Overview and Brief History - Sampling Spatial Units for Agricultural Surveys - page 39

similar attributes than distant ones. Random units might be clustered in some areas
and missing from others, and may potentially miss spatial hotspots. In this case, the
choice of neighboring locations adds less additional information about the target
area. So, it is clear that sampling schemes for spatial units cannot be reasonably
defined unless we take spatial dependence into account (see Chap. 7 for a detailed
description of the key topic of spatial sampling).
The next step when designing sampling schemes for spatial units is to recognize
the existence of fundamental arbitrariness when aggregating areal data. This has
been referred to as the modifiable areal unit problem (MAUP, Openshaw 1977 ,
1984 ; Arbia 1989 ). This uncertainty seems to be particularly important because
areal units are not generally defined for the particular purpose of spatial data
analysis, but on the basis of some other exogenous criteria. This means that data
are available with respect to a given spatial configuration, whose structure is not
necessarily based on the underlying model of spatial variation.
Agricultural sampling methods can be based on list or spatial reference frames
(see Chap. 5 for more details). Designs based on a list frame are the most commonly
used sampling procedures for agricultural surveys. The list frame is produced by an
enumeration of elements of the population under investigation. In agricultural
surveys, this is often formed by holdings or holders addresses.
A spatial 1 sample survey is a study in which the final stage-sampling units are
land areas. The selection probabilities are generally proportional to their area, and
the land areas are generally denoted as segments. The segments of a spatial frame
can be areas (i.e., portion of territory), points, or transects (i.e., lines of a certain
length). The sampling units should not overlap, and must cover the entire survey
area under investigation.
List and spatial frames have advantages and disadvantages. In particular, list
frame surveys are cheaper, because the sampled farms provide a large amount of
information on crop area and yields, livestock, inputs, and socio-economic vari-
ables, and use only one interview.
Spatial reference frame samples (Cotter et al. 2010 ) are better protected against
non-sampling errors that are caused by the frame having missing or overlapping
units. They do not exhibit problems linked to gaps (i.e., bias of the estimates). It is
also possible to use a spatial frame to update and verify the rate of coverage of
existing archives on farms. Furthermore, the researcher can use auxiliary informa-
tion (e.g., remote sensing), can issue timely and higher precision estimates on
cultivated areas and expected production, and can reduce the burden on farmers.
Finally, spatial sampling displays longevity of the frame (only updates for land use
changes are necessary), is versatile (multiple variables can be considered in one
survey), is objective (land cover/land use and measures of areas are directly
observed by surveyors in the field), and has a low non-response rate (only for
unreachable areas). The major disadvantages are the high cost of setting up the
1 In literature, this frame is commonly called an area frame. We consider this expression to be
misleading, and in this topic we prefer the terms spatial reference or geo-coded frames.