Modelling dNBR

Modelling dNBR

Predicting dNBR based on elevation and northness

In MATLAB 2014a I used the group statistics [grpstats()] to calculate the average dNBR by elevation (10m bin sizes) and northness (0.1 bins). Next I used Matlab's CurveFittingTool to fit the distributions using least-squares regression. A Gaussian model was fit to a function of the average dNBR (from MTBS) and elevation. There was a strong correlation between dNBR and elevation 

Increasing northness showed a positive correlation with 

dNBR versus elevation Gaussian model
MEDIAN
General model Gauss1:
     f(x) =  a1*exp(-((x-b1)/c1)^2)
Coefficients (with 95% confidence bounds):
       a1 =       258.7  (252.4, 265)
       b1 =        1974  (1950, 1999)
       c1 =         957  (912.3, 1002)
Goodness of fit:
  SSE: 1.233e+05
  R-square: 0.8319
  Adjusted R-square: 0.83
  RMSE: 26.1


MAXIMUM

General model Gauss1:
     f(x) =  a1*exp(-((x-b1)/c1)^2)
Coefficients (with 95% confidence bounds):
       a1 =        1146  (1123, 1169)
       b1 =        2099  (2073, 2125)
       c1 =       967.8  (926, 1010)
Goodness of fit:
  SSE: 1.726e+06
  R-square: 0.9089
  Adjusted R-square: 0.9079
  RMSE: 97.65
STANDARD DEVIATION
General model Gauss1:
     f(x) =  a1*exp(-((x-b1)/c1)^2)
Coefficients (with 95% confidence bounds):
       a1 =       257.8  (252.1, 263.6)
       b1 =        2183  (2158, 2209)
       c1 =       830.2  (793.1, 867.3)
Goodness of fit:
  SSE: 9.43e+04
  R-square: 0.9292
  Adjusted R-square: 0.9284
  RMSE: 22.83

 dNBR versus Northness Gaussian model

MEDIAN
General model Gauss1:
     f(x) =  a1*exp(-((x-b1)/c1)^2)
Coefficients (with 95% confidence bounds):
       a1 =       189.3  (164.6, 213.9)
       b1 =        1.55  (-0.9604, 4.061)
       c1 =       5.156  (0.9536, 9.359)
Goodness of fit:
  SSE: 984.7
  R-square: 0.7558
  Adjusted R-square: 0.7286
  RMSE: 7.396

MAXIMUM
General model Gauss1:
     f(x) =  a1*exp(-((x-b1)/c1)^2)
Coefficients (with 95% confidence bounds):
       a1 =        1177  (1157, 1197)
       b1 =     0.01042  (-0.1668, 0.1876)
       c1 =       4.312  (2.904, 5.72)
Goodness of fit:
  SSE: 1.471e+04
  R-square: 0.3692
  Adjusted R-square: 0.2991
  RMSE: 28.58

STANDARD DEVIATION
General model Gauss1:
     f(x) =  a1*exp(-((x-b1)/c1)^2)
Coefficients (with 95% confidence bounds):
       a1 =       190.8  (185.4, 196.3)
       b1 =      0.3672  (0.1814, 0.553)
       c1 =       2.497  (1.958, 3.037)
Goodness of fit:
  SSE: 1274
  R-square: 0.7897
  Adjusted R-square: 0.7664
  RMSE: 8.413

Incorporating existing vegetation into future dNBR predictions

The predicted models of dNBR (shown above) do not take into account the influence existing vegetation on the likely dNBR should it burn in a future fire.

In order to account for the impact of vegetation on dNBR I suggest using one of the Gridmetric layers for vegetation - either mean height above ground or max height above ground