Method for dataset preparation for soil data analysis in decision support applications

Abstract

Nowadays, publically available soil health cards (SHC) are easily understandable and useful means of knowledge transfer about soil conditions to farmers for sustainable soil management in farms. The SHCs are the true representative of precise and useful information about the physicochemical properties and nutrient status of agricultural soil. Before SHCs, the soil test records generally lie with soil test laboratories (STLs) and their availability in form of soil datasets is very limited. Moreover, the datasets are not big enough to be utilized as test datasets in soil investigation and development of futuristic and intelligent decision support and advisory systems. To address the issue, this paper presents a simple method for soil dataset preparation from SHCs for machine learning (ML) and artificial intelligence (AI) based soil research and analysis applications. The method collects wide number of SHCs for particular location and applies simple parsing and data cleaning concepts, implemented by utilizing open-source python libraries, to set of SHCs to prepare soil dataset. The dataset prepared by the proposed method can be used in a variety of intelligent data-driven decision support applications such as soil classification, nutrient prediction, and fertilizer recommendation. The dataset thus generated by the proposed method would help both researchers and soil scientists in their future research on soil health monitoring and decision support for sustainable farm management.