, 4639–4662, https://doi.org/10.5194/gmd-13-4639-2020, 2020.

Or the Github repository that houses the IPART module:

https://github.com/ihesp/IPART

# 1. 概述
本数据集为北半球大气河流（atmospheric river, AR）检测目录，基于6小时间隔的ERA5再分析数据集（ERA5 reanalysis dataset）与基于图像处理的大气河流追踪（Image-Processing based Atmospheric River Tracking, IPART）算法构建。

### 数据时间范围
1979年1月1日至2020年12月31日，时间分辨率为6小时。

### 数据空间范围
北半球全域（包含陆地与海洋），空间分辨率为0.25°×0.25°经纬度。

ERA5的输入数据包含：
- 北向水汽通量垂直积分，单位为kg/(m·s)
- 东向水汽通量垂直积分，单位为kg/(m·s)

北半球范围（0°~90°N）、0.25°×0.25°经纬度分辨率的数据可从https://www.ecmwf.int/en/forecasts/datasets/reanalysis-datasets/era5 获取。

用于大气河流检测与追踪的IPART Python模块v3.0.8版本存储于https://zenodo.org/record/4164826，采用知识共享署名4.0国际许可协议（Creative Commons Attribution 4.0 International）开源，其开发仓库位于GitHub：https://github.com/ihesp/IPART。

# 2. 文件命名规范
数据文件遵循以下命名规则：
`ar_YYYYMM.nc`
其中：
- YYYY：4位年份数字
- MM：2位月份数字

例如`ar_199902.nc`代表1999年2月的大气河流检测结果。
月份采用自然月，闰年包含2月29日。

# 3. 数据格式
数据以netCDF格式（netCDF）存储。
每个数据文件包含一个三维数组，形状为`(t, 360, 1440)`，各维度含义如下：
- `t`：时间维度长度。由于数据为6小时间隔，t等于当月天数×4。
- `360`：纬度维度，范围为0°~90°N，步长0.25°。
- `1440`：经度维度，范围为80°E~440°E（向东偏移80°以将太平洋与大西洋纳入同一区域），步长0.25°。

每个时间切片对应一幅北半球地图，网格单元内为整数值，可能的取值包括：
- 0：代表该网格单元未检测到大气河流
- 1, 2, …, n：整数标签，分别对应一个大气河流实体的区域范围。

# 4. IPART算法核心参数
使用IPART Python模块从ERA5数据中检测大气河流时，所用核心参数如下：
1. THR滤波核：`[16, 13, 13]`。其中16代表16个时间切片，对应6小时间隔下的4天；13代表13个网格单元，对应0.25°经纬度下的约325 km。需注意，该时空长度均为滤波核尺寸的一半。
2. 最小面积：`50×10^4 km²`，即大气河流候选区域的最小尺寸。
3. 最大面积：`1800×10^4 km²`，即大气河流候选区域的最大尺寸。
4. 最小长宽比：`2.0`，即大气河流候选区域的最小长度/宽度比值。
5. 最小长度：`2000 km`，即大气河流候选区域的最小长度。
6. 最小纬度：`20°`，即大气河流候选区域几何质心的最小纬度。
7. 最大纬度：`80°`，即大气河流候选区域几何质心的最大纬度。

如需了解更多参数细节及IPART算法的相关内容，请参考我们已发表的研究：
Xu, G., Ma, X., Chang, P., and Wang, L.: Image-processing-based atmospheric river tracking method version 1 (IPART-1), Geosci. Model Dev., 13, 4639–4662, https://doi.org/10.5194/gmd-13-4639-2020, 2020.

或IPART模块的GitHub开发仓库：
https://github.com/ihesp/IPART