Weather shocks and rice (Oryza sativa) yield response to fertilizer: Representative field-level evidence from Bangladesh

The fertilizer response of yield has been one of the major indicators of agricultural productivity in both developed and developing countries. Filling the evidence gap remains vital regarding fertilizer response in South Asia, given the emergence of intensifying weather shocks. Nationally representative evidence at field levels reflecting farmers’ actual production environments is particularly scarce. We fill this knowledge gap by using three rounds of nationally representative panel data of farm households with plot-level rice (Oryza sativa) production information and assessing how the shapes of response functions are affected by shocks in temperatures, droughts, and rainfall, using common yield response functions including both quadratic function and stochastic linear response plateau (LRP). Notably, in the stochastic LRP model, we find that one standard deviation (1SD) increases in the percentiles of growing degree days (GDD) and high nighttime temperature (HNT) relative to their historical distributions reduce sub-plateau yield response by 50% or more and yield plateau by up to 0.4 t/ha in Boro and Aman irrigated system. In the Aman rainfed system, 1SD increases in GDD and HNT percentiles reduce sub-plateau linear responses by roughly 30%. Similarly, 1SD increases in drought severity and decreases in rainfall shift down the overall linear response function by 0.1–0.2 t/ha and yield plateau by about 0.1 t/ha. Furthermore, results for stochastic LRP are also consistent for both maximum likelihood estimation of Maddala–Nelson Switching Regression, as well as Bayesian regression models in which researchers’ prior beliefs are updated by posterior information obtained from the data based on the Bayes’ rules.