Evaluation of Efficiency between Classification Methods and Spectral Indices in Cropped Area Estimation of Shush County

Introduction Agriculture is the essential sector for promoting food security. Crop area estimation (CAE) can meet the requirements of the crop monitoring plan. The organizing basis of the cultivation pattern is recognizing the types of crops and examining the condition of their crop area. Shush county in Khuzestan Province has 300,000 hectares of the crop area. It is one of the agricultural hubs of Iran because it has a record annual production of more than two million tons of strategic crops such as wheat, sugar beet, and corn. CAE affects the amount of net production and shortage or surplus of produce for market steadiness. Traditional approaches for CAE are time-consuming and costly and are not widely enforceable. Remote sensing (RS) data provide good information for decision-makers by determining the crop type and the crop area. RS data has made it possible to avoid continuous reference to agricultural lands with less time and cost than another usual method and accurate CAE. Also, the use of multi-time images during the growing season of agricultural products allows the use of spectral curves when related to the crop calendar of each crop. This spectral curve is almost separate for each product and increases the ability to distinguish between products. Therefore, multi-temporal images support segregation based on multispectral images of products. The current study follows a speedy method with appropriate accuracy established on satellite image classification algorithms and spectral indices to identify and separate crops with RS data in Shush County.Materials and Methods Landsat-8 data with path/row coordinates 166/38 extracted from the USGS website were used to identify and separate the cultivated lands of the region. The reason for choosing Landsat images is the relatively suitable temporal and spatial resolution, availability, and the appropriate time distribution with the product growth period. The Landsat 8 carries 2-sensors, OLI (Operational Land Imager) and TIRS (Thermal Infrared Sensor). The OLI sensor with a spatial resolution of 30 meters has 8-bands in the visible spectrum, near-infrared (NIR), short-wavelength infrared (SWIR), and a panchromatic band with a spatial resolution of 15 meters. The TIRS sensor can record thermal infrared radiation with a spatial resolution of 100 meters with the help of 2-bands in atmospheric windows of 10.6 to 11.2 micrometers for band 10 and 11.5 to 12.5 micrometers for band 11. This research used bands 1-7 of the Landsat-8 OLI sensor with a spatial resolution of 30 meters after the initial corrections of satellite images. The spectral similarity between the region's dominant crops has made it impossible to select a single image to differentiate and extract the cultivation pattern. Wheat and barley have a high spectral similarity. The peak of the greenness of these products is in the first four months of the year, which has high NDVI values at this time. Therefore, choosing a good time to separate the crops was feasible by referring to the Khuzestan Organization Agriculture-Jihad (KOAJ) and receiving the regional crops calendar in 2018-19. Then, the low-level cloud cover images on April 24, June 27, and August 30, 2019, were selected for classification based on the crop calendar. Planting, harvesting, maximum greenness, and ripening information of the dominant crops in the area were pivotal in obtaining image dates. In dates selected related to the images were considered planting, harvesting, maximum greenery, and ripening information of the region's dominant crops.Results and Discussion According to the results, from total crop area in Shush county (163313.7 hectares) is allocated about 103513.2 hectares (63.4% of the county's crop area) to the ANN, about 102875.1 hectares (63.0% of the county's crop area) to the SVM, and about 102,277.3 hectares (62.6% of the county's crop area) to the NDVI, which in comparison with the KOAJ statistics, has an error of 0.11, 6.2 and 1.8%, respectively.This difference is the similarity of the reflective spectrum in some places, which affects the separability and recognition of phenomena and increases the error in estimating the area under cultivation of different crops. The highest and lowest errors in estimating the area under cultivation in the artificial neural network method were in barley and rice crops, respectively, in the support vector machine method were in wheat and rice crops, respectively, and in NDVI index were in wheat and barley crops, respectively. The difference between the cropped area obtained from classification methods and NDVI index with cropped area statistics of Agricultural-Jihad Organization may be due to the following: First, the cultivation history of different has caused problems such as reflections of diverse agricultural lands in one image. Second, the agricultural lands in this area are small. Most of them are under one hectare. Also, the crops in this area are diverse. Third, the smallest region that the image used in the present study can distinguish is about 900 square meters, which is a large number for the agricultural lands of the study area and causes errors.Conclusion The study results showed that the support vector machine method had the lowest error in CAE than the artificial neural network method, which indicates the higher accuracy of the support vector method in identifying and separating crops in the region. Comparing the area obtained from the NDVI index with the statistics of the Agricultural-Jihad Organization of Khuzestan province and evaluating the accuracy of this method indicated the higher efficiency of spectral indices in CAE for the region compared to classification methods. The NDVI index minimizes the error values of the results due to having a threshold and better identification of vegetation density. Therefore, based on the accuracy assessment results and comparing the cropped area with the KOAJ statistics, the utilization of the NDVI index provides the best CAE in the region.