In the climate change studies, we have to use CMIP5 data sets in NETCDF files. Should we interpolate between two points when getting prcp data for a point OR get the nearest point ? What is the best ? I usually use MATLAB, but I need to what ArcGIS do when we request for a point. Thanks in advance
Hi Madusanka,
I'm not sure it's valid to assume your interpolation is valid for point-scale, without extreme caution. (I'm guessing you're trying to make the CMIP5 data comparable with a raingauge, for example.) A CMIP5 model grid box might be of order 100-200km size. This means that the rainfall therein is spread over the entire box, and your rain rates will never reach the high intensities found at a single point station. A common manifestation of this is that in a histogram of the rain rates from model data, high intensity rainfall is too infrequent, whereas very low rainfall amounts (drizzle) are overestimated. Apologies if I have misinterpreted your question.
Andy.
Hi Madusanka,
I agree with Andy's advice as to being very cautions when analyzing the data. I also want to add that due to the scale of an individual grid box, the degree of comparability significantly depends on the temporal scale you want to analyze. As Andy mentioned, when comparing precipitation from GCM and observations on daily or even sub-daily timescales, it is naive to assume that the GCM should perform well in terms of getting the distribution of rainfall right. However, comparing monthly, seasonal or annual means is possible and can give you some skill (depending on the model of course).
Overall, you have to remember that in GCMs, modelled surface precipitation is what comes out of a vast number of parameterisation schemes (which all carry uncertainties and ambiguities) and it is fair to say that there is almost no other quantity that GCMs disagree more on.
See also Chapter 9 of IPCC's 5th Assessment Report.
Karsten
Dear Andy and Karsten, For the bias correction for each station, I am using quantile mapping using qmap package in R. Anyway before the bias correction we have to derive the model prcp data from NC files. Actually my question is about that derivation of raw model data. Anyway both of you have given the answer.
Rainfall is given for a grid, not for a point, so there is no need to interpolate between the middle points of the grids in order to achieve the rainfall for a gauge point.
Please let me know whether I am correct.
Yes, the value given for one "grid box" is essentially to be understood as the mean value averaged over that grid box. Estimates of subgrid-scale values are generally not provided I think.
How to find representative row and column of study area in a NetCDF file? If I have netCDF file having Precipitation data in this form 480*260*25 i.e Longitude*latitude*time then how i can extract precipitation data of a specified area?
yes I have the same question, how we can extract data for specified area from netcdf?
This a very popular question. I would appreciate if someone can answer this question with an evidence. Which interpolation method do you recommend for interpolation of GCM cmip5 outputs?
The nearest neighbour give the average value of the grid to each point located in the grid. According to the attached the author claimed that the NN method is better than methods such as IDM because:
"The major reason is that we intended to preserve the
original climate signal of the GCMs even though the large grid spacing.
Involving more GCM grid cell data on the interpolation procedure
(as in Inverse Distance Weighting–IDW) may result to significant
information dilution, or signal cancellation between two or more grid
cell data from GCM outputs."
But in my opinion maybe the IDM is a better choice because I think as Karsten Peters said "estimates of subgrid-scale values are generally not provided".
I disagree with Jayshir Patel about using the elevation because the model is in the global scale.
- Badges
- Science topic