MODELLING TEA (Camellia (L) O. Kuntze) YIELD USING
SATELLITE DERIVED LAI, LANDUSE AND METEOROLOGICAL
DATA
hours and average daily temperature maps for the estate were prepared. Tea plantation parameters concerning acreage, variety (clone), planting density, soil cover by the canopy, pruning cycle, planting date and yield were collected. These data were collected from tea estate records.
Each and every tea field leaf samples were collected for measuring LAI. Sampling was done in 50 fields of the estate. Sample size was 1 square meter. Ten random sampling points per field were selected from each field. Leaf area was measured by using Decagon pseudocolor Ag vision image analyzer. Following equation was used to calculate LAI by point-by-point basis:
LAI = Leaf Area / Sample surface Area
However, when considering satellite images actual coverage of tea plant canopy has to be considered. Therefore measured leaf area readings and fields by field plant density records were used to calculate LAI by using following formula. Leaf area and sample ground surface area should be in same units.
LAI = Leaf Area / (Sample Surface Area X Actual Coverage)
Actual Coverage = Actual Plant Density / Optimum Plant Density
Where; For Vegetatively
propagated tea with 0.6mx1.2m spacing; Optimum tea
plant density = 13800 plants
per
ha For seedling tea with 0.9mx1.2m spacing;
Optimum tea plant density =
9250 plants per
ha Sample surface area =
1m 2
2.3. Satellite data
Radiometric normalization was done to remove inherent noise in multi-temporal LANDSAT-TM images. Linear transformation was done for radiometric correction. The IRS-1C LISS III (Linear Imaging Self Scanner) acquired on 12 th February 2000 and LANDSAT-TM image acquired on 6 th March 1995 were used in the study. After performing geometric correction, radiometric normalization and masking for assigning null value to non vegetation areas of images NDVI values were calculated by using field boundary vector file as ROI (region of interest).
Non-vegetation areas include water, buildings and roads etc. Masking was done by detecting all NDVI responses less than zero and assigning the pixels a null value. NDVI values were calculated with formula expressed below;
NDVI = (NIR - VR) / (NIR + VR)
NDVI is a measure of chlorophyll abundance and energy absorption (Mynen iet. al., 1995). Band 4 of LANDSAT-TM and band 3 of IRS-1C receives the maximum
of the chlorophyll reflectance, while band 3 of Landsat-TM and band 2 of IRS-1C is the chlorophyll absorption band. Vector layer of the estate field boundary map was overlain and used as regions of interest to get statistics of mean NDVI value for each field.
2.4. Model 1- Model for estimating LAI using satellite derived NDVI
NDVI values derived from IRS-1C satellite image were used for this model development. Out of Landsat-TM and IRS-IC only IRS-1C satellite image was used for this as it was acquired on one of the sampling date. Several univariate linear and nonlinear prediction models were used. Gong, P. et. al.(1995) also tried several functions for develop a model for estimate coniferous forest LAI using NDVI and conclude a hyperbolic is the best model for estimating LAI using NDVI.
2.5. Model 2 - Model for estimating tea yield
All digital coverages were used for assessing modeling parameters specific to tea plants. In order to estimate LAI from all existing variables multivariate relationships between tea yield and other variables were investigated. These variables were slope, aspect, elevation, age of the tea plantation, type of tea, relative humidity, annual cumulative rain fall, average daily temperature, average daily sunshine hours, soil depth, rock cover percentage of the field and leaf area index. However, average daily temperature and daily sunshine hours parameters were not considered for the model development as temperature and daily sunshine hours do not show any variations in field-by-field basis of the estate. To find the parameter weights univariate relationship between LAI and all existing variables were investigated. Using these correlations weights were assigned to each variable and empirical relationship between average yield and other parameter weights was investigated.
3. Results and Discussion
3.1. Model 1- Model for estimating LAI using satellite derived NDVI
The
spectral data corresponding to 40 LAI measured
fields were extracted from the IRS-1C February
2000 image. NDVI were calculated for those images.
Several linear and non-linear models were tried to
the relationship between LAI and NDVI derived from
IRS-1C satellite image. Out of all a logarithmic
function considers as the best goodness of fit by
considering all cases for estimate LAI using
satellite images. Models adopted to find the
relationship between NDVI and LAI for satellite
image presents in Figure 1.
