I want to to downscale precipitation and temperature for a basin (6820 km2) using SDSM model. I have 14 station with almost 100% full data (no missings) covering 25 years within the basin. After review the literature, I get confused whether to select a few representative stations and downscale each one then average their results or average the stations data then downscale. However, the basin is semi-arid with rainy days are very scarce, so poor correlation when downscale each station while averaging stations then downscale showed much more correlation. So my question is which is proper to downscale each meteorological station or averaged value of all stations in the study area?
This question is very important, also depends on the type of study, whether point based or regional based. If point based, straightaway can be taken for that location and if on regional based, you have to delineate land units within the region on the basis of bio-physical (soil general characters and problem soils, weather normals-seasonal or annual) and socio-economic (admin boundaries, irrigated areas etc) and then run the model for those land units and within the land unit the respective weather stations are identified.
you can refer 'Alternate land use options- Haryana case study' a study carried out by our group at IARI, New Delhi, India
Dear Alham
It is better to downscale individually. If your model is not performing well, you should increase your predictors domain size or increase the number of predictors.
By the way how much NCEP grid points are you considering about your study area?
Thanks
This question is very important, also depends on the type of study, whether point based or regional based. If point based, straightaway can be taken for that location and if on regional based, you have to delineate land units within the region on the basis of bio-physical (soil general characters and problem soils, weather normals-seasonal or annual) and socio-economic (admin boundaries, irrigated areas etc) and then run the model for those land units and within the land unit the respective weather stations are identified.
you can refer 'Alternate land use options- Haryana case study' a study carried out by our group at IARI, New Delhi, India
Dear Alham,
the wise way will be to divide between temperature analysis and precipitation analysis. The temperature is a field variable (or continuous variable), i.e. differentiable at each point, and you get the best estimates using the averages. The precipitation is a non-field variable, i.e. having both the points where it is differentiable and points of singularity. In the model language it is called sub-grid variable (same for cloudiness for example). So, with precipitation, you will rather use a separate-station approach. Good luck!
Dear Alham,
It is still possible to use the average of the stations especially if they lay in one GCM scale and the correlation is much better, means if the GCM predcitor for all the sations is the same, but if stations are in different GCM box it good to downscale individually. This paper will give you more information.
http://www.sciencedirect.com/science/article/pii/S2212096315000194
Good luck.
- Badges
- Science topic
- Similar topics
- Geoscience
- Atmospheric Sciences
- Climate Science