I want to reproject Global Precipitation Measurement (GPM) satellite gridded precipitation data from WGS84 to UTM33 for Norway (60N, high latitude). GPM is in WGS84, Geographic coordinate system. Grid resolution is 0.1x0.1 degree. Which pixel size need I select for output size to preserve the rainfall values of areal location?
Rule of thumb is that 0.1 degree is approximately 10km. However, it is known that 0.1degree latitude is approximately 11km but 0.1degree longitude varies and it is approximately 11km at equator and 6km at 60-degree North latitude location. When I reproject GPM data from WGS84 to UTM33 using ArcMap, ArcMap gives 8421.88m x 8421.88m pixel size as default values for reprojected UTM33. I try to do the same with gdalwarp command line and I do not set the output resolution and the gdalwarp gives 7278.69m x 7278.69m as default pixel size. Which pixel size in km can be approximately equal to 0.1x0.1 degree?
In addition to that, there are different resampling methods (Nearest neighbour, Bilinear, Cubic and Majority) available. Which method is suitable for this case? As rainfall is continuous data, I can use bilinear but for example, if one pixel has extreme value and the neighbour does not have. Then bilinear diminishes the value. In this situation Nearest neighbour is suitable but reprojection and selection of pixel size distort the original pixel shape and reprojected pixel may fall into the neighbouring pixel.
My aim is to compare the Global Precipitation Measurement (GPM) satellite rainfall data with Radar rainfall data. Radar data is in UTM33 projected coordinate system and grid resolution is in 1kmx1km. To bring both data to the same coordinate system, I reproject GPM data from WGS84 to UTM33 and I plan to aggregate radar 1kmx1km to the GPM output pixel size.
It can be noted that GPM (0.1x0.1 degree) is coarser than Radar (1kmx1km). Alternatively, is it advisable to keep GPM in WGS84 coordinate and convert 1kmx1km radar data from UTM33 to WGS84 and then aggregate to 0.1x0.1 degree for comparison?
I am not clear about the right procedure for reprojection. Is there any standard practice for reprojection/resampling climate (precipitation) data?
You can calculate using this online calculator: http://www.csgnetwork.com/degreelenllavcalc.html that for 60 degrees North gives the following:
Length Of A Degree Of Latitude In Meters: 111412.24020000001
Length Of A Degree Of Longitude In Meters: 55799.979000000014
There is a Power Point presentation in Internet with graphics and numbers. I´ve attached it to this answer...
You can also search for an Excel Spreadsheet:
https://www.uwgb.edu/dutchs/UsefulData/UTMFormulas.HTM
Hope it helps.
Thank you for your answer with information and useful conversion tools.
In my case, I need reproject the grids from WGS84 to UTM33 and resample the data to the projected UTM33 grids.
I want to have reprojected(UTM33) square grids, for example 8kmx8km grids. I believe 8kmx8km is more appropriate at this location(Latitude-60N) than 10kmx10km but not sure.
When I reproject, there will always be a distortion in shape and uncertainty in resampling. I read that this problem is known as Modifiable Areal Unit Problem (MAUP) in spatial analysis. We can find a good description of MAUP at this link (http://gispopsci.org/maup/). However, spatial climate data (continuous raster data, e.g., Satellite precipitation data) is often reprojected and resampled from one coordinate to another coordinate system. How do they do? Is there any standard procedure for that?
My main question is that, how can I decide the optimum/appropriate pixel size for reprojected grids? Here, my objective is to minimise the uncertainty caused by reprojection and resampling of raster dataset.
Alternatively, Is it more advantageous if I keep the coarser grid (GPM satellite WGS84 grid) and reproject and aggregate the finer grid (Radar UTM33 1kmx1km grid) to coarser (WGS84 0.1x0.1degree)?
The correct resolution will be half the sample's spacing (no matter the coordinate datum you are using). Please read on Nyquist-Shannon sampling theorem. Using higher resolution than the ideal, you will get unrealistic artifacts, lower resolution you will get a poor image. Based on this principle, most GIS sotfwares nowadays will automatticaly calculate an average spacing fo your data, and will sugest you the "most correct" value to use.
I agree with the whole Marco´s comment on the sample´s spacing. Regarding the distortion, any cartographic projection system wil introduce a distortion, depending on the scale and on the shape and extension of the projected area. The projection systems like UTM (or Gauss Krüger) are conformal transformations, it means that they maintain the angles unchanged, you vill project a square area and you will get a projected square with the same angles. For UTM the width of the area (it means the UTM Zone) in which you will not have a big distortion is six longitude degrees wide . Other transversal projections like Gauss Kruger have zones of three degrees wide. For this kind of proyection you will get bigger distortion if you project areas with a bigger size than one degree and a half (measuring from the central meridian of the zone. For UTM you will get bigger distorsion if you project points that are outside the UTM zone of six degrees, let say if you move more than three degrees from the central meridian of the zone. This distortion will be more evident if you work at 1.50.000 scale and less evident if you work at 1.500.000 scale. For 8 Km. cell size the projection scale should be 1:40.000.000.
Regards
Gabriel
- Badges
- Science topic