Data-driven behavioural characterization of dry-season groundwater-level variation in Maharashtra, India.

This paper looks at the crucial issue of dry-season groundwater-availability in the state of Maharashtra, India. We look at the two key hydro-climatological measurements which are used to implement groundwater policy in the state, viz., water levels in 5000+ observation wells across the state and aggregate rainfall data. We see that there is substantial variation in groundwater levels within and across the years in most wells. We argue that for a large number of these observation well locations, aggregate rainfall data is inadequate to model or to predict groundwater levels. For this, we use a novel random rainfall coefficient model for the purpose of modelling the effect of rainfall in a composite setting where extraction and changing land-use data is unknown. The observed high variance of this coefficient points to significant variations in groundwater levels, which may only be explained by unmeasured anthropogenic factors. Next, we see that the uncertainty in actual groundwater levels along with scarcity are two distinct features of groundwater availability and will elicit different behaviours from the typical user. Finally, we recommend that quantitative groundwater assessment protocols of the state should move to incorporating data from which extraction and land-use may be modelled. We believe this is one of the first studies where large spatio-temporal scale data gathered by state agencies have been analysed for scientific adequacy. [ABSTRACT FROM AUTHOR]