ted by the
1e authors
irre would
y recently
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B8, 2012
XXII ISPRS Congress, 25 August - 01 September 2012, Melbourne, Australia
OPTIMIZATION OF DECISION-MAKING FOR SPATIAL SAMPLING IN THE NORTH
CHINA PLAIN, BASED ON REMOTE-SENSING A PRIORI KNOWLEDGE
Jianzhong Feng * ^ *, Linyan Bai", Shihong Liu**, Xiaolu Su“, Haiyan Hu c
? Key Laboratory of Agri-information Service Technology, Ministry of Agriculture, Beijing, 100081, China -
fengjzh4680@sina.com
b Center for Earth Observation and Digital Earth Chinese Academy of Sciences, Beijing, 100101, China -
lybai@ceode.ac.cn
? Institute of Agriculture Information, Chinese Academy of Agricultural Sciences, Beijing 100081, China —
fengjzh0338@sina.com
Commission VII, WG VIV4
KEY WORDS: Agricultural Spatial Sampling, Remote Sensing, a Priori Knowledge, Spatial Structure Characteristics,
RIP(s)/RIV(s), Sampling efficiency
ABSTRACT:
In this paper, the MODIS remote sensing data, featured with low-cost, high-timely and moderate/low spatial resolutions, in the North China
Plain (NCP) as a study region were firstly used to carry out mixed-pixel spectral decomposition to extract an useful regionalized indicator
parameter (RIP) (i.e., an available ratio, that is, fraction/percentage, of winter wheat planting area in each pixel as a regionalized indicator
variable (RIV) of spatial sampling) from the initial selected indicators. Then, the RIV values were spatially analyzed, and the spatial
structure characteristics (i.e., spatial correlation and variation) of the NCP were achieved, which were further processed to obtain the scale-
fiting, valid a priori knowledge or information of spatial sampling. Subsequently, founded upon an idea of rationally integrating
probability-based and model-based sampling techniques and effectively utilizing the obtained a priori knowledge or information, the spatial
sampling models and design schemes and their optimization and optimal selection were developed, as is a scientific basis of improving and
optimizing the existing spatial sampling schemes of large-scale cropland remote sensing monitoring. Additionally, by the adaptive analysis
and decision strategy the optimal local spatial prediction and gridded system of extrapolation results were able to excellently implement an
adaptive report pattern of spatial sampling in accordance with report-covering units in order to satisfy the actual needs of sampling surveys.
1. INTRODUCTION
For a long time, cropland area estimate using remote sensing
technique is an important research topic of accurately estimating
large-area crop yields by remote sensing approaches (Wang et al.,
2008). It is very important for macro-economic decision-making
departments of governments to timely know the related crop
production and make scientific and sound decisions, etc. Spatial
sampling technology is used to be able to effectively resolve this
problem that a balance is realized between the cropland area
accurate estimation and limited investigation budget (Li et al., 2004).
It is essential for spatial sampling that the basic principles and
methods of statistical sampling are applied to the regionalized
attributes of geographical objects. Therefore, there are the
probability spatial sampling techniques, based on traditional
statistics, (such as simple random, systematic, stratified, and cluster
spatial sampling) and model spatial sampling that are tied to spatial
variability theory (Stevens Jr. and Olsen, 2004; Li et al., 2004;
Dobbie and Henderson, 2008; Jia et al., 2008; Jiang et al., 2009).
Due to the complexity of geographical features in a region, the
traditional large-scale spatial sampling of resources or
environmental investigation (especially on agricultural monitoring
with remote sensing) are operated not well to use the a priori
knowledge or information of spatial structure characteristics of
some research region, but, relied on a certain sampling schema,
* Corresponding author.
more often to use common probability sampling models and
methods, such that they cannot meet actual requirements of spatial
sampling investigation because their sampling efficiencies are not
very high and ranges of sampling application are not large (Jiao et
al., 2002, 2006; Wu et al., 2004).
In this paper, founded upon an idea of rationally integrating
probability-based and model-based sampling techniques and
properly using the relevant sampling models and methods, the
valid spatial sampling design was developed in the North China
Plain (NCP) as a study region, which is associated with remote
sensing a priori knowledge or information, and the existing spatial
sampling schemes were suggested to improve and optimize in
better order to service the large-scale cropland remote sensing
monitoring in the NCP. In short, this study can further effectively
enhance the level of spatial sampling survey and decision-making
analysis of the related departments of national and local
governments.
2. METHODOLOGY
2.1 Models of a priori information acquisition
Geographical spatial structure characteristics are represented
with the spatially correlated and heterogeneous properties of
regionalized geographical elements (of surface things or
phenomena) (Feng, 2010).