Haiti - Accumulated precipitation Hurricane Matthew

  • 800+ Downloads
  • This dataset updates: Never


Related Showcases

Source Global Precipitation Measurement
Date of Dataset October 13, 2016-October 13, 2016
Updated 9 August 2018
Expected Update Frequency Never
Methodology Direct Observational Data/Anecdotal Data
Caveats / Comments

Data source: https://pmm.nasa.gov/data-access/downloads/gpm Admin areas: https://data.humdata.org/dataset/hti-polbndl-adm1-cnigs-zip

Steps/assumptions: - the entire accumulated rainfall for 3, 4 and 5 october. (the data for 6 october indeed showed that the rain was gone). - I used the .tif/.twf files from the given source and used 3 of those combinations in total: 20161003 23.30-23.59-1day, 20161004 23.30-23.59-1day and 20161005 23.30-23.59-1day. Since these respectively give the accumulated rainfall for 1 whole day ending at 23.59. - I used 'Zonal statistics' in QGIS to compute the average per administrative area and used the resulting 'mean' as output variable. Often, the administrative level 3 area was smaller than one raster point, but the resulting mean values seemed to reflect the average of the neighbouring raster point(s) when i inspected a few. - In the csv/xlsx I multiplied by 0.1, since the units in the data are in 0.1mm according to https://pps.gsfc.nasa.gov/Documents/README.GIS.pdf and the total amounts seemed to make sense like this (up to 700mm over 3 days in the worst hit areas, instead of 7000mm).

The data output is similar to this video: https://svs.gsfc.nasa.gov/cgi-bin/details.cgi?aid=12389#59042 From this video you can also find the time frames in which Haiti incurred rain in relation to the typhoon.

You can get GPM IMERG data here: https://pmm.nasa.gov/data-access/downloads/gpm. Note that IMERG tends to produce too much rainfall over land. We are working to address this problem in the next version of the algorithm, but unfortunately that won't be until near the end of 2017. You can also look at an older algorithm from the TRMM Program: https://pmm.nasa.gov/data-access/downloads/trmm. That algorithm used to involve calibration using TRMM data, but TRMM ended in 2015, so now the algorithm uses a climatological calibration. For both algorithms, there is a real-time version and later science versions. The real-time versions use whatever data is available within a short time after the observation time. The science versions have longer latency, but use more data, including raingauge observations. The data sets come in a number of different formats listed on the pages.

For the issues behind the over-estimation, I will refer you to the product developer, George Huffman, cc'd here. I don't think it is a uniform over-estimation, but I don't fully know the nature of the problem. Also, IMERG and TMPA have somewhat different estimates and deficiencies. In a conversation with George a moment ago, it sounded like the bigger issues were over the Southeastern U.S. rather than Haiti.

In terms of latency, TMPA-RT (also known as 3B42-RT) has a real-time version available in about 8 hours after the observation time while the research product TMPA (3B42) has a latency of about 2 months.

For IMERG, the real-time version is available in about 6 hours, a near-real-time-late product (with a bit more data) in 18 hours, and the final research version in about 4 months.

These are near global products, so should work for just about any region except high latitudes. There are other rainfall products available elsewhere that attempt to do a similar type of analysis (such as NOAA's CMORPH algorithm), but IMERG represents a combination of the best techniques in use by other products.

IMERG is in the awkward position of having known flaws in the current version (3) and the new version (4) still being finalized. GPM hopes for a release by the end of November 2016.

The current overestimation problem depends on the rain rate, when the rate goes above about 10 mm/hr, IMERG increasingly overestimates the rate. We picked up on this last October in South Carolina, USA, when the actual accumulations were in excess of 250 mm, but IMERG was high by a factor of two. Since then, we've seen other such examples. The real-time TMPA (product 3B42RT) doesn't have this problem, but in fact the 3B42RT accumulations over land during Matthew were notably low (while IMERG Early or Late were notably high). A quick check over Haiti using Giovanni seems to show that (land and ocean both) 3B42RT is about 60% of IMERG-L, although, as you imply, it's really hard to know what's happening there.

We compared the LATE to the EARLY GDM data and found first of all that the relative distribution over admin-areas does not change at all. second (see attached sheet) i find that the absolute amounts do not actually change that much (from a minimum factor 0.91 to a max of 1.24). what is weird is that on average the late rainfall is 5% higher than the early rainfall, which is the other way around from what the people at NASA expected. I can check with them if you like.

File Format