Comparative geospatial approach for agricultural crops identification in inter- fluvial plain- A case study of Sahiwal district, Pakistan

Agricultural crop cover identification is a major issue and time-consuming effort to verify the crop type through surveys of the individual field or using prehistoric methods. To establish the scenario of crop identification, the stage of crop provides diverse spatial information about the variety of crops due to its spectral changes. The main aim of this study was to the identify the crop types and their behavior using remote sensing and geographical information system-based approach. Moreover, two main methods were applied to the Sentinel-2 satellite data in which one is random forest based supervised classification and another was Normalize Difference Vegetation Index (NDVI) density estimation method through the google earth engine to procure the data in time-efficient way. This study also established the comparison between classified and vegetation index based seasonal compositional datasets for wheat, cotton, maize, and fodder crops. Study discussed the best fit technique for crops identification in the light of observed methods. Furthermore, the vegetation index ranges by the zonal statistics of the field samples were established according to crop precision. Results showed that-22.94,-43.72, 20.61, and 32.49 % dissimilarities existed in wheat, fodder, cotton, and maize results respectively, after comparison of both techniques. Although, the accuracy assessment was performed on the classified dataset for validation of results by confusion matrix accuracy assessment process using field sample data. Moreover, the vegetation index was used to evaluate crop land surface temperature to estimate the crop growth stage valuation that revealed noticeably enthralling outcomes. The results determined that the classified accuracies of wheat, cotton, maize and fodder were 84, 80, 81 and 71 % respectively. This study also revealed that the random forest classifier has used more features and information potentially during the classifier trainings but vegetation index just implies the limited number of features such as crop growing status.