When running the MRMS data, the following error occurs

[HY-10]

Hello, I have a question regarding processing MRMS data. When running the MRMS data, the following error occurs, how should I adjust the parameters?

rainrate, metadata = dimension.aggregate_fields_space(rainrate, metadata, 2000) ValueError: space_window does not equally split R

[dnerini]

hi @HY-10, the method is basically complaining that one of your spatial dimensions (which are assumed to be the last two dimensions of your input) is not a multiple of 2000, see the code for details: https://github.com/pySTEPS/pysteps/blob/master/pysteps/utils/dimension.py#L194

In essence, you have to make sure that the shape of your x and y dimensions, multiplied by the ypixelsize and xpixelsize respectively, is a multiple of 2000.

In your example, what's the shape of rainrate and the content of the metadata dictionnary?

[HY-10]

I'm sorry, but I still don't understand how to get the appropriate value for the space_window. The MRMS data I'm using has a shape of rainrate.shape=(3,875,1750). Thank you for your guidance.

[dnerini]

what's the resolution of your pixels? can you copy/paste here the content of the metadata dictionary?

[HY-10]

(3, 875, 1750) {'institution': 'NOAA National Severe Storms Laboratory', 'xpixelsize': 0.04, 'ypixelsize': 0.04, 'unit': 'mm/h', 'accutime': 2.0, 'transform': None, 'zerovalue': 0, 'projection': '+proj=longlat +ellps=IAU76', 'yorigin': 'upper', 'threshold': 0.0125, 'x1': -130.00000000042866, 'x2': -60.000001999571296, 'y1': 20.000000999571306, 'y2': 55.00000000042869, 'cartesian_unit': 'degrees', 'timestamps': array([datetime.datetime(2019, 6, 10, 0, 56), datetime.datetime(2019, 6, 10, 0, 58), datetime.datetime(2019, 6, 10, 1, 0)], dtype=object)}

像素的分辨率是多少？您可以在这里复制/粘贴字典的内容吗？metadata

[aperezhortal]

Hi @HY-10,

Due to the large size of the dataset (3500 x 7000), the importer reduces the original 1kmx1km resolution to 4km x 4km.

Hence, we end with a grid size of (875, 1750). Now, in the metadata, the xpixelsize/ypixelsize is expressed in degrees. That is why you see 0.04 (that is roughly 4km in CONUS).

Now, the aggregate_fields_space can only reduce (or keep) the resolution of the input data. So, let's say you want to downscale the data by 5 (to 20km).

Both 875 and 1750 are divisible by 5. Hence, we use *window=0.045** (remember that the window needs to be in the same units as the pixel size ).

upscaled_rainrate, upscaled_metadata = dimension.aggregate_fields_space(rainrate, metadata, 0.04*5)
print(upscaled_metadata)

{'institution': 'NOAA National Severe Storms Laboratory',
 'xpixelsize': 0.2,
 'ypixelsize': 0.2,
 'unit': 'mm/h',
 'accutime': 2.0,
 'transform': None,
 'zerovalue': 0,
 'projection': '+proj=longlat  +ellps=IAU76',
 'yorigin': 'upper',
 'threshold': 0.0125,
 'x1': -130.00000000042866,
 'x2': -60.000001999571296,
 'y1': 20.000000999571306,
 'y2': 55.00000000042869,
 'cartesian_unit': 'degrees',
 'timestamps': array([datetime.datetime(2019, 6, 10, 0, 0)], dtype=object)}

In which resolution do you need the data? For example, if you need 2km, you need to load the data in the original 1km resolution and then upscale it using a window of pixelsize*2 (0.02).

Let us know if this helps.

Andres

[HY-10]

Hi @HY-10,

Due to the large size of the dataset (3500 x 7000), the importer reduces the original 1kmx1km resolution to 4km x 4km.

Hence, we end with a grid size of (875, 1750). Now, in the metadata, the xpixelsize/ypixelsize is expressed in degrees. That is why you see 0.04 (that is roughly 4km in CONUS).

Now, the aggregate_fields_space can only reduce (or keep) the resolution of the input data. So, let's say you want to downscale the data by 5 (to 20km).

Both 875 and 1750 are divisible by 5. Hence, we use *window=0.045** (remember that the window needs to be in the same units as the pixel size ).

upscaled_rainrate, upscaled_metadata = dimension.aggregate_fields_space(rainrate, metadata, 0.04*5)
print(upscaled_metadata)

{'institution': 'NOAA National Severe Storms Laboratory',
 'xpixelsize': 0.2,
 'ypixelsize': 0.2,
 'unit': 'mm/h',
 'accutime': 2.0,
 'transform': None,
 'zerovalue': 0,
 'projection': '+proj=longlat  +ellps=IAU76',
 'yorigin': 'upper',
 'threshold': 0.0125,
 'x1': -130.00000000042866,
 'x2': -60.000001999571296,
 'y1': 20.000000999571306,
 'y2': 55.00000000042869,
 'cartesian_unit': 'degrees',
 'timestamps': array([datetime.datetime(2019, 6, 10, 0, 0)], dtype=object)}

In which resolution do you need the data? For example, if you need 2km, you need to load the data in the original 1km resolution and then upscale it using a window of pixelsize*2 (0.02).

Let us know if this helps.

Andres

hi @aperezhortal , Thank you very much! This has been very helpful to me. Now I understand how the value of space_window is obtained.

[HY-10]

hi@

Hi @HY-10,

Due to the large size of the dataset (3500 x 7000), the importer reduces the original 1kmx1km resolution to 4km x 4km.

Hence, we end with a grid size of (875, 1750). Now, in the metadata, the xpixelsize/ypixelsize is expressed in degrees. That is why you see 0.04 (that is roughly 4km in CONUS).

Now, the aggregate_fields_space can only reduce (or keep) the resolution of the input data. So, let's say you want to downscale the data by 5 (to 20km).

Both 875 and 1750 are divisible by 5. Hence, we use *window=0.045** (remember that the window needs to be in the same units as the pixel size ).

upscaled_rainrate, upscaled_metadata = dimension.aggregate_fields_space(rainrate, metadata, 0.04*5)
print(upscaled_metadata)

{'institution': 'NOAA National Severe Storms Laboratory',
 'xpixelsize': 0.2,
 'ypixelsize': 0.2,
 'unit': 'mm/h',
 'accutime': 2.0,
 'transform': None,
 'zerovalue': 0,
 'projection': '+proj=longlat  +ellps=IAU76',
 'yorigin': 'upper',
 'threshold': 0.0125,
 'x1': -130.00000000042866,
 'x2': -60.000001999571296,
 'y1': 20.000000999571306,
 'y2': 55.00000000042869,
 'cartesian_unit': 'degrees',
 'timestamps': array([datetime.datetime(2019, 6, 10, 0, 0)], dtype=object)}

In which resolution do you need the data? For example, if you need 2km, you need to load the data in the original 1km resolution and then upscale it using a window of pixelsize*2 (0.02).

Let us know if this helps.

Andres

hi @aperezhortal , I want to load the data in the original 1km resolution and then upscale it using a window of pixelsize*2(0.02). So, what should I do

[HY-10]

嗨，

由于数据集的体积较大 （3500 x 7000），导入器将原来的 1kmx1km 分辨率降低到 4km x 4km。

因此，我们以 （875， 1750） 的网格大小结束。现在，在元数据中，xpixelsize/ypixelsize 以度表示。这就是为什么你会看到 0.04（在 CONUS 中大约是 4 公里）。

现在，aggregate_fields_space只能降低（或保持）输入数据的分辨率。因此，假设您想将数据缩小 5（到 20 公里）。

875 和 1750 都可以被 5 整除。因此，我们使用 *window=0.045**（请记住，窗口需要与像素大小的单位相同）。

upscaled_rainrate, upscaled_metadata = dimension.aggregate_fields_space(rainrate, metadata, 0.04*5)
print(upscaled_metadata)

{'institution': 'NOAA National Severe Storms Laboratory',
 'xpixelsize': 0.2,
 'ypixelsize': 0.2,
 'unit': 'mm/h',
 'accutime': 2.0,
 'transform': None,
 'zerovalue': 0,
 'projection': '+proj=longlat  +ellps=IAU76',
 'yorigin': 'upper',
 'threshold': 0.0125,
 'x1': -130.00000000042866,
 'x2': -60.000001999571296,
 'y1': 20.000000999571306,
 'y2': 55.00000000042869,
 'cartesian_unit': 'degrees',
 'timestamps': array([datetime.datetime(2019, 6, 10, 0, 0)], dtype=object)}

您需要哪种分辨率的数据？例如，如果需要 2km，则需要以原始的 1km 分辨率加载数据，然后使用像素大小*2 （0.02） 的窗口对其进行放大。

如果有帮助，请告诉我们。

安德烈斯

hi @aperezhortal , I want to load the data in the original 1km resolution and then upscale it using a window of pixelsize*2(0.02). So, what should I do？