So sample size calculation is a bit more tricky.
There are two things you need to consider:
1. **Technical** : What is the level of
accuracy you want to achieve.
Specifically for what variable you
are collecting. Different variables
will have different uncertainties in
the data collected as the
uncertainty is measured through the
number of times the value appears in
the study. Normally you will achieve
higher accuracy in the "main"
classes like the IPCC Land Use
Category (Forest, Settlement,
Cropland...) than in the sudvisions
(type of forest, type of cropland,
type of grasslands...). So if you need a high accuracy on data
of
Land Use Change, you will need to
collect a lot more plots that if you
just want to have a high level of
accuracy on the IPCC classes for the
current Land Use.
2. **Practical** : how
much time you have for the data
collection activity. Depending on
the level of complexity of the
survey, the internet speed and the
knowledge of the area you should be
able to collect between 100 (slow
internet, difficult survey) to 300
(fast internet, simplified survey)
plots per day. How many days and how
many people will be collecting data? Also, are there a lot of Google Earth very high resolution images in the AOI or will you need to use Landsat or Sentinel for assessing the plots (this takes longer)
This will make the decision easier.
Obviously, at the end you will get a
certain amount of plots and from
that you will be able to calculate
the uncertainties.
So there are two approaches to try to accomodate these aspects :
1. Building a multiple grid so that the
data collection can start from plots
at say 10x10 km and then if the
uncertainties is too large, refine
it with a grid one level down, in
this case 5x5 km. This grid has 4
times as many plots than the initial
one. Since you already collected the
plots at 10x10km (say 100 plots)
then it means that you will have to
collect the remaining ones (in this
example 300). If this data is still
not accurate enough then you move to
a 2.5x2.5 km grid and so on...
2. The previous approach can be a bit
complex to set up. A very similar
approach which is much easier to
design and has identical results is
to just use a random sampling
design. In this case you will
collect data in completely random
plots. Once you have collected
enough plots you may evaluate the
uncertainties. If you are happy with
that, you are done! If not, then you
need to keep collecting data. To
generate a random sampling design
you can use this Google Earth Engine
Script : [IMPROVED GRID GENERATOR][1]
[1]: https://code.earthengine.google.com/764eb6a6b5e7075d1faa051040f93031https://code.earthengine.google.com/c4a9f26e3b2242fca5571750531da1cc