Research Papers

Improving sampling efficiency of crop acreage estimation using wheat planting rule from historical remote sensing

[+] Author Affiliations
Jinshui Zhang

Beijing Normal University, State Key Laboratory of Earth Surface Processes and Resource Ecology, Beijing 100875, China

Beijing Normal University, College of Resources Science and Technology, Beijing 100875, China

Shuang Zhu

Beijing Normal University, State Key Laboratory of Earth Surface Processes and Resource Ecology, Beijing 100875, China

Beijing Normal University, College of Resources Science and Technology, Beijing 100875, China

Xiufang Zhu

Beijing Normal University, State Key Laboratory of Earth Surface Processes and Resource Ecology, Beijing 100875, China

Beijing Normal University, College of Resources Science and Technology, Beijing 100875, China

Guanyuan Shuai

Beijing Normal University, State Key Laboratory of Earth Surface Processes and Resource Ecology, Beijing 100875, China

Beijing Normal University, College of Resources Science and Technology, Beijing 100875, China

Dengfeng Xie

Beijing Normal University, State Key Laboratory of Earth Surface Processes and Resource Ecology, Beijing 100875, China

Beijing Normal University, College of Resources Science and Technology, Beijing 100875, China

J. Appl. Remote Sens. 8(1), 083663 (Mar 17, 2014). doi:10.1117/1.JRS.8.083663
History: Received June 16, 2013; Revised February 11, 2014; Accepted February 11, 2014
Text Size: A A A

Open Access Open Access

Abstract.  The technology of remote sensing combining sampling is an effective way to estimate crop acreage (CA) at large scale. Previous research proved that if the crop proportion within a sampling unit is sufficiently stable from year to year, pixels classified from historical remote sensing images could offer reliable regression estimators for current CA. However, previous works explored various approaches for CA estimation using one-year historical data, which makes it difficult to determine which year has the highest correlation to the targeted year, especially for a region where background information about the cultivated planting system is scarce. We estimated the winter wheat acreage of Beijing in 2009 by using two stratification variables including the coefficient of variance and the mean CA of the sampling units via a two-stage stratified sampling method with multiple historical remotely sensed data. Results show that: (1) our method has higher sampling average accuracy and lower standard errors of sample averages than simple random sampling or the one-stage stratification method with CA as the auxiliary variable, which are the usual methods employed in the previous studies. (2) Fewer samples are required to get the predefined accuracy, which reduces costs. (3) Generally, using mean CA derived from multiyear historical remotely sensed data as the auxiliary variable has a higher accuracy than those data using CA derived from one-year historical remote sensing images as the auxiliary variable in one-stage sampling.

Figures in this Article

Reliable and timely assessment of crop planting acreage is critical for ensuring a nation’s food security and deciding upon import/export policies, as well as managing food price optimization. Considering the advantages of remote sensing for wide-range and short-duration detection of land cover characteristics, information regarding the distribution of different crop types could be derived from remote sensing images by a series of classification methods.1,2 Multiplying the number of pixels identified as one specific crop by the area represented by each pixel provides a figure for the total area of a crop in that region. However, the accuracy of crop acreage (CA) estimation by pixel counting method cannot be 100% guaranteed due to potential errors either made during the initial classification or from omission or relevant pixels as well as mixed pixels,3,4 where only part of the pixels contains the desired information. Therefore, several methods were developed for improving CA estimation accuracy;5,6 these include regression, calibration, and small-area estimators, which combine classification maps derived from remote sensing images with ground-truth data. For example, National Agricultural Statistics Service periodically classifies the AWiFS imagery for the Crop Data Layer, and then regional crop planting area is calculated followed by a survey of ground samples in June survey statistics.7

Remote sensing images play an important role throughout the entire sampling design in order to improve both estimation accuracy and surveying efficiency.3,5,814 At the design stage of sampling for agricultural statistics, remote sensing images are usually used for the definition and stratification of sampling units. Remote sensing for stratification of an area’s sampling frame has been widely used in many countries such as the US and within the EU.15,16 With the stratified boundaries derived from stratification on the remote sensing imagery, a wall-to-wall frame is then divided into final sampling units, called segments.17 Land cover distribution derived from remote sensing images offers an efficient and low-cost method of conducting this stratification. For example, Corine Land Cover, a harmonized land cover map for nearly all countries of central and Western Europe, has been tested as a basis for stratification in Spain with satisfactory results.1719

Relative efficiency is a specific indicator for demonstrating the significance of the effectiveness of a stratification indicator constructed using remote sensing. For dominant crop types in a region, the relative sampling efficiency was around 3, which benefit from the crop identifications made from the remote sensing classifications as the basis for stratification indicator to improve sampling efficiency. Thus, the classified imagery serves as a proxy for representing the spatial pattern of land cover, which can be used as an effective auxiliary to improve sampling efficiency for CA estimation.8

Despite the extensive applications for remote sensing in ecological resource detection,20 an inevitable-limiting condition in its use for agricultural detection is the frequent occurrence of rapid, short-duration changes in land cover distribution.2 Remote sensing images are difficult to acquire during growing seasons due to the impact of the long-revisit period or adverse atmospheric conditions. Besides, obtaining images acquired before the sowing period may be impossible, and even if sufficient appropriate images could be acquired in the mid- or later-stage of the growing season, the time required to preprocess and classify the images for updating the sampling frame would render it difficult to ensure timely crop detection at a large scale. As a result, some researchers have proposed historical regression estimators constructed from land cover data obtained from remotely sensed images to improve the cost efficiency of CA estimation.10,21,22 Previous studies have shown that if the proportion of land devoted to a specific crop is sufficiently stable or if the correlation between different crop types is relative high from year to year, then historical data could offer reliable regression estimators for CA. Therefore, taking full advantage of historical images for agricultural statistics could provide an estimated result quickly through a process that integrates remote sensing and ground surveys, with the added benefit of decreased expenses for image acquisition and processing.10 However, the researches only used 1 year’s historical data for CA estimation, with the assumption that crop proportion is relatively stable and therefore represents a strong correlation between estimators and targets. Obviously, this assumption is limited to regions with different planting patterns.10,21 Moreover, it is difficult to determine which year has the highest correlation with the target year, especially for an area being investigated without sufficient background information regarding the planting system.

According to previous studies, multiyear crop distributions could effectively reflect the crop planting patterns. Therefore, this article used two stratification variables, the coefficient of variance of crop acreage (CVCA) and the mean CA of the sampling units, to estimate CA via a two-stage stratified sampling (TSS) method with multiple historical remotely sensed data. These two variables will indicate both fluctuation in and average level of CA among different years within parcels, respectively.

To test the performance of the two stratification variables, we used winter wheat as our target, selected the main planting region in Beijing as the study area and estimated the wheat acreage for 2009 using a TSS method with those two auxiliary variables. The rest of this article is organized as follows. Section 2 describes the study area and materials. The proposed method is presented in detail in Sec. 3. Experimental results are given in Sec. 4, followed by discussions in Sec. 5, and conclusions in Sec. 6.

Study Area

Our study area is located in Beijing, China, which is situated on the North China Plain. Urban agriculture plays an important role in both food supply and ecological protection. High-crop planting cover could decrease the risk of dust storms and protect the urban eco-environment. The Beijing government has paid much attention on the crop situation in the city, and the agricultural subsidiary policy was proposed to focus on the agricultural economy’s development. During spring in Beijing, bare cropland directly results in dust blowing. As the only winter crop, winter wheat could effectively reduce the amount of dust generated from bare cropland. Thus, accurate and precise detection of winter wheat planting areas is particularly important. The four counties chosen as our study area, namely Fangshan, Shunyi, Daxing, and Tongzhou, which cover about 4.98×103km2 (Fig. 1), are the main wheat planting regions, accounting for more than 90% of the wheat planting area in Beijing.

Data Sets

The data used in this study mainly include two elements: four cloud-free Landsat images (acquired on April 7, 2006, December 3, 2006, March 27, 2008, and April 15, 2009) and wall-to-wall aerial orthophotograph images with a fine spatial resolution (0.4 m) obtained in September 2004 in Beijing. Coregistration between the Landsat image acquired on April 7, 2006 and the aerial orthophotograph images were carried out first, yielded a mis-registration error smaller than 0.5 thematic mapper (TM) pixels. After that, the remaining three Landsat images were coregistered to the initial geo-corrected Landsat image. Landsat images were used for mapping historical wheat distribution from 2006 to 2009. Aerial orthophotograph images were also used to construct the area’s sampling frame with artificial photointerpretation method.

The technology flowchart of our study is shown below, which includes three main steps (Fig. 2). First, wheat thematic maps are derived from multihistorical Landsat images by support vector machine (SVM) classification method; second, wheat acreage estimation is obtained by a TSS method employing our proposed auxiliary variables; finally, comparison experiments in which the conventional sampling methods of simple random sampling (SRS) and one-stage stratified sampling are used.

Graphic Jump LocationF2 :

Flowchart of the proposed method.

Classification for Wheat Distribution Mapping

The SVMs is a nonparametric-supervised classification method reported to have great advantages over other classifiers, such as maximum likelihood classification,20,2325 which has been widely used to classify different land cover types from remote sensing images.

In this study, five different land cover schemes were defined: winter wheat, bare land, forest, water, and urban. Winter wheat as the sole winter crop throughout the study area could be extracted from the remote sensing easily. Training and validation samples for each land cover type were all derived via visual interpretation with help of the high-resolution orthophotograph image. Four land cover maps were generated by using SVM and each land cover map was further converted into a binary map (the one indicates wheat, and zero indicates nonwheat).

Here, we must emphasize that the wheat distributions from images obtained on April 7, 2006, and December 3, 2006, depict wheat harvested in different years. The wheat distribution from the image obtained on April 7, 2006, is for wheat harvested in 2006, whereas the wheat distribution from the image obtained on December 3, 2006, is for the year 2007; this potential for confusion arises from the fact that in Beijing, wheat is sown in October and harvested in the following June. Therefore, our wheat thematic maps were derived for wheat growing years 2006, 2007, 2008, and 2009.

Classification accuracy assessments were carried out using visual interpretations, and the producer accuracy and user accuracy of the wheat thematic maps derived from four Landsat images are reported in Table 1. The overall accuracy of each wheat thematic map is satisfactory.26 We used the first 3 years’ maps (2006, 2007, and 2008) to define the auxiliary variables and estimated the wheat acreage of 2009 by a TSS method, which will be introduced later (Sec. 3.3). The wheat acreage calculated from our wheat thematic map covering 2009 was taken as the true value in order to assess the accuracy and efficiency of our sampling method, regardless of any omission or commission errors because of two reasons. The producer’s accuracy (91.7%) in 2009 is the second highest among the four study years. The user’s accuracy (89.9%) in 2009 is the highest among the four study years. The producer’s accuracy refers omission errors, while the user’s accuracy refers to commission errors (Table 1). Omission and commission errors offset each other. Therefore, it is reasonable to make conclusion that the wheat acreage calculated from wheat thematic map in 2009 is quite closed to the actual wheat acreage in 2009, which is suitable to be used as the actual estimator to validate the proposed method. The similar method has been used in the Broich’s study to assess the sampling efficiency to estimate deforestation and without considering the system error of the testing data.27

Table Grahic Jump Location
Table 1Accuracy assessment of the historical images classification.
Area Sampling Frames Construction

A sampling frame defined by a geographical space, in which the units could be points, transects or area units, often termed segments.3 Satellite images can be used to determine segments with physical boundaries.8,17 Aerial photography-produced remote sensing images are the perfect data source for area frame construction. In our study, the area-sampling frame, including 11,540 cultivated parcel populations almost 24 ha for each segment, was constructed by visual interpretation of aerial photography covering all the arable land with physical boundaries defined by roads, water, and so on.

Sampling Design

Stratified sampling methods have been widely used in previous studies for their higher accuracy and efficiency of estimation comparing with other sampling method.12,2730 A TSS method (double stratification), which has been evaluated in previous studies,31 was employed in our research in order to improve sample efficiency. The auxiliary variable used in the first stage is the CVCA with multihistorical years [Eqs. (1)–(8)]. The auxiliary variable used in the second stage is CA. The coefficient of variance, also known as “relative variability,” is defined as the ratio of the standard deviation to the mean value, which is a useful statistical index for defining the degree of variation and used as an effective measurement for evaluating the accuracy of unbiased acreage estimations in previous studies.12,32 In this article, CVCA, an innovative indicator used to represent the fluctuation of CA across multihistorical years for each sampling unit, is adopted for first-stage stratification in the TSS method Display Formula

s¯=i=1Tsi/t,(1)
Display Formula
sd=i=1T(SiS¯)2/T,(2)
Display Formula
CVCA=sd/S¯,(3)
where T is the number of historical years. Here, that number is three, i.e., 2006, 2007, and 2008. Si is the CA within each sampling unit in i year. s¯ and sd are the mean and standard deviation of the CA of a given sampling unit within T year, respectively.

An effective stratification method is the basis for improving sampling efficiency.3 The partition of the population of the area frame (R) in the TSS method can be presented as Display Formula

R=R1R2RiRM(i=1,2M),(4)
Display Formula
Ri=Ri1Ri2RijRiN(j=1,2N).(5)

R was subdivided into M nonoverlapping strata (Ri, i=1,i,M) in terms of the auxiliary variable CVCA. The stratum Ri was further subdivided into N strata by the auxiliary variable CA. Therefore, Rij is j’th of the second stage stratification within i’th stratum of the first-stage stratification. An increase in the number of strata involves extra work in planning and drawing the sample and increases the number of weights used in computation. Cochran33 analyzed that an increase in strata beyond six necessitates any substantial decrease of the sample number in order to keep the cost constant, and the increase will seldom be profitable. Therefore, in this study, the number of strata was set to six for both first- and second-stage stratification, i.e., M=N=6. Each stratum’s boundaries were determined by the Dalenius–Hodges rule, which is the cumulative of the square root of the frequency method.33,34

Sample Selection

The samples were selected in each stratum using a random sampling method. The sample size in each stratum is subject to a proportional allocation rule. In other words, the sample size nh is proportionate to the population size Nh in h stratum, which is given by Display Formula

f=nh/Nh=n/N,(6)
where N is the total number of sampling population, n is the total number of samples, is Ni the total number of samples for stratum h, and f is the sampling ratio for samples selection. In order to test the sampling variation of different methods, different sampling rates ranging from 1% to 10% with a fixed interval of 1% were implemented for CA estimations in our study.

Acreage Estimation

Here, we used a simple ratio estimator for evaluating the feasibility of the TSS method.12 The formulas used to estimate total wheat acreage are given below: Display Formula

Y^tRC=R^icXi=y¯ist/x¯istXi,(7)
Display Formula
Y^RC=i=1MY^iRC,(8)
where Y^tRC is estimated acreage of wheat in stratum i, and Y^RC is the estimated acreage of wheat in the whole study area. R^ic is the ratio between the true value and auxiliary value for stratum i. Xi is the wheat acreage of the classification derived from the remote sensing image in stratum i. y¯ist and x¯ist are the estimated mean value of the target and auxiliary variable for the stratum i, respectively. M is the total number of strata.

Comparison Experiments

In order to assess the performance of our proposed method, we estimated the wheat acreage of our study area in 2009 via two sampling methods: one using SRS, without any auxiliary variables; the other method is a one-stage stratum sampling method using CA as the stratum variable, named as OSS thereafter. Both methods have been widely used for land cover acreage estimation in previous studies.27,28,35 In addition, to avoid the random errors that might go un-noticed in one experiment, we carried out 30 experiments under each sampling rate. Two indices, sampling average accuracy and standard errors of sample averages, were adopted for assessing the performance of each sampling method. Sampling average accuracy (d) is defined as one minus the sampling average error [Eq. (9)]. The standard error of sample averages (sde) is defined as the standard deviation of sampling errors [Eq. (10)] showing the extent of variation between the estimator and the truth value. Display Formula

d=1|Y^¯Y|/Y,(9)
Display Formula
sde=iN(Y^iY)2/N,(10)
where Y is true value of wheat acreage derived from the TM image 2009, Y^¯ is mean value of estimated wheat acreage with N times experiments, and N is the total number of experiments (30 times in our study).

Stratifications

The entire study area was divided into 36 layers through two-stage stratification in the TSS method. The stratified parameters were presented in Table 2 showing the higher the number of layers, the greater the fluctuation of wheat within each unit. The wheat planting acreage of the former three layers accounted for more than 70% of the whole wheat area.

Table Grahic Jump Location
Table 2The stratum boundaries definition with coefficient of variance of crop acreage calculated by historical multiyear wheat distribution maps.
Accuracy Assessment

Figure 3 showed the comparisons among the three sampling methods (TSS, OSSmean, and SRS). From Figs. 3(a) and 3(b), the accuracy of all these three sampling methods was increasing with the sampling ratio enlargement. Overall, our proposed method (TSS) had the highest average accuracy, followed by OSSmean, whereas the SRS had the lowest average accuracy, which had been proved in those of previous studies.3 Among the three sampling methods, our proposed method (TSS) had the lowest standard deviations of sampling errors. The average accuracy and standard deviations of sampling errors of TSS are 97% and 0.02, respectively, when sampling ratio reaches at 3%. The average accuracy of TSS became stable when the sampling ratio is higher than 3%, while TSS’s standard deviations of sampling errors becomes stable when the sampling ratio is higher than 4%. However, for the SRS method, sample accuracy did not show a trend toward stability even though the sample rate reaches 10%.

Graphic Jump LocationF3 :

Results of comparison experiments, (a) average accuracies of three sampling methods. (b) Standard variance of sampling errors of three sampling method. (c) Average accuracies of OSS with different stratified variables. (d) Standard variance of sampling errors of OSS with different stratified variables.

Figures 3(c) and 3(d) showed comparisons among OSS with wheat acreage calculated from different historical wheat maps as an auxiliary variable. From Figs. 3(c) and 3(d), we could see that OSSmean had higher average accuracy and a lower standard deviation of sampling errors than OSS2007 and OSS2006, but had similar accuracy to OSS2008 in our study area. The reason why OSSmean and OSS2008 had similar accuracy will be discussed in Sec. 5.

Feasibility of Diachronic Remote Sensing Images used in the Sampling Processes

The success of historical remote sensing images used in sampling processes depends on the stability of the crop planting system. For wheat in our study area, a correlation coefficient matrix was computed. Table 3 gives the correlations between two different years in which we could observe that correlations coefficients were higher than 0.65, which is similar to the findings in intensive agriculture segments,10,21 which indicated that the crop planting system in our study area is stable overall.

Table Grahic Jump Location
Table 3Correlations coefficient for wheat in our study area.

Table 4 summarizes the fluctuation of wheat planting in the parcels (sampling units) of our study area. None_3 means that for a given parcel in 2006 to 2009, it was planted with wheat in just 1 year (such as 2006) and not planted in the other 3 years (such as 2007, 2008, and 2009). One_3 means that a parcel was planted with wheat in 1 year (such as 2006), and in the other 3 years, it was planted just one additional time (2007, 2008, or 2009), for a total of 2 years being planted. Two_3 and ALL_3 have the similar meanings. The number ratio is defined as the number of parcels satisfying the prerequisites (such as planted in 2006 and not planted in other years) divided by the total number of parcels in our study area. Similarly, the acreage ratio is defined as the area of parcels satisfying the prerequisites divided by the total area of parcels in our study area. From this table, we can see that the number ratio and acreage ratio are much higher in ALL_3 than in other cases, which further indicates that the crop planting system is sufficiently stable in most parcels of our study area. To sum up, both Tables 2 and 3 indicate that using historical remote sensing images to construct the auxiliary variable of stratified sampling is suitable for use in our study area.

Table Grahic Jump Location
Table 4Statistical analysis of wheat among multihistorical years.
Advantages of the Stratified Variables used in our Study

From Table 3, we can see that the wheat acreage of 2009 is highest correlated with the mean wheat acreage of 2006, 2007, and 2008, with 2008 as the year showing the next highest correlation levels. It explains clearly, why OSSmean and OSS2008 have similar sampling accuracy levels and why these levels are higher than those of OSS2006 and OSS2007. However, in our proposed method (TSS), we used mean wheat acreage of 2006 to 2008 as the auxiliary variable in the second stage of stratified sampling rather than the wheat acreage of 2008 for two reasons. First, the correlation between the wheat acreage of 2009 and the mean wheat acreage of 2006 to 2008 is highest; therefore, it is expected that using the mean wheat acreage as auxiliary variable could achieve higher accuracy. Besides, in terms of relative efficiency, which is approximately 1/(1r2), where r is the correlation coefficient,10 the mean wheat acreage of multihistorical years could improve the regression estimation precision about 2.5 times. Second, it is difficult to determine which year has the highest correlation with a target year, especially for an investigated area without enough background information about planting systems. Hence, the mean wheat acreage of multihistorical years can be defined as an ideal variable for stratum division and acreage estimation, which can reduce the indeterminacy of auxiliary variable definition using only historical single-year wheat acreage data.

In the first stage of TSS, we used the CVCA as the auxiliary variable by which to stratify the parcels for stratum definition, with those having similar crop planting systems clustered in the same stratum. In Fig. 4, an example is given for analyzing the stratum difference. The parcels (Parcel A and Parcel B) highlighted by cyan lines show the differences of stratums defined by CVCA and CA variables, respectively. In Figs. 4(a) and 4(b), the numbers 1 to 6 are the stratum numbers. In Fig. 4(a), a lower number indicates that the mean wheat area of the parcel is smaller. In Fig. 4(b), a lower number indicates that the wheat area of the parcel has a higher fluctuation. From Figs. 4(a) and 4(b), we can see that Parcels A and B will be grouped into a same stratum, when using CA as the auxiliary variable yet into different stratums when using CVCA. Observing Fig. 4(c), Figs. 4(d) and 4(e), we can see that Parcel A was planted twice whereas Parcel B was planted three times during 2006 to 2008 although the mean wheat sowing area of 2006 to 2008 was similar between Parcels A and B. The planting pattern of Parcel A is more stable than that of Parcel B. Hence, using CVCA as auxiliary variable can better identify local crop planting patterns.

Graphic Jump LocationF4 :

Stratification by coefficient of variance of crop acreage (CVCA) and crop acreage (CA). (a) Different stratum defined by CA. (b) Different stratum defined by CVCA. (c) Zooming in wheat thematic map of 2006. (d) Zooming in wheat thematic map of 2007. (e) Zooming in wheat thematic map of 2008. (f) Zooming in wheat thematic map of 2009.

In this study, we used a TSS method to estimate the 2009 winter wheat acreage in Beijing. The auxiliary variables used in the first and second stages were CVCA and CA, respectively. Multiple historical remote sensing images were used to create these stratification variables. To examine the performance of our proposed method, two other sampling methods were also used to estimate Beijing’s 2009 winter wheat acreage: SRS, and a one-stage stratification method with CA as the auxiliary variable. Our results show that:

First, the average accuracy of our proposed method becomes stable when the sampling ratio is higher than 3%, while the average accuracies of other methods do not show an obvious stable trend when the sampling ratios increase from 1% to 10%. This finding further indicates that our method requires fewer samples to obtain the predefined accuracy; therefore, it is more efficient and can keep costs down.

Second, using mean CA derived from multiple year historical remote sensing data as an auxiliary variable has higher accuracy than using CA derived from one historical year remote sensing data as the auxiliary variable in one stage sampling. Besides, this method does not take into account that determining which year has the highest correlation with the target year is difficult, especially for a study area (investigated area) where sufficient background information about the planting system is lacking.

With the great abundance of historical remote sensing images, more and more useful historical crop thematic maps can be employed to create the auxiliary variables proposed in this study. Therefore, the method proposed in this study is promising in areas with sufficient stable crop planting system.

In a future study, we will try to use our TSS method to estimate acreage for other crops, such as corn or soybeans, which usually have lower classification accuracy due to the more complicated cropping systems that exist in autumn. The influences of classification errors from the historical remote sensing images will be explored for CA estimation. Besides, in our method, the CVCA and the mean CA of the sampling units were used as stratification variables. CV was defined as the ratio of the standard deviation to the mean. Extreme values could be a big influence on CV. The CV is more reliable when more data are used for computation. In this study, only 3 years’ wheat acreage values were used for the computation of the CV. More years’ data should be adopted in the future to further validating the performance of our proposed method. Furthermore, the TSS method will also be popularized and validated at a large scale such as a national scale, which might provide the basis for national regular operation of CA investigations.

This research was supported by the Natural Science Foundation of China (Grant No. 41301444) and the Major Project of High-resolution Earth Observation System. We thank the anonymous reviewers and editors for their valuable comments and suggestions on improving the quality of this paper.

Duveiller  G., Defourny  P., “A conceptual framework to define the spatial resolution requirements for agricultural monitoring using remote sensing,” Remote Sens. Environ.. 114, (11 ), 2637 –2650 (2010), CrossRef. 0034-4257 
Pan  Y. et al., “Winter wheat area estimation from MODIS-EVI time series data using the Crop Proportion Phenology Index,” Remote Sens. Environ.. 119, , 232 –242 (2012), CrossRef. 0034-4257 
Gallego  F. J., “Remote sensing and land cover area estimation,” Int. J. Remote Sens.. 25, (15 ), 3019 –3047 (2004), CrossRef. 0143-1161 
Czaplewski  R. L., Catts  G. P., “Calibration of remotely sensed proportion or area estimates for misclassification error,” Remote Sens. Environ.. 39, (1 ), 29 –43 (1992), CrossRef. 0034-4257 
Hanuschak  G. A. et al., “Crop-area estimates from LANDSAT; transition from research and development to timely results,” IEEE Trans. Geosci. Remote Sens.. GE-18, (2 ), 160 –166 (1980), CrossRef. 0196-2892 
Walsh  T. A., Burk  T. E., “Calibration of satellite classifications of land area,” Remote Sens. Environ.. 46, (3 ), 281 –290 (1993), CrossRef. 0034-4257 
Bailey  J. T., Boryan  C. G., “Remote sensing applications in agriculture at the USDA National Agricultural Statistics Service” (2010).
Carfagna  E., Gallego  F. J., “Using remote sensing for agricultural statistics,” Int. Stat. Rev.. 73, (3 ), 389 –404 (2005), CrossRef. 0306-7734 
Carfagna  E., Marzialetti  J., “Continuous innovation of the quality control of remote sensing data for territory management,” in Statistics for Innovation. , Erto  P., Ed., pp. 145 –160,  Springer-Verlag ,  Italia Milan  (2009).
Gallego  F. J., “On the feasibility of diachronic regression estimators with ground survey and Landsat TM data,” Int. J. Remote Sens.. 19, (8 ), 1621 –1625 (1998), CrossRef. 0143-1161 
Gallego  F. J., “Crop area estimation in the MARS project,” in  Conference on Ten Years of the MARS Project ,  Brussels, Belgium  (1999).
Gallego  F. J., “Stratified sampling of satellite images with a systematic grid of points,” ISPRS J. Photogramm. Remote Sens.. 59, (6 ), 369 –376 (2005), CrossRef. 0924-2716 
Gallego  F. J., Stibig  H. J., “Area estimation from a sample of satellite images: the impact of stratification on the clustering efficiency,” Int. J. Appl. Earth Obs. Geoinf.. 22, , 139 –146 (2013), CrossRef. 0303-2434 
Gallego  F. J., Bamps  C., “Using CORINE land cover and the point survey LUCAS for area estimation,” Int. J. Appl. Earth Obs. Geoinf.. 10, (4 ), 467 –475 (2008), CrossRef. 0303-2434 
Holko  M. L., Sigman  R. S., “The role of Landsat data in improving U. S. crop statistics,” in  18th Int. Symp. Remote Sens. Environ. ,  Paris, France , pp. 307 –320 (1985).
Delincé  J. A., “European approach to area frame survey,” in  Proc. Conf. Agri. Environ. Stat. Appl. Rome (CAESAR) , Vol. 2, pp. 463 –472 (2001).
Cotter  J. J., Tomczak  C. M., “An image analysis system to develop area sampling frames for agricultural surveys,” Photogramm. Eng. Remote Sens.. 60, (3 ), 299 –306 (1994). 0099-1112 
Gallego  F. J., Carfagna  E., Peedell  S., “The use of CORINE land cover to improve area frame survey estimates,” Res. Off. Stat.. 2, (2 ), 99 –122 (1999).
Deppe  F., “Forest area estimation using sample surveys and Landsat MSS and TM data,” Photogramm. Eng. Remote Sens.. 64, (4 ), 285 –292 (1998). 0099-1112 
Lu  D., Weng  Q., “A survey of image classification methods and techniques for improving classification performance,” Int. J. Remote Sens.. 28, (5 ), 823 –870 (2007), CrossRef. 0143-1161 
Gonzalez-Alonso  F. et al., “Remote sensing and agricultural statistics: crop area estimation in north-eastern Spain through diachronic Landsat TM and ground sample data,” Int. J. Remote Sens.. 18, (2 ), 467 –470 (1997), CrossRef. 0143-1161 
Toukiloglou  P., “Comparison of AVHRR, MODIS and VEGETATION for land cover mapping and drought monitoring at 1 km spatial resolution,” Ph.D. Thesis,  Cranfield University  (2007)
Foody  G. M., Mathur  A., “Toward intelligent training of supervised image classifications: directing training data acquisition for SVM classification,” Remote Sens. Environ.. 93, (1–2 ), 107 –117 (2004), CrossRef. 0034-4257 
Foody  G. M., Mathur  A., “The use of small training sets containing mixed pixels for accurate hard image classification: training on mixed spectral responses for classification by a SVM,” Remote Sens. Environ.. 103, (2 ), 179 –189 (2006), CrossRef. 0034-4257 
Melgani  F., Bruzzone  L., “Classification of hyperspectral remote sensing images with support vector machines,” IEEE Trans. Geosci. Remote Sens.. 42, (8 ), 1778 –1790 (2004), CrossRef. 0196-2892 
Wynne  R. H. et al., “Optical remote sensing for forest area estimation,” J. For.. 98, (5 ), 31 –36 (2000). 0022-1201 
Broich  M. et al., “A comparison of sampling designs for estimating deforestation from Landsat imagery: a case study of the Brazilian Legal Amazon,” Remote Sens. Environ.. 113, (11 ), 2448 –2454 (2009), CrossRef. 0034-4257 
Stehman  S. V., Sohl  T. L., Loveland  T. R., “Statistical sampling to characterize recent United States land-cover change,” Remote Sens. Environ.. 86, (4 ), 517 –529 (2003), CrossRef. 0034-4257 
Foody  G. M., “Status of land cover classification accuracy assessment,” Remote Sens. Environ.. 80, (1 ), 185 –201 (2002), CrossRef. 0034-4257 
Tsiligirides  T. A., “Remote sensing as a tool for agricultural statistics: a case study of area frame sampling methodology in Hellas,” Comput. Electron. Agric.. 20, (1 ), 45 –77 (1998), CrossRef. 0168-1699 
Cozzucoli  P. C., “Simultaneous confidence intervals on partial means of classes in the two-stage stratified sampling,” Stat. Pap.. 51, (3 ), 673 –685 (2010), CrossRef. 0932-5026 
Pradhan  S., “Crop area estimation using GIS, remote sensing and area frame sampling,” Int. J. Appl. Earth Obs. Geoinf.. 3, (1 ), 86 –92, (2001), CrossRef. 0303-2434 
Cochran  W. G., Sampling Techniques. , 3rd ed.,  John Wiley & Sons ,  New York  (1977).
Tomppo  E., Czaplewski  R., “The role of remote sensing in global forest assessment,” in  Proc. FAO Expert Consult. Global For. Resour. Assess. Kotka , Vol. 889, pp. 267 –281 (2002).
Hansen  M. C. et al., “Humid tropical forest clearing from 2000 to 2005 quantified by using multitemporal and multiresolution remotely sensed data,” Proc. Natl. Acad. Sci. U. S. A.. 105, (27 ), 9439 –9444 (2008), CrossRef. 0027-8424 

Jinshui Zhang is an associate professor in the State Key Laboratory of Earth Surface Processes and Resource Ecology and College of Resources Science and Technology at Beijing Normal University, Beijing, China. His research interests include land cover and use detection, spatial sampling for classification error calibration, and crop mapping using remote sensing.

Shuang Zhu is a doctor in the State Key Laboratory of Earth Surface Processes and Resource Ecology and College of Resources Science and Technology at Beijing Normal University, Beijing, China. Her research interests include land cover and use detection and crop identification of remote sensing.

Xiufang Zhu is an associate professor in the State Key Laboratory of Earth Surface Processes and Resource Ecology and College of Resources Science and Technology at Beijing Normal University, Beijing, China. Her research interests include the response research of human activities and climate change and application of remote sensing.

Guanyuan Shuai is a master’s student of the State Key Laboratory of Earth Surface Processes and Resource Ecology and College of Resources Science and Technology at Beijing Normal University, Beijing, China. His research interests include agricultural statistics of remote sensing and pattern recognition of remote sensing.

Dengfeng Xie is a master’s student of the State Key Laboratory of Earth Surface Processes and Resource Ecology and College of Resources Science and Technology at Beijing Normal University, Beijing, China. His research interests include agricultural statistics of remote sensing and land cover and use detection.

© The Authors. Published by SPIE under a Creative Commons Attribution 3.0 Unported License. Distribution or reproduction of this work in whole or in part requires full attribution of the original publication, including its DOI.

Citation

Jinshui Zhang ; Shuang Zhu ; Xiufang Zhu ; Guanyuan Shuai and Dengfeng Xie
"Improving sampling efficiency of crop acreage estimation using wheat planting rule from historical remote sensing", J. Appl. Remote Sens. 8(1), 083663 (Mar 17, 2014). ; http://dx.doi.org/10.1117/1.JRS.8.083663


Figures

Graphic Jump LocationF2 :

Flowchart of the proposed method.

Graphic Jump LocationF3 :

Results of comparison experiments, (a) average accuracies of three sampling methods. (b) Standard variance of sampling errors of three sampling method. (c) Average accuracies of OSS with different stratified variables. (d) Standard variance of sampling errors of OSS with different stratified variables.

Graphic Jump LocationF4 :

Stratification by coefficient of variance of crop acreage (CVCA) and crop acreage (CA). (a) Different stratum defined by CA. (b) Different stratum defined by CVCA. (c) Zooming in wheat thematic map of 2006. (d) Zooming in wheat thematic map of 2007. (e) Zooming in wheat thematic map of 2008. (f) Zooming in wheat thematic map of 2009.

Tables

Table Grahic Jump Location
Table 1Accuracy assessment of the historical images classification.
Table Grahic Jump Location
Table 3Correlations coefficient for wheat in our study area.
Table Grahic Jump Location
Table 2The stratum boundaries definition with coefficient of variance of crop acreage calculated by historical multiyear wheat distribution maps.
Table Grahic Jump Location
Table 4Statistical analysis of wheat among multihistorical years.

References

Duveiller  G., Defourny  P., “A conceptual framework to define the spatial resolution requirements for agricultural monitoring using remote sensing,” Remote Sens. Environ.. 114, (11 ), 2637 –2650 (2010), CrossRef. 0034-4257 
Pan  Y. et al., “Winter wheat area estimation from MODIS-EVI time series data using the Crop Proportion Phenology Index,” Remote Sens. Environ.. 119, , 232 –242 (2012), CrossRef. 0034-4257 
Gallego  F. J., “Remote sensing and land cover area estimation,” Int. J. Remote Sens.. 25, (15 ), 3019 –3047 (2004), CrossRef. 0143-1161 
Czaplewski  R. L., Catts  G. P., “Calibration of remotely sensed proportion or area estimates for misclassification error,” Remote Sens. Environ.. 39, (1 ), 29 –43 (1992), CrossRef. 0034-4257 
Hanuschak  G. A. et al., “Crop-area estimates from LANDSAT; transition from research and development to timely results,” IEEE Trans. Geosci. Remote Sens.. GE-18, (2 ), 160 –166 (1980), CrossRef. 0196-2892 
Walsh  T. A., Burk  T. E., “Calibration of satellite classifications of land area,” Remote Sens. Environ.. 46, (3 ), 281 –290 (1993), CrossRef. 0034-4257 
Bailey  J. T., Boryan  C. G., “Remote sensing applications in agriculture at the USDA National Agricultural Statistics Service” (2010).
Carfagna  E., Gallego  F. J., “Using remote sensing for agricultural statistics,” Int. Stat. Rev.. 73, (3 ), 389 –404 (2005), CrossRef. 0306-7734 
Carfagna  E., Marzialetti  J., “Continuous innovation of the quality control of remote sensing data for territory management,” in Statistics for Innovation. , Erto  P., Ed., pp. 145 –160,  Springer-Verlag ,  Italia Milan  (2009).
Gallego  F. J., “On the feasibility of diachronic regression estimators with ground survey and Landsat TM data,” Int. J. Remote Sens.. 19, (8 ), 1621 –1625 (1998), CrossRef. 0143-1161 
Gallego  F. J., “Crop area estimation in the MARS project,” in  Conference on Ten Years of the MARS Project ,  Brussels, Belgium  (1999).
Gallego  F. J., “Stratified sampling of satellite images with a systematic grid of points,” ISPRS J. Photogramm. Remote Sens.. 59, (6 ), 369 –376 (2005), CrossRef. 0924-2716 
Gallego  F. J., Stibig  H. J., “Area estimation from a sample of satellite images: the impact of stratification on the clustering efficiency,” Int. J. Appl. Earth Obs. Geoinf.. 22, , 139 –146 (2013), CrossRef. 0303-2434 
Gallego  F. J., Bamps  C., “Using CORINE land cover and the point survey LUCAS for area estimation,” Int. J. Appl. Earth Obs. Geoinf.. 10, (4 ), 467 –475 (2008), CrossRef. 0303-2434 
Holko  M. L., Sigman  R. S., “The role of Landsat data in improving U. S. crop statistics,” in  18th Int. Symp. Remote Sens. Environ. ,  Paris, France , pp. 307 –320 (1985).
Delincé  J. A., “European approach to area frame survey,” in  Proc. Conf. Agri. Environ. Stat. Appl. Rome (CAESAR) , Vol. 2, pp. 463 –472 (2001).
Cotter  J. J., Tomczak  C. M., “An image analysis system to develop area sampling frames for agricultural surveys,” Photogramm. Eng. Remote Sens.. 60, (3 ), 299 –306 (1994). 0099-1112 
Gallego  F. J., Carfagna  E., Peedell  S., “The use of CORINE land cover to improve area frame survey estimates,” Res. Off. Stat.. 2, (2 ), 99 –122 (1999).
Deppe  F., “Forest area estimation using sample surveys and Landsat MSS and TM data,” Photogramm. Eng. Remote Sens.. 64, (4 ), 285 –292 (1998). 0099-1112 
Lu  D., Weng  Q., “A survey of image classification methods and techniques for improving classification performance,” Int. J. Remote Sens.. 28, (5 ), 823 –870 (2007), CrossRef. 0143-1161 
Gonzalez-Alonso  F. et al., “Remote sensing and agricultural statistics: crop area estimation in north-eastern Spain through diachronic Landsat TM and ground sample data,” Int. J. Remote Sens.. 18, (2 ), 467 –470 (1997), CrossRef. 0143-1161 
Toukiloglou  P., “Comparison of AVHRR, MODIS and VEGETATION for land cover mapping and drought monitoring at 1 km spatial resolution,” Ph.D. Thesis,  Cranfield University  (2007)
Foody  G. M., Mathur  A., “Toward intelligent training of supervised image classifications: directing training data acquisition for SVM classification,” Remote Sens. Environ.. 93, (1–2 ), 107 –117 (2004), CrossRef. 0034-4257 
Foody  G. M., Mathur  A., “The use of small training sets containing mixed pixels for accurate hard image classification: training on mixed spectral responses for classification by a SVM,” Remote Sens. Environ.. 103, (2 ), 179 –189 (2006), CrossRef. 0034-4257 
Melgani  F., Bruzzone  L., “Classification of hyperspectral remote sensing images with support vector machines,” IEEE Trans. Geosci. Remote Sens.. 42, (8 ), 1778 –1790 (2004), CrossRef. 0196-2892 
Wynne  R. H. et al., “Optical remote sensing for forest area estimation,” J. For.. 98, (5 ), 31 –36 (2000). 0022-1201 
Broich  M. et al., “A comparison of sampling designs for estimating deforestation from Landsat imagery: a case study of the Brazilian Legal Amazon,” Remote Sens. Environ.. 113, (11 ), 2448 –2454 (2009), CrossRef. 0034-4257 
Stehman  S. V., Sohl  T. L., Loveland  T. R., “Statistical sampling to characterize recent United States land-cover change,” Remote Sens. Environ.. 86, (4 ), 517 –529 (2003), CrossRef. 0034-4257 
Foody  G. M., “Status of land cover classification accuracy assessment,” Remote Sens. Environ.. 80, (1 ), 185 –201 (2002), CrossRef. 0034-4257 
Tsiligirides  T. A., “Remote sensing as a tool for agricultural statistics: a case study of area frame sampling methodology in Hellas,” Comput. Electron. Agric.. 20, (1 ), 45 –77 (1998), CrossRef. 0168-1699 
Cozzucoli  P. C., “Simultaneous confidence intervals on partial means of classes in the two-stage stratified sampling,” Stat. Pap.. 51, (3 ), 673 –685 (2010), CrossRef. 0932-5026 
Pradhan  S., “Crop area estimation using GIS, remote sensing and area frame sampling,” Int. J. Appl. Earth Obs. Geoinf.. 3, (1 ), 86 –92, (2001), CrossRef. 0303-2434 
Cochran  W. G., Sampling Techniques. , 3rd ed.,  John Wiley & Sons ,  New York  (1977).
Tomppo  E., Czaplewski  R., “The role of remote sensing in global forest assessment,” in  Proc. FAO Expert Consult. Global For. Resour. Assess. Kotka , Vol. 889, pp. 267 –281 (2002).
Hansen  M. C. et al., “Humid tropical forest clearing from 2000 to 2005 quantified by using multitemporal and multiresolution remotely sensed data,” Proc. Natl. Acad. Sci. U. S. A.. 105, (27 ), 9439 –9444 (2008), CrossRef. 0027-8424 

Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging & repositioning the boxes below.

Related Book Chapters

Topic Collections

PubMed Articles
Advertisement
  • Don't have an account?
  • Subscribe to the SPIE Digital Library
  • Create a FREE account to sign up for Digital Library content alerts and gain access to institutional subscriptions remotely.
Access This Article
Sign in or Create a personal account to Buy this article ($20 for members, $25 for non-members).
Access This Proceeding
Sign in or Create a personal account to Buy this article ($15 for members, $18 for non-members).
Access This Chapter

Access to SPIE eBooks is limited to subscribing institutions and is not available as part of a personal subscription. Print or electronic versions of individual SPIE books may be purchased via SPIE.org.