Cultivo de arveja 1. Días a la floración

In order to determine the Forestry Commission sub-compartments in which to locate a field plot site, a basic remote sensing analysis was carried out in order to delineate areas which corresponded to a range of different potential forest structural types. The authors Kalacska et al. (2004) and Manes et al (2010) state that Normalised Difference Vegetation Index (NDVI) values and their variability correlate with different forest canopy structural types. Thus, field plot locations were selected across a range of NDVI values.

An NDVI image was calculated from airborne CASI-2 multispectral imagery acquired in 2007 for the New Forest study site in ENVI 4.7 (ITT Visual Information Solutions) image analysis software. The details of this dataset are presented in section 3.2.3. All non-forest areas were masked and removed based on an NDVI threshold discriminating forest and non- forest. The NDVI image was then subset to produce images of deciduous or coniferous areas based on Forestry Commission mapping. This allowed an equal number of sample sites for deciduous and coniferous woodland types to be identified, thereby reducing potential bias towards one broad structural type.

Figure 3.5 – Field plot locations. Blue points represent those plots surveyed in 2010 whilst red indicate the plots surveyed in 2012. Base Map layer is © Crown Copyright/database right 2010. An Ordnance Survey/EDINA supplied service.

Figure 3.5 – Field plot locations. Blue points represent those plots surveyed in 2010 whilst red indicate the plots surveyed in 2012. Base Map layer is © Crown Copyright/database right 2010. An Ordnance Survey/EDINA supplied service.

Figure 3.5 – Field plot locations. Blue points represent those plots surveyed in 2010 whilst red indicate the plots surveyed in 2012. Base Map layer is © Crown Copyright/database right 2010. An Ordnance Survey/EDINA supplied service.

Figure 3.6 – A comparison of field plot summary statistics between four sample plot sizes.

Eight NDVI classes were created manually for the two images. The classes covered an equal NDVI range which occupied the values in-between 0.4 to 0.8 (i.e. the range of values for green vegetation in the imagery). A stratified random sampling technique was then applied to both of the classified images where an equal number of sample points were produced for each class-strata. An arbitrary minimum distance was enforced between sample points (100m) in addition to a minimum distance from the perimeter, or of no-data value (100m). The Forestry Commission sub-compartments in which a sample point was located were identified and visited within the field. Attempts were then made to locate a position as close as possible to the sample point coordinates in which to establish a field plot.

The 21 training data plots were visited during the months of June to September 2010, whilst an additional 20 plots were enumerated to provide validation data during June to October in 2012.Plot locations are presented in Figure 3.5.The plot size was established as 30m x 30m, with a 10m x 10m subplot located in the south-west corner, illustrated in Figure 3.7. To locate the extents of each of the field plots accurately, in order to best line up the data with remote sensing datasets, the Forestry Commission sub-compartment was located in the field, and the approximate coordinates identified from the generated sample point. A differential GPS (dGPS) system, the Leica GPS 500 (Leica Geosystems, part of Hexagon Group), was set up in a suitable clearing or outside of the forest in order to ensure signal errors caused by the forest canopy was reduced (Campbell and Wynne, 2011). The dGPS unit was employed to determine accurate reference points as benchmarks for locating a Sokkia 6F total station

Figure 3.6 – A comparison of field plot summary statistics between four sample plot sizes.

Eight NDVI classes were created manually for the two images. The classes covered an equal NDVI range which occupied the values in-between 0.4 to 0.8 (i.e. the range of values for green vegetation in the imagery). A stratified random sampling technique was then applied to both of the classified images where an equal number of sample points were produced for each class-strata. An arbitrary minimum distance was enforced between sample points (100m) in addition to a minimum distance from the perimeter, or of no-data value (100m). The Forestry Commission sub-compartments in which a sample point was located were identified and visited within the field. Attempts were then made to locate a position as close as possible to the sample point coordinates in which to establish a field plot.

The 21 training data plots were visited during the months of June to September 2010, whilst an additional 20 plots were enumerated to provide validation data during June to October in 2012.Plot locations are presented in Figure 3.5.The plot size was established as 30m x 30m, with a 10m x 10m subplot located in the south-west corner, illustrated in Figure 3.7. To locate the extents of each of the field plots accurately, in order to best line up the data with remote sensing datasets, the Forestry Commission sub-compartment was located in the field, and the approximate coordinates identified from the generated sample point. A differential GPS (dGPS) system, the Leica GPS 500 (Leica Geosystems, part of Hexagon Group), was set up in a suitable clearing or outside of the forest in order to ensure signal errors caused by the forest canopy was reduced (Campbell and Wynne, 2011). The dGPS unit was employed to determine accurate reference points as benchmarks for locating a Sokkia 6F total station

Figure 3.6 – A comparison of field plot summary statistics between four sample plot sizes.

Eight NDVI classes were created manually for the two images. The classes covered an equal NDVI range which occupied the values in-between 0.4 to 0.8 (i.e. the range of values for green vegetation in the imagery). A stratified random sampling technique was then applied to both of the classified images where an equal number of sample points were produced for each class-strata. An arbitrary minimum distance was enforced between sample points (100m) in addition to a minimum distance from the perimeter, or of no-data value (100m). The Forestry Commission sub-compartments in which a sample point was located were identified and visited within the field. Attempts were then made to locate a position as close as possible to the sample point coordinates in which to establish a field plot.

The 21 training data plots were visited during the months of June to September 2010, whilst an additional 20 plots were enumerated to provide validation data during June to October in 2012.Plot locations are presented in Figure 3.5.The plot size was established as 30m x 30m, with a 10m x 10m subplot located in the south-west corner, illustrated in Figure 3.7. To locate the extents of each of the field plots accurately, in order to best line up the data with remote sensing datasets, the Forestry Commission sub-compartment was located in the field, and the approximate coordinates identified from the generated sample point. A differential GPS (dGPS) system, the Leica GPS 500 (Leica Geosystems, part of Hexagon Group), was set up in a suitable clearing or outside of the forest in order to ensure signal errors caused by the forest canopy was reduced (Campbell and Wynne, 2011). The dGPS unit was employed to determine accurate reference points as benchmarks for locating a Sokkia 6F total station

(SOKKIA TOPCON Co. Ltd.). The total station was used to traverse from these reference points to each of the plot corners beneath the canopy, recording each in British National Grid (BNG) coordinates.

Post-processing position corrections were applied to the dGPS coordinate information using UK Ordnance Survey RINEX data using the Leica Geo Office software. Overall this provided a positional accuracy for dGPS positions of ≤0.03m overall horizontal accuracy. The appropriate coordinate corrections were then applied to total station measurements. When conducting the surveying work with the total stations, reference points were set up along the traverse, and sighted before and after the total station was relocated to provide a measure of accuracy. Through back-sighting the horizontal error was calculated as ≤8.11cm overall.