International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
harvest crops in North China, Such as early rice, middle rice,
later rice in south China, spring wheat, soybean, single rice and
maize in North East China. But now this had change very much.
In south China, growing stage has change to 2 growing stage
for economic reason, this provide many convenience for data
acquisition.
4.5 Crop proportion monitoring precise assessment
In order to assess the precise of crop proportion monitoring
using remote sensing, two test-regions, Kaifeng and Taigu were
selected.
In Kaifeng region, one Landsat TM image was accepted on Apr.
1 in 2001 for summer-harvested crop proportion monitoring. In
Taigu region, Landsat TM image was accepted on Oct. 14 in
2003 for fall-harvested crop proportion monitoring.
At the same time, an IKONOS image was accepted in Kaifeng
region on Mar. 21 in 2001. After classification, we get another
crop proportion to compare to that from Landsat TM. In Taigu
region, we accepted a Quickbird imagery to draw a plant patch
map. By filling crop type or land-use type in every patch, we
get another crop proportion by summing up the total crop area
and divided with arable area extracted from the very high
spatial resolution images.
After the comparisons between crop proportion from Landsat
TM and from very high spatial resolution images, we can draw
conclusion that Landsat TM can be sure to monitor crop
proportion very well. Monitoring accuracy in Kaifeng region
was up to 99% and 97% in Taigu region. And this is acceptable
for the running of China Crop Watch System (Li Qiangzi, Wu
Bingfang 2004).
5. CROP TYPE PROPORTION SURVEY
China has a complicated topography and results in the small
patch of land in much area. This has given much difficult when
carrying out agricultural survey because of the low proportion
of arable land. At the same time, actual land use system in
China had pricked up the difficulty of agricultural survey,
because arable area are divided into small patches and every
patch is owned by personnel and is used at discretion.
Occasionally, a farm is dived into many thin patches and used
to plant several crops severally. In order to carry out
agricultural survey to estimate crop type proportion, sampling
methods must solve the following problems.
---- Sampling efficiency enough to complete the survey during
the limited time slot.
---- Sampling accuracy enough to complete unbiased estimate
crop type proportion
CCWS selected transect sampling frame to solve the problems.
Transect sampling frame is actually a two-step sampling
technique. During the first sampling step, CCWS use tessellated
plane sampling to build area frame and selected PSU randomly.
During the second sampling step, CCWS uses transect sampling
technique to select transect belt based on roads in PSU.
5.1 Transect Sampling Frame Design
For first step sampling, CCWS built its area sampling frame
randomly on the third level crop proportion stratification, and
the frame size set to 4 KM by 4 KM. And sample size
calculation use the sample size equation of simple random
sampling.
During sample calculation, crop type proportion p is important
for sample size. Here we use simple random sampling equation
to calculate sample size, and p is set to 0.5 to ensure enough
samples. For some strata, sample size calculated is larger than
5% sampling ratio, then used 5% sampling ratio to selec
samples. For all China, 3579 PSUs are selected for all 102 strata
with an average sampling ratio 1.82%.
For the second sampling step, we selected transect belt from
road segments in every PSU by experience. Given D is the
width of transect belt, sampling ratio n is,
LD
n=
(6)
Where: L is the transect belt length, K is the size of PSU. When
sample selecting, K and D is known, then L changes with
sampling ratio,
L= K
5 (7)
When selecting transect sampling belt, CCWS used 2%
sampling ratio to carry out crop type proportion survey. Here,
PSU size K = 4 km, belt width D = 0.1 km, sampling ratio n =
2%, then belt length L = 3.2 km, the total transect belt length is
11.5 thousands kilometre all China.
5.2 Ground survey and GVG survey system
Every year, ground survey is carried out based on transect
sampling design results. In order to ensure completing ground
survey in limited time, CCWS, divided the agricultural region
into 9 sampling regions and built special sampling team for
every one. At the same time, CCWS built a quality-control
mechanism by regulating sampling schedule and frequency for
every sampling region.
In order to improve survey efficiency, CCWS designed a patent
product, GVG surveying system to equip every survey team to
assist ground survey. GVG survey system is the integration of
GPS receiver, Video camera and GIS analysis system (Wu
Bingfang, 2000). Using video camera to snap pictures along the
transect lines and locate the geo-location of every picture by
GPS, then interpreting crop type proportion of every picture
inside and summing up for every stratum using GIS system.
5.3 Crop type proportion calculation
After ground survey, crop type proportion is interpreted for
every picture, and then accounted using equation (8).
i=l i=1,2,3..M (8)
Where: P; is the proportion of crop j, aj; is the proportion of crop
jin — i. A; is the total proportion of all crops in photo i. Nis
the total number of validate photos, M is the total crop types in
survey area.
Crop type proportion calculation include 3 levels, firstly it is
accounted on PSU level, and then account to strata level by
average. At last, crop type proportion on province level is
accounted using weighted average method, the weight here is
the area of every strata within the province.
5.4 Feasibility analysis of transect sampling based on road
segments
The main argumentation for the feasibility of transect sampling
frame is whether the road density is high enough to support the
sampling required. According to the data in 2002 from China
256
Internatio
Hue.
State Stat
administré
Shandong
cities exce
the 2% sa
less than
fact by ex
villages c
ground su
KM by 5
be used a:
2004).
Another a
structure 1
transect al
This phen
are more v
have not
mapped oi
keeping av
frames anc
kilometres
55 Crop
In order to
began to se
province ai
Shuangyan
in Heber p:
in 2002. FA
In every te
randomly t
help of G'
statistics, '
transect sai
every field
spatial reso
calculation
proportion.
value from
sampling te
TRE AM ED:
Test are
Dehui
Shuangy:
Luanche
Xinzho
Jiangnir
RE ORUM
Table 2 "Tre
We found tl
89.1% and
Winter whe
can be 98.9
test results s
transect sam
meet the run
