Classifier comparison machine learning-MNlogit with iso-mean for agriculture, forestry, and other land use

Rapid assessment of land-use land-cover (LULC) is important for resource planning and management. One of the core functions of the Food and Agriculture Organization (FAO) is to collect and disseminate information related to the subject areas covered by the organization, and amongst these natural resources, land use, agriculture and forests stand important. This is done by collection of information through different reporting processes and its compilation which takes two to three years. The reliability, scalability and definition of classification are major issues to produce statistics on Agriculture, Forestry, and Other Land Use (AFOLU). The ‘Earth Observation System’ provides inputs for mapping such details at spatial and temporal scales efficiently and effectively. This paper demonstrates utility of Moderate Resolution Imaging Spectroradiometer (MODIS) vegetation index at 16 days of revisit for AFOLU classification for the South Asian Association for Regional Cooperation (SAARC) countries. We compared the Machine Learning based classifier Multi-nominal logistic regression (MN logit) with Iterative Self-Organizing (ISO) Mean Classifier based algorithm for classification and assessment of AFOLU for SAARC countries with more that 80% and 55% accuracy and provide the option for random accuracy assessments for end users. We find Machine Learning based Mnlogit based classification reports minimized error of resources inventory and increased reliability and global uniformity.