Evidence – Chlorophyll Content

The NDVI (plant stress) imagery generated by the AgriDrone system is extremely accurate. In a recent series of experiments we sampled chlorophyll content measurements from a Yucca Pallida plant and compared them with the NDVI values generated by our system. We averaged a correlation coefficient of above 95% for four tests. We also fitted linear regressions to the data with adjusted R-squared figures between 90 and 95. The significance of a linear relationship between the data sets also simplifies calibration of measured NDVI back to estimated clorophyll content measure.

Below we’ve included some data and  imagery from one of the tests, using 10 data points. The first table below shows collected chloropyll content measurements on the left side and NDVI measurements from our imagery on the right side. The correlation coefficient of 97.29 is calculated for the averages of the two sets of five samples.

 

Clorophyl Content Measure NDVI Scaled Value ex Floating Point Bitmap
No Sample 1 Sample 2 Sample 3 Sample 4 Sample 5 5 Sample Avg CC Value 5 Sample Avg NDVI Value Sample 1 Sample 2 Sample 3 Sample 4 Sample 5
1 39.5 41.2 41.3 40.7 39.6 40.46 65.4 67 65 65 65 65
2 36.8 36.6 36.4 36.3 36.5 36.52 52.8 56 53 52 53 50
3 35.5 34 35 35 35 34.9 55.8 54 55 52 58 60
4 34.9 34.8 34.9 35.3 35.2 35.02 60.2 53 58 65 66 59
5 36.7 37.4 37.3 37.6 37.8 37.36 62.4 62 63 62 62 63
6 39.7 40.2 40.8 39.4 40.3 40.08 65.4 66 59 63 70 69
7 40.5 41.8 43.2 43.9 44.3 42.74 75.6 68 78 75 84 73
8 41.1 43.5 43.7 43.1 41 42.48 72.6 71 74 75 73 70
9 43.6 44 43.4 42.4 41.1 42.9 77 77 78 79 77 74
10 62.9 63.8 72.3 64.5 57 64.1 108.4 96 111 111 105 119
Correl Coeff 0.972953361

 

Next we performed OLS regressions of NDVI on CC measurement. We chose this form so that we could infer NDVI from a CC measure for any given sample, and thereby calibrate our NDVI imagery more accurately.

This regression shows significance for the regression as a whole and the estimated coefficient. The adjusted R squared of 94 implies a good predictive capability of the relationship.

OLS PREDICT NDVI
Regression Statistics
Multiple R 0.972953361
R Square 0.946638242
Adjusted R Square 0.939968022
Standard Error 3.876893042
Observations 10
ANOVA
df SS MS F Significance F
Regression 1 2133.101603 2133.102 141.9201 2.26603E-06
Residual 8 120.2423973 15.0303
Total 9 2253.344
Coefficients Standard Error t Stat P-value Lower 95% Upper 95%
Intercept -6.160593035 6.473274702 -0.9517 0.369115 -21.08799125 8.766805183
CC Value 1.817759579 0.152585931 11.91302 2.27E-06 1.465895793 2.169623366

 

The plot below shows the predicted relationship versus the scatter plot of observations. Visually, the relationship is strong:

Scatterplot

 

Next we show the imagery associated with this test, starting with the false-colour IR image and associated sample points:

Points - NRG

 

 

The floating point NDVI image with the same sample points shown:

Points - FloatNDVI

 

 

 

 

 

 

 

 

 

 

And the NDVI applied to our LUT:

Points - NDVI LUT

 

Conclusion

This set of experiments have established that our remote sensor system is capable of representing plant health extremely well. The high correlation coefficients and R squared statistics from significant regressions implies that the sensor is equivalent to an airborne clorophyll content meter. Provided that appropriate controls are applied for varying light conditions, comparable time series datasets can be obtained using this system. For highly sensitive calculations – such as when combining data from successive days or weeks – it would be important to calibrate NDVI measurements from different days using an invariant measure such as chlorophyll content and calibrating the NDVI values for each sample period.