# DEM, points values and elevation to create a map of temeperature

I am using SAGA - Universal Kriging (Global) for this, but I am stucked. I have grid (.tif) for Serbia with elevation values, and .shp file with points that have mean yearly temperature. I want to create a map using these two files and to take in to the account that Temperature gradient in Serbia is 0.6°C. Temperature gradient means that for every 100m temperature is changed for 0.6°C Also i have the elevation for the .shp file with points, but i don't know how to combine all this. So I want to display a map of the temperature, extrapolated around the known points according to the DEM. Is that possible?

• Given you are trying to create a trend surface which later used as an External Drift in UK, may i suggest following steps: (1) multiply temp gradient (-0.6/100m) with elevation grid, then (2) calibrate the map from (1) with your points by `Multiple Regression Analysis` tool (in `Spatial and Geostatistics | Regression`). I think `Regression` output grid is what you need for UK process. (I have not been successful with my laptop as this is intense computing). – Kazuhito Aug 28 '17 at 9:23
• I don't know to the first step. I need more detailed answer please. – Markoneo Aug 28 '17 at 11:58
• Yep,sorry... For the Step (1), please use `Grid Calculator` (in Tools | Grid | Calculus). With your elevation as Grids (g1), input formula will look like `20 - 0.006 * g1` (20-degC as just an indicative temp at elev=0). Calculated temperature itself is less important (we only need trend) but it helps us to visually compare it with the point data. (Please note it is probably better to smooth your elevation grid beforehand). – Kazuhito Aug 28 '17 at 12:25
• @Markoneo How about (1) create residual grid, and (2) subtract (or add) this residual? In detail; Step (1) Calculate the difference (9.6 - 9.3) at the point and store it to the attribute table. Then interpolate (e.g. `Grid - Spline Interpolation | Multilevel B-Spline`) to generate the grid. Step (2) Subtraction - it can be done either by `Grid Calculator` or by `Grid difference` (both in `Grid - Calculus`). But...because the regression grid was meant to be trend surface, it looks within reasonable range (to me). Do you really have to match them? – Kazuhito Oct 9 '17 at 9:25

Let me try to summarize the workflow you have taken to obtain a trend surface to be an input grid to the UK algorithm.

Background of this topic would have been like this: (A) There is DEM and measured temperature data available. It is given the temperature generally follow the `-0.6 deg-C per 100m` trend. (B) In perfect world the temperature curve should be like the gray curve, but the recorded temperature show certain deviation. So a workflow (detailed below) taken was, create a surface first by Multiple Regression Analysis considering both measured temperature and elevation, and then removing residual errors.

1. Create a "pseudo- temperature grid" by multiplying temperature gradient with DEM. [Grid calculator] Recommended equation was `20 - 0.006 * g1` where 20 was notional (intercept) temperature. (It would have been trial and error to find best intercept value).
2. Create a Regression grid from measured temperature data and pseudo-temperature grid. [Multiple Regression Analysis Tool]
3. Create a residual grid by (a) Calculate difference between measured temperature and the regression grid. [Subtract] and store the difference into the attribute table of the measured point shapefile. (b) Interpolate the difference. Recommended tool was [Multilevel B-Spline] but any gridding tool does this job. (c) Subtract this residual grid from the regression grid. [Grid calculator]

Final output is a temperature trend surface (red dots; above picture), conditioned at the measured locations.

Markoneo, please edit as you think necessary. I likely have missed some steps you have taken.