reflectance was fixed. When LAI<2 the TSAVI noise was of equal magnitude to that when soil reflectance was 
fixed, but was significantly larger than that when the ALA was fixed. In contrast when LA1<2 the NDVI noise 
was larger than that for fixed soil reflectance and of equal magnitude to that for a fixed ALA. 
When VI is close to VI mu the relative equivalent noise approaches or exceeds one, and the signal is then 
obscured by the noise. Neither vegetation index can reliably and accurately predict the LAI of dense canopies. 
The range of accurately predictable LAI is further reduced when the canopy chlorosis distribution is unknown. 
3.2. The sensitivity of REN P to chlorosis distribution, ALA and soil reflectance 
Fig. 2a shows REN P of both indices for an ALA of 55°. NDVI was noisier than TSAVI for sparse canopies 
because it was strongly affected by the soil reflectance.The noise associated with TSAVI was always less than 
0.05. For canopies in which P>0.75 and the chlorosis distribution was known the noise associated with NDVI 
was less than that with TSAVI. The relative equivalent noise for NDVI had a similar magnitude for all values 
of ALA, but the magnitude of TSAVI relative equivalent noise decreased as the ALA increased (became more 
erectophile). 
Fig. 2b shows the trends of REN P vs. P were similar for both indices when the red soil reflectance was 
fixed-at 20%. The REN P of both indices was close to 0.1 when P<0.6 but decreased significantly at higher 
values. In dense canopies the noise associated with both indices was greater when the chlorosis distribution was 
unknown. In comparison with Fig. 2a the TSAVI noise was greater over the full range of P whilst the NDVI 
noise was lower when P<0.6 and greater when P>0.6. 
Fig. 2c shows the variation in REN P when neither ALA nor soil reflectance were known. Though NDVI- 
related noise was greater than 0.1 for sparse canopies (P<0.6), TSAVI-related noise was less than 0.1 at all 
canopy densities. The shapes of the curves were similar to those for known soil reflectance when P>0.6. But at 
low canopy densities the NDVI noise was considerably increased by uncertaintity in the soil reflectance. 
4. CONCLUSION 
Noise in the LAI and P signals were found to be related to uncontrolled variation in the ALA, the soil reflectance 
and the chlorosis distribution; their relative importance varied with the canopy density. At low LAI the greatest 
contribution to the noise in the signal came from uncontrolled variation in the soil reflectance. At high LAI the 
chlorosis distribution and the ALA contributed equally to the total noise level of both indices, though the effect 
was greater for NDVI than TSAVI. When either index was used to predict P the signal always exceeded the 
noise, but since the signal to noise ratio was always greater than 10:1 for the TSAVI it was a more reliable 
predictor of P than the NDVI. When TSAVI was used to predict P the noise only marginally increased when 
both the soil reflectance and the ALA were unknown, but when NDVI was used the noise increased 
considerably for sparse canopies if the soil reflectance was unknown. In dense canopies (P>0.85) there was a 
further marginal increase in the noise level of both indices when the chlorosis distribution was unknown. 
It is important to stress that this analysis was based on synthetic data generated with a canopy 
reflectance model. The results indicate the relative sensitivity of vegetation indices to chlorosis distribution, ALA, 
and soil reflectance; but real vegetation canopies do not always conform to the simple assumptions about canopy 
structure and optical properties made in the ABSAIL model. Important deviations from these assumptions 
include horizontal inhomogeneity, specular leaf reflectance, and finite sized leaves which cause the shadowing 
responsible for the hot spot effect. Though canopy inhomogeneity will obviously increase the importance of soil 
related noise at high LAI, the possible effects of specular reflectance and the hotspot phenomenon upon the 
signal-noise ratio of vegetation indices require further investigation with appropriate canopy reflectance models. 5 
5. ACKNOWLEDGEMENTS 
This research was carried out while in receipt of a NERC postgraduate studentship at the University of 
Nottingham (UK).