Hello,
I have high-resolution DEM data (10x10 m) and low-resolution climate data (1km2) in netcdf. I need to downscale the climate data to the same resolution as DEM. Actually, I got lost in so many approaches and still didn't find the right one. I am trying to downscale it in R, according to the advice by Ákos Bede-Fazekas: https://www.researchgate.net/post/Can_I_downscale_climate_data_from_Worldclim_which_will_be_used_in_species_distribution_modelling, but I got stuck in step 3. Maybe regression kriging in GIS would be applicable too, but it usually works with climatic data as points which I do not have - as I have found in the manuals. Do you have any advice on where to start and what is the best option? Or, maybe, could you, please come up with any proceeding that would be the most suitable? I am a beginner (starting PhD student) with a knowledge of R and ArcGIS.
The aim of it all is to extract climate and other variables for the points indicating occurence of the two studied plant species and map their distribution based on that.
Thank you very much!
Hi,
I think probabbly the easiesst way is to use Rescale module in ArcGIS. You can find the procedure to follow in this link: http://desktop.arcgis.com/en/arcmap/10.3/tools/data-management-toolbox/rescale.htm
Hope you can solve the issue!
Hi Terzie,
Although upscaling can work well with simple re-sampling, downscaling climatic spatial data with simple re-sampling or statistical interpolation methods can often carry inaccuracies, due to the heterogenity of parameters affecting the spatial variance of the climatic data over the domain (eg. land use, topography, etc).
If your concern is mainly about rainfall data, there are number of methods used by the researchers to downscale the datasets:
https://agupubs.onlinelibrary.wiley.com/doi/pdf/10.1029/2003WR002504
https://www.sciencedirect.com/science/article/pii/S0034425711002331
https://www.mdpi.com/2073-4441/10/4/409
Regards!
If you need to compare both datasets you can't downscale climate data to dem resolution. I think this is a wrong approach. You must upscale dem to climate data may be using a mode moving filter. Comparisons are made to the worst scale. In the contrary you are comparing high frequency with low frequency data despite the interpolation. If you need more information about this subject I think you can probably will find something learning on sampling theories.
But if your necessities comes from the field of cartography, you can choose a linear or inverse distance interpolation method combined with smooth filtering methods (that are necesarry for removing "hills & valles" noise)
Downscaling climate data is never straightforward. You can either upscale as @Gabriel recommends (in your case simply choose the closest known climate location in your netcdf) or try a dynamical downscaling with some sort of climate model. If your terrain is generally flat, then your climate data resolution is pretty good, you don’t need to bother with downscaling (temperature for example varies very little at these scales) In high slope/mountains the history is different (may be very different) and you need to search for climate models, try WRF from UCAR witch is one of the most supported. Good luck
Terezie Simakova
You can downscale the data in ArcGIS using the following approach:
1 Add netcdf layer as a file in arctool box
2 Make your netCDF as raster
3 Convert raster to points
4 Interpolate point data with required cell size into a raster surface
Cheers
Dear Terezie Simakova .
Ricardo Delgado Tellez, Indishe Senanayake and Gabriel Asato are totally right. There are several post-processing methods (e.g. external drift kriging) or dynamical downscaling approaches available for "improving" the spatial resolution of your climate data, but you always have to think about the gain of these approaches. Please keep in mind that climate data at a spatial resolution of 1km^2 can already assumed to be "high-resolution" so I would not worry too much about bringing that to even higher resolutions. Furthermore, all of these approaches also come with their own bag of problems so you might even introduce further uncertainties or biases in your output.
If you only want to compare some other data to your climate output, you have to follow the advice from Gabriel Asato and bring the higher-resolved data to the resolution of your climate data.
Cheers,
Christof
Thank you all for your answers! I forgot to stress an important fact that the region that I've been working with is mountaineous and the species were collected along the elevation gradients. Therefore, the climate variable (temperature and precipitation) would be e.g. the same in the lowland areas as on the top of the mountain because of the coarse resolution. That's not what I need and it's exactly the main reason why I've been trying to downscale the climate layers; re-scaling would not help. I do really need accurate climate data for each "point" for further species distribution modelling. Please, correct me if this approach is completely wrong. Thanks again!
Best wishes
Terezie
Terezie Simakova, We´re working on something similar here. As Christof Lorenz said, a resolution of about 1 km is good enough for most uses. We´re operating with about 0.6 km in small and step mountains. The question in your case is, if you go by the dynamical downscaling or with a static atmosferic model (finding gradients along elevation profiles for your variables). In the first case, you will need access to some HPC platform. Fort short modelation time and not very large target domains (about 74000sqkm) we got 1.6 real time modelation ratio (1 hour realtime will take 1.6 hours to compute). For the second case, you may need to go to the field and do some experiments along elevation profiles with sensors.
Hope u find a suitable solution
Ricardo