Identifying opportunities to improve digital soil mapping in India: A systematic review

Abstract

Soaring food demand, population pressure, land degradation, small size of agricultural land holdings, and diversified soil types in India require advanced digital soil mapping (DSM) for sustainable land management. This paper systematically reviews the common trends of SCORPAN based DSM in India to identify the important research gaps and opportunities to improve in future. A systematic literature search from 2000 to October 2021 has yielded 35 numbers of peer reviewed articles, which have performed DSM in India following the SCORPAN approach. The increased number of published articles from 2017 onwards suggests that there is a growing interest for DSM in India over the past few years. However, only two articles have prepared digital soil maps at the national extent. Moreover, the local and regional extent DSM are being limited to only a few parts of the country. There still remains 50% of the states and Union Territories of the country where no DSM studies have been performed so far except the national and global level interventions. Among the target variables, soil carbon related attributes have been predicted most frequently, whereas soil classes have been rarely predicted. Environmental covariates representing organism (O) and relief (R) have been widely included for DSM, whereas the use of other covariates has been limited. Among different machine learning (ML) algorithms, regression kriging has been adopted most frequently followed by random forest and quantile regression forest. Most articles have adopted data splitting (76%) as the model and map evaluation approach, whereas independent validation has been limited to only 5% of the articles. Only 34% of the articles have presented the uncertainty maps. Major research gaps identified by this review include lack of standardized digital soil databases, poor sampling density, coarse resolution, limited use of environmental covariates, insufficient comparative studies among ML algorithms, inadequate independent validation, and undersupply of uncertainty maps. Key evidences from this review will be helpful for improving future DSM activities by scientists and practitioners involved with DSM in India and abroad.