MODELLING TEA (Camellia (L) O. Kuntze) YIELD USING
SATELLITE DERIVED LAI, LANDUSE AND METEOROLOGICAL
DATA
Figure 1 - The relationship between NDVI and LAI for IRS-1C image
3.2. Model 2 - Model for estimating tea yield
Variable weights were assigned by finding the univariate relationship between each variable with yield. Weights for classes within each variable were assigned by using ARCVIEW analysis technique, which is yield summarized within the zones of reclass of each variable. Class weights were assigned by considering the yield of each tea field summarized within the zones of each variable. After assigning weights for all classes in available variables the relationship between yield and weights of classes were found using backward regression analysis. 5 models were obtained using backward regression analysis technique. Following model considered as the best model because, soil depth and rock cover percentage are very important factors for tea plant growth. Slope is also very important factor but Tea Research Institute of Sri Lanka is not recommending to plant tea in more than 55% slope lands in mid country region and more than 70% slope in up country region. Therefore, this factor is excluded.
Model 2
Yield = -603.923 +50.124wd - 23.5wr - 14.049wl + 65.845wi + 513.54wa
+39.54wh + 65.695wf + 46.338we
For Model 2-3 r 2 = 0.742; F = 1575.083
Where;
wd = Soil depth weight wr = Rockiness cover weight
wl = Landuse type weight wi = LAI weight
wa = Age of tea plantation weight wh = Relative humidity weight
wf = Rainfall weight we = Elevation weight
3.3. Model validation and application
Validate the model 1
Model developed for estimate LAI using IRS-1C image was validated using t
test by using holdout samples. The calculated t value (0.1367) is less than the tabulated t value (1.7247). Therefore the null hypothesized cannot be rejected and can be conclude that there is no significant difference between actual and measured LAI values. Therefore the model developed for estimate LAI by using satellite image derived NDVI values is valid with 95% significant level.
Validate the model 2 by using holdout samples and February, 2000 IRS-1C
image
The selected model 2 was validated using t test by using holdout samples. As the calculated t value (-0.197) is less than the tabulated value (1.745), the null hypothesis of no difference between actual and predicted yield cannot be rejected. Therefore actual yield does not differ significantly with predicted yield using the model developed for predict yield by using measured LAI values, other meteorological, agronomic parameters and is valid with 95% significant level.
After normalized March 1995 Landsat-TM image NDVI values of the tea fields were calculated. Model 2-B-3 was used to estimate 1995 yield. Estimated and actual tea yields were compared using t-test. The calculated t value (1.467) is less than the tabulated t value (1.669). Therefore the null hypothesized cannot be rejected and can be conclude that there is no significant difference between actual and estimated yields. Therefore this model can be used to estimate yield with 95% significant level. Actual tea yield for the whole estate is 1144 made tea kg ha -1 annum -1 for 1995.
Estimated tea yield for the whole estate for year 1995 is 1031 made tea kg ha -1
annum -1 .
4. Conclusion
Logarithmic function gives a good relationship between NDVI and LAI. This model can be adopted to estimate LAI of particular tea field using NDVI value of satellite images. The model develops by using weights of all variables give best correlation of determination. In this multivariate regression analysis all variables were considered. This method gives five models with backward regression and the model with eight variables selected as the best one out of these five as it gives 95% validity with holdout samples and could be able to estimate 1995 yield. Therefore this model (Model 2) is used for estimating tea yield.
5. Acknowledgement
The authors are grateful to Kahawatte plantation Ltd, Sri Lanka for giving permission to utilize Westhall estate for the study, Institute of surveying and mapping, Diyatalawa, Sri Lanka for providing a GPS system and Tea Research Institute of Sri Lanka for providing necessary facilities to carry out the study.
6. References
Cleves, J.G.P.W., Buker, C., Van Leeuwan, H.J.C., and Bouman, B.A.M.
(1994) A Framework for Monitoring Crop Growth by Combining Directional and Spectral Remote Sensing Information, Remote Sensing Enviroment, 50: 161-170
Gong, P., Ruiliang, Pu, and Miller, J.R. (1995) Coniferous Forest Leaf Area Index Estimation along the Oregon Transect using Compact Airborne Spectrographic Imager Data, PE&RS, 61-9:1107-1117
Myneni, R. B., Hall, F. G., Sellers, P. J., & Marshak, A. L. (1995). The interpretation of spectral vegetation indexes. IEEE Transactions on Geoscience and Remote Sensing, 33(2), 481-486
Selleres, P.J., 1985, Canopy
reflectance, photosynthesis and transpiration,
International Journal of Remote Sensing, 6,
1335-72
