Crop hectare estimation by
means of satellite image analysis & sampling techniques
Z –
Shori Introduction Agricultural estimates using remotely sensed data is a subject has been investigated and tested around the world during the last decade. In this sense it important to point out that Remote Sensing techniques do not replace traditional procedures for gathering agricultural data, considering these procedures objects and reliable. However, depending on the actual situation in each country, Remote Sensing could become a decisive element of the system. Objective The main objective of this project is wheat / barely identification and hectare estimation. Project area The project area is the GORGAN-GONBAD region of MAZANDARAN province in the north-east part of IRAN. It is one of most important areas for crop production in the country, presenting a typical intensive agricultural pattern. It is a almost flat region upto the foot of the ALBORZ mountains with soils ranging from high to regular capacity for agricultural purposes. Main crops in the region area wheat, barely and cotton being followed by second harvest corps like rice, soyabean, broad bean, potatoes, different vegetables and fruits complete the agricultural pattern in some specific area. Wheat and cotton usually are irrigated but almost all barely production is just rainfed. Field sizes vary accordingly from approximately 20-50 ha 1-5 ha. The project has been done in 3 phases as follow: PHASE – I *STRATIFICATION Stratification is a well known procedure by which a region is divided into subarea or strata with the objective of grouping homogeneous areas, with respect to one or more specific variables. The purpose of the stratification in agricultural surveys is toe increase the precision of the sample survey estimates. * PROJECT AREA STRATIFICATION The stratified sampling design used in this project provides separate estimates of mean and variance for each stratum generally given a more precise estimates of these parameters for the entire population than an in the case of a random sample design for the whole area. Due to the necessity of estimating the sampling error within each stratum, attention should be given to the fact that the stratum size permits the allocation of a proper number of sample units to compute that error. The GORGAN-GONBAD area was stratified using topographic maps and Landsat MSS and TM images from different dates of the major crops growing season. Four strara were identified and delineated on the Landsat TM image. A field trip allowed the boundary control and description: LEGEND STRATUM 1: correspond top the piedmont of ALBORZ mountains landuse is characterized by a high agricultural activity pattern with small to medium field size (1-10 ha). Main crops are wheat and cotton also including vegetables, fruits, potatoes etc. total area – 109700 ha STRATUM 2 : correspond to the alluvial plain of the Gorgan river it is a large area with intensive agricultural activity. Field size varies from medium to large (5 – 25 has) being wheat and cotton almost exclusive crops during the spring season. Wheat is largely the major crop and only around villages other crops appear in smaller fields. Total area = 142200 ha. STRATUM 3 : Correspond to the alluvial plain of Gorgan river this area presents an extensive agricultural pattern being barely the excluyent crop few wheat field are observed only where soil conditions are better. Field sizers are the largest of the region (more than 20 ha) when smaller is due to drainage channels pattern near villages. Total area = 108500 STRATUM 4 : Correspond to the alluvial plain of Gorgan river it is characterized by an intensive agricultural practice devoted mainly to summer crops (cotton). Fields size is medium to large and few winter crops are found. Total area = 20800 ha *SAMPLE DESIGN 1. MAP PREPARATION Basic map is 1:250000 topographic maps of GORGAN-GONBAD region. The villages and roads was updated from the road map scale 1:300000. Next step is to transfer the boundaries into the original 1:250000 maps and measure strata extension to know final number of sample to be allocated. 2. SAMPLE UNIT SIZE The determination of the sample size is one important decision in the designing of an agricultural survey. Sample size is commonly a function of accuracy desired in the estimate as well as of the cost of the field survey and budget restrictions. In the case of GORGAN-GONBAD project the design is a stratified random sampling. Direct expansion and ratio and regression estimators is used and the sample unit size was established in 100 hectare or (1 sq km). 3. SAMPLE SELECTION The most frequent methods used for sample selection is known as “equal probability”, where each sample is given the same probability of being drawn. Being the total number of possible sample units (N) for each stratum and the number of samples to be selected (n), the probability that every sample has to be drawn is 1/N considering the “with replacement” approach. The selection procedure usually requires the generation of a grid cell where each cell represents the size of the sample unite, this grid should be transferred onto topographic or cadastral maps or satellite images in order to geographically locate the samples that have been identified and labeled. The selected samples which its number has been previously computed are drawn by mean of tables of random numbers. For GORGAN – GONBAD area after the final adjustments of the strata boundaries based on the new Landsat TM 1:100000 print was done the number of possible samples (N) and the number of selected samplers (n) for different strata resulted as follows: STRATUM 1 :number of possible samples N1 – 1097, number of selected samples n1=10 STRATUM 2 : number of possible samples N2 = 1422, number of selected samples n2=16 STRATUM 3 : number of possible samples N3 = 1085, Number of selected samples N3 = 13 STRATUM 4 : number of possible samples N4 = 208, number of selected samples n4 = 5 PHASE II *SAMPLE EXTRACTION AND DELINEATION The next step the extraction of the sample areas and boundary delineation. The selected units should have their corresponding X and Y coordinates maps or aerial location in the corresponding maps or aerial photographs where the boundary delineation are going to be done. This procedure could also be done using enlargements of high resolution satellites images – Landsat TM or SPOT. This possibility becomes particularly interesting when the satellite images acquired during the surveyed crop season are available at this stage of the crop estimate program. In GORGAN-GONBAD survey, given the fact of the lack of detailed or cadastral maps and aerial photos and the availability of a real time Landsat TM image, the X and Y coordinates of each selected sample were plotted on the Landsat TM 1:100000 image acquired during the wheat/barely growing period. Each sample delineation was done on the respective color enhanced print enlargements obtained from the monitor. This method was used to delineate the 44 samples of project area. *FIELD DATA COLLECTION The quality of the final data, the accuracy of the final results will largely depends on the quality of the data obtained during the field survey. The main purpose of field trip was to collect data for training and testing the classifications algorithm run during the automatic crop identifications analysis and also to produce a direct expansion estimates of wheat/barely (field data only). In the agricultural estimates of GORGAN – GONBAD area lf Landsat TM quadrant acquired during the wheat/barely growing season – flowering stage was available at project site, both in digital and 1:100,000 scale color print format. Based on this material the selected samples were plotted. The field upto trip was designed and the itinerary of reaching the segments was prepared. The total 44 selected samples were surveyed and the corresponding field data recorded. Field measurements were done back in the Remote Sensing laboratory using a grid cell over the Landsat based actual land use map previously controlled in the field. PHASE III *SATELLITE IMAGE PROCESSING AND DIGITAL ANALYSIS PROCEDURES 6 TM spectral bands and higher sensor sensitivity provides adequate information for the spectral analysis, due to extension of the area the expected spectral confusion mainly between wheat and barely and field size variation ranging from 1 upto 50 or more hectares, one Landsat TM quadrant of April 22, 1991 was selected. For surveying areas with small field size and where spectral confusion is expected because of high spectral variation, merging SPOT/PAN and Landsat TM provided excellent information in project to be developed during the satellite image processing could be grouped in to the following categories. 1. IMAGE PREPARATION When sampling approach is being used sample unite should be located and identified in the image and then extracted from it. For a more efficient use of computing time a large unique file including all sample unit is generated for displaying and analysis. 2. SPECTRAL SIGNATURE EXTRACTION A specific statistic file is created for each crop and spectral values extracted and their statistical characteristics means, variances, etc generated to achieve a representative signature pattern for respective class. 3. SPECTRAL SEPERABILITY ANALYSIS The analysis of separability of the signatures files will provide a measurement of the spectral confusion of the best and combination, as well as expected performance of the classification algorithm. Band 3, 4 and 5 Landsat TM have been proved to be the most useful set for crop identification and vegetation analysis. 4. CLASSIFICATION AND ACCURACY EVALUATION Supervised classification maximum likelihood algorithm is used. Samples form which crop training signature have been taken are classified and error measured by the confusion matrix (a priori/posteriori or blind test). The result was 90% accuracy of wheat and 88% accuracy for barely. *CROP ESTIMATES BASED ON FIELD AND SATELLITE DATA After the joint analysis of the information gathered during the field data collection program and the from satellite image, with the data available from the ground survey and from the satellite data classification of project area, several statistical procedures of crops in the GORGAN-GONBAD region. In all case the use of satellite data throughout a regression estimator two phase reduces the variance of the estimate and consequently increases the statistical precision of final estimates. In this senes it is important to remake that the higher the correlation between satellite and field data / the higher the confidence of the estimator and the lower the cost of the whole program. Final estimates based on regression estimator two phases for the whole studies region is produced as follows.
CONCLUSION Use of Remote Sensing as a modern tool, with high performance capability, particularly in the large region in which, the ground statistics data collection and crop hectarage estimation using the traditional methods causes the serious difficulties can be efficient to reduce the number of ground samples, cost / time consummation and handle the expected and timely results. The apply of timely satellite data in a regression estimator double sampling which has been used in described project, can be concerned to establish a reliable and operative crop estimation program. | ||||||||||||||||||||