Project

General

Profile

« Previous | Next » 

Revision 100e95ca

Added by Adam M. Wilson almost 11 years ago

Adding a new option 2b

View differences:

climate/research/LST_evaluation.md
28 28 
# TMax~LST relationship seasonal variability
29 29 

  		




  30
  30
  
    
First let's explore the variability of the TMax~LST relationship by month, grouped by tile.  In this plot grey lines indicate a Tmax~LST regression within each tile (stations may be present in multiple tiles). Variability among grey lines represents spatial variability in the fitted lapse rate and the heavy black line is overall mean relationship (by month).


  		




  31
  
  
    
![plot of chunk unnamed-chunk-4](http://i.imgur.com/RZbujnI.png) 


  		




  
  31
  
    
![plot of chunk unnamed-chunk-4](http://i.imgur.com/KYSRkMV.png) 


  		




  32
  32
  
    

  		




  33
  33
  
    

  		




  34
  34
  
    

  		




  35
  35
  
    
# Comparison of monthly means with 3-month moving window


  		




  36
  36
  
    
Here we compare the monthly means with a three month moving window (e.g. January mean LST with December-February mean LST).  Note that the relationship is very good (R^2 >0.95) but slightly weaker in winter months, likely due primarily to seasonal minimums.  Heavy black line is 1:1, and thin line is the fitted regression.


  		




  37
  37
  
    

  		




  38
  
  
    
![plot of chunk unnamed-chunk-5](http://i.imgur.com/aAmbMaz.png) 


  		




  
  38
  
    
![plot of chunk unnamed-chunk-5](http://i.imgur.com/kInQcBr.png) 


  		




  39
  39
  
    

  		




  40
  40
  
    

  		




  41
  41
  
    
# Comparison of monthly means with 2-month moving window that does not include the month


  		




  42
  42
  
    
Here we compare the monthly means with a two month moving window that does not include the month of interest (e.g. January mean LST with December and February mean LST, but not including the January LST).  Heavy black line is 1:1, and thin line is the fitted regression.


  		




  43
  43
  
    

  		




  44
  
  
    
![plot of chunk unnamed-chunk-6](http://i.imgur.com/AeJ0AFq.png) 


  		




  
  44
  
    
![plot of chunk unnamed-chunk-6](http://i.imgur.com/8AfNmiq.png) 


  		




  45
  45
  
    

  		




  46
  46
  
    

  		




  47
  47
  
    
Now let's look at the differences between the 1-month and 2-month LST values.  These represent a measure of how wrong we would be if we only had data from the two surrounding months and not the month in question.  


  		




  ......




  51
  51
  
    
histogram(dlst2$lst - dlst2$lst2m, col = "grey", xlab = "Anomolies (1 month - 2 month means)")


  		




  52
  52
  
    
```


  		




  53
  53
  
    

  		




  54
  
  
    
![plot of chunk unnamed-chunk-7](http://i.imgur.com/qIlsdWo.png) 


  		




  
  54
  
    
![plot of chunk unnamed-chunk-7](http://i.imgur.com/7E9SHnc.png) 


  		




  55
  55
  
    

  		




  56
  56
  
    
The 95th quantile of the absolute value is only 4.2, so the differences are quite small. From this analysis, it appears that broadening the temporal window will maintain a relatively consistent estimate of LST (R^2 ranged from 0.88-0.9) even when using only data from the surrounding months.


  		




  57
  57
  
    

  		




  ......




  60
  60
  
    

  		




  61
  61
  
    

  		




  62
  62
  
    

  		




  63
  
  
    
![plot of chunk unnamed-chunk-8](http://i.imgur.com/Ve0mC1C.png) 


  		




  
  63
  
    
![plot of chunk unnamed-chunk-8](http://i.imgur.com/wOsUVGT.png) 


  		




  64
  64
  
    

  		




  65
  65
  
    

  		




  66
  66
  
    
# Processing Options


  		




  ......




  145
  145
  
    
```


  		




  146
  146
  
    

  		




  147
  147
  
    

  		




  148
  
  
    
Let's see what that looks  for the 10 example stations above. The blue line is the observed LST value and the pink line is the predicted LST from the sinusoidal function.


  		




  149
  
  
    
![plot of chunk unnamed-chunk-10](http://i.imgur.com/oNteXlf.png) 


  		




  
  148
  
    
Let's see what that looks  for the 10 example stations above. The red line is the observed LST value and the black line is the predicted LST from the sinusoidal function.


  		




  
  149
  
    
![plot of chunk unnamed-chunk-10](http://i.imgur.com/WYgS7qx.png) 


  		




  150
  150
  
    

  		




  151
  151
  
    
And a histogram of the residuals for these stations:


  		




  152
  
  
    
![plot of chunk unnamed-chunk-11](http://i.imgur.com/1z4IHQs.png) 


  		




  
  152
  
    
![plot of chunk unnamed-chunk-11](http://i.imgur.com/qRekbih.png) 


  		




  153
  153
  
    

  		




  154
  154
  
    

  		




  155
  155
  
    
Not bad... Now let's drop 25% of the observations from each station and do it again:


  		




  156
  
  
    
![plot of chunk unnamed-chunk-12](http://i.imgur.com/6zkd5CB.png) 


  		




  
  156
  
    
![plot of chunk unnamed-chunk-12](http://i.imgur.com/n6ZP6yB.png) 


  		




  157
  157
  
    

  		




  158
  158
  
    

  		




  159
  159
  
    
And the  of residuals for these 10 stations with 25% of the observations removed:


  		




  160
  
  
    
![plot of chunk unnamed-chunk-13](http://i.imgur.com/c1Wyh9J.png) 


  		




  
  160
  
    
![plot of chunk unnamed-chunk-13](http://i.imgur.com/8Z1Niwd.png) 


  		




  161
  161
  
    

  		




  162
  162
  
    

  		




  163
  163
  
    
### Apply this function to the full CONUS dataset described above.


  		




  164
  164
  
    
Now look at the distributions of the residuals for the full conus dataset.


  		




  165
  
  
    
![plot of chunk unnamed-chunk-14](http://i.imgur.com/od4iR2S.png) 


  		




  
  165
  
    
![plot of chunk unnamed-chunk-14](http://i.imgur.com/iSdwpOe.png) 


  		




  166
  166
  
    

  		




  167
  167
  
    
And summarize the residuals into RMSE's by station:


  		




  168
  
  
    
![plot of chunk unnamed-chunk-15](http://i.imgur.com/KIV2rBk.png) 


  		




  
  168
  
    
![plot of chunk unnamed-chunk-15](http://i.imgur.com/M29nxV1.png) 


  		




  169
  169
  
    

  		




  170
  170
  
    
So the vast majority of stations will have a RMSE of <5 using this simple method even if 25% of the data are missing.


  		




  171
  171
  
    

  		












Also available in:  Unified diff