How this dataset was created
A. A series of 31
satellite images was acquired between February and October 2017. These were
from Sentinel-2A and -2B and Landsat-8. Each image was cloud-masked and then
classified separately into land cover classes.
B. A
rule-set was used to draw together evidence from the time-series of 31
per-pixel land cover classifications into a single per-pixel land use map
representing winter 2017.The term ‘land use’ pertains primarily to the
agricultural classes that are the main focus of this work. The map also includes
classes outside the agricultural area which are not formally land use classes, but are simply the
dominant land cover class at that pixel through time. The classes are:
Agricultural classes
– winter forage
WF kale
WF other brassica and swede
WF fodder beet
WF cereal
Unknown/other winter grazing – Intensive winter
grazing on pasture or other unidentified winter crop
Agricultural classes
– non-forage
Agricultural bare soil – Bare for an extended period
April to July
Crop residue, dead vegetation – Crop residue or dead
vegetation for an extended period April to July
Autumn-planted crop or pasture – Includes arable
crops and pasture renewal
Pasture – improved
Pasture – unimproved/poor quality/low cover –
Generally grazed, though there is a meeting point between this class and the
non-agricultural (mostly ungrazed) class that includes tussockland.
Unknown agriculture – Including mixed classes,
scrubby pasture, narrow shelterbelts within a pasture/crop pixel
Non-agricultural
classes
Forest and scrub – Both native and exotic, including
forestry
Tussockland, herbfield, alpine shrubland –
Essentially ungrazed, though there is a meeting point between this class and
the agricultural unimproved pasture class.
Water – Lakes, rivers, estuaries
Snow, ice
Rock, scree, gravel, sand
Hill country/non-agricultural bare soil – For
example, land slips. Can be an overlap with the rock/gravel class in silty
scree margins
Deep shadow – Insufficient reflection of light to
determine land cover class – generally in steep areas
Urban, industrial
C. A
GIS-format paddock boundary map (polygons) was derived from a selected subset
of the time-series satellite images. The paddock boundary map covers only those
areas considered to be ‘agricultural’ (rather than the whole Southland region,
which also includes large alpine and forested areas). This polygon layer was
imposed on the per-pixel land use map described in B above. A rule set was used
to decide on a final land use classification for each paddock polygon, based on
the land use classes of the set of pixels within it. The final output uses the
same set of classes as listed in B.
Created for Environment Southland by Landcare Research - Heather North, Stella Bellis & David Pairman, May 2018
