3. RESULTS AND DISCUSSIONS 
3.1 Comparison of NDVI Between Sensors 
Experiment plot means of NDVI measured by different sensors 
for the same locations and dates were used to compare the 
sensors. Table 1 shows results of regression of the ACS 470 
sensor NDVI values with the ACS 210 and Greenseeker, based 
on corresponding plot means. The sensors show linear 
relationships with high r^ values of 0.89 and 0.75, and standard 
errors of 0.06 and 0.07 NDVI units. Neither corresponding set 
of measurements shows a 1:1 relationship, with slopes of 1.3 
and 0.91, and offsets of -0.1170 and 1.2799. Given that the 
measurements were made over the growing season (and a range 
of dates), these results would indicate systematic differences in 
the NDVI values generated by the sensors. 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B8, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
  
  
  
Data Used N r? Std. Bo» B1 
Error 
ACS 470 on ACS 210 | 106 0.88 0.06 -0.1170, 
(Horsham 2010) 1.2799 
ACS 470 on 637 0.75 0.07 0.1248, 
Greenseeker (Lubeck 0.9126 
2009, 2010) 
  
  
  
  
  
  
  
Table 1. Regression of NDVI values between sensors 
3.2 Comparison of NDVI with Fractional Green Cover 
Experimental plot means of NDVI were compared with 
corresponding values of fractional green cover from photos for 
each of the sensors over a range of dates. Fig. 1 shows the 
means for NDVI and fractional green cover with each point in 
the graph representing an experimental plot mean value. The 
fractional green cover data were well described by NDVI using 
a linear model (Table 2). Pooling all of the data with sensor 
NDVI and corresponding fractional green cover (N=2355 plot 
means) resulted in an 1r? values of 0.70, and a standard error is 
0.12 (note fractional cover is unit less). Subsetting the data to 
include only the stem elongation phase of crop growth, or by 
considering individual dates, did not drastically alter the results 
(not shown). 
  
  
            
  
  
   
  
  
  
  
  
  
1 
9 08. 
Ó 
& ZW 
$ 06- A NUS 
Sr ain fy oe. 
S ! e ACS 210 
9 S CO " 8 | 
S 0.2 | KA eue) uem S40... 
be TRA A | à Greenseeker 
LJ pu L AAO 
0 D ; : 
0 0.2 0.4 0.6 0.8 1 
Sensor NDVI 
  
  
  
Figure 1. Fractional Green Cover (measured) plotted against 
the mean NDVI measured by plot, for all datasets 
(N=2355). The regression fitted yield r^ 2 0.70 and a 
standard error of fractional cover of 0.12. Note that 
bare soil plots are included (fractional cover of 0.0). 
  
  
  
  
  
Sensor N rr 
All 2355 0.70 | 
CC210 311 0.76 
Greenseeker 643 0.59 
CC470 1302 0.74 
  
  
  
  
  
Table 2. Regression results for fractional green cover on 
NDVI 
3.3 Comparison of NDVI with Above Ground Biomass 
Experimental plot means of NDVI and corresponding measured 
biomass were compared for the three sensors across sites and 
dates. Fig. 2 shows biomass (dry matter in kg ha) 
measurements and corresponding NDVI means for dates up 
until flowering, or Zadoks 60 (Zadoks et al. 1974) plotted by 
sensor. These dates are shown as this would be the maximum 
amount of green above ground biomass. The measured biomass 
was fitted to sensor NDVI using a non-linear regression model: 
Biomass = Bo + p ,*(p ; PV (1) 
where By, B ; and B , represent fitted parameters. The regression 
results are shown in Table 3. Overall, the sensor NDVI did not 
explain as much of the variance in biomass as it did for 
fractional green cover as indicated by the r” values. Pooling all 
of the data (N=1858) resulted in an r? value of 0.13 and 
standard error of 3459 kg ha'!. For two of the sensors (ACS 470 
and Greenseeker) the r improved when the data was subset to 
include dates up until flowering; the ACS 210 Horsham data 
was only acquired during this phase, so no comparison was 
available. The ACS 210 Horsham data yielded the highest r 
values for biomass fit to the sensor NDVI. One possible 
explanation for this result was the close proximity of the sensor 
measurements and the biomass sampling due to the small 
experimental plot size. 
As can be seen in Fig. 2, the relationship between sensor NDVI 
and biomass is more linear at low values of biomass. Linear 
regression was applied to the ACS 210 Horsham data for 
biomass values between 0 and 1000 kg ha” (inclusive), which 
resulted in an r? of 0.67 and a standard error of 115 kg ha” 
(Table 3.) 
  
  
    
  
  
  
  
   
     
  
  
   
  
  
  
   
   
   
     
    
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
    
   
  
  
  
  
    
Ini 
12! 
10 
Biomass, as Dry Matter kg ha-1 
o 
Sei 
| 
All S 
AC: 
Greer