Hi Amrit,

An easy way to increase the spatial resolution of gridded data is by using "spatial interpolation". A good overview of spatial interpolation techniques is presented in: Li, J., Heap, A,. 2008. A Review of Spatial Interpolation Methods for Environmental Scientists. Geoscience Australia, Record 2008/23 (https://www.researchgate.net/publication/246546630_A_Review_of_Spatial_Interpolation_Methods_for_Environmental_Scientists)

An implementation of the Random Forest method (one of the most advanced methods discussed in the reference above) to increase the resolution of climatic variables can be found in:

https://www.researchgate.net/publication/257298539_Spatial_Interpolation_of_McArthur%27s_Forest_Fire_Danger_Index_accross_Australia_Observational_study

Hope it helps,

Augusto

Book A Review of Spatial Interpolation Methods for Environmental Scientists

Article Spatial Interpolation of McArthur's Forest Fire Danger Index...

However, always bear in mind that you can dress up a coarse-resolution dataset as fine resolution, but underlying it is information that is only coarse. That is, it may look finer, and it may appear to be much more detailed, but the downscale is only as good as the method and information that was added in the downscaling process. For temperature, such downscaling is feasible, thanks to the known properties of the adiabatic lapse rate. For precipitation, the properties are far more complex. So just a word of caution. All the best, Town Peterson

Andrew's point is a very important consideration when selecting a method for spatial interpolation of gridded data. That is why I recommended the Random Forest method. It belongs to the so-called machine learning techniques. The method uses auxiliary variables to do the spatial interpolation, i.e. if you have gridded data at the right resolution, this data can be entered as  further information for the method to find the correct value at interpolated points. A good discussion of the Random Forest method is presented in the reference provided above, in particular notice that in this method we use as auxiliary variables, to help with the interpolation, a number of existing gridded variables such as: location parameters (lat, lon), Elevation above sea level, Enhanced vegetation Index, Evaporation, Precipitation, Temperature, etc.

Hope this helps,

Augusto

Dear Hugo Hidalgo,

Are there r -packages or some tools for PRISM method?

I fully agree to Andrew's point in this discussion. Temperature might be relatively easy to interpolate on a finer grid since it mostly depends on the terrain (elevation, exposition), precipitation however is usually difficult. 

I do not want to comment on the suggested methods such as the Random Forest method to improve the resolution of precip, however, for me it would be important that the local weather is related to the large-scale circulation. That's why I would suggest a weather type approach for downscaling using large-scale climate data and e.g. pressure variables (SLP, geopotential heighs in different levels) as predictors. This also has the advantage that identified relationships can be used to downscale future climate projections, because these pressure variables are known to be more "reliable", compared to large-scale precipitation data, under the assumption of stationarity, of course.

If you are "only" interested in correcting large-scale gridded observation data with station data (only for the past, and not for climate projections), simple bias correction approaches might be relevant for you, e.g. the local intensity scaling for precip (e.g. Schmidli, 2006) or a power law transform. For temperature, you also have such easy-going options such as the monthly scaling.

Just regridding the data on a finer grid is certainly not a good option because since it usually does not contain more information than the large-scale fields (see Andrew's comment).

Patrick     

Dear Amrit,

With Andrews comments in mind, I would recommend you explore the SDSM (Statistical downscaling of Global Models) tool by Rob and Chris. This tool can be used to synthesize plausible daily weather series as well as exotic variables (such as tidal surge) and if you like gridded data. The SDSM is best described as a "conditional weather generator, because atmospheric circulation indices and regional moisture variables are used to estimate time-varying parameters describing daily weather at

individual sites (e.g. precipitation occurrence or daily mean temperatures). Since these are based on calibrated predictand-predictor relationships at individual sites, you are likely to achieve a high resolution data set for sites of your region of interest based on the gridded data sets and atmospheric conditions. The challenging part however is that your data set should be at a daily time step.

Richard