Creating a taro yield emulator and using it to study climate change impact on taro in Papua New Guinea

This paper develops a taro yield emulator to assess climate change impacts on taro production in Papua New Guinea (PNG). The emulator uses a fixed-effects polynomial regression trained on DSSAT simulations to capture non-linear, stage-specific responses of taro yields to rainfall and temperature. Historical MSWX weather and future climates from CMIP6/ISIMIP3b and a large MIT-IGSM/AgERA5 sample are used to drive the emulator and assess changes in average yields, yield variability, and the frequency of low-yield years across PNG. Our analysis shows that taro appears to be sensitive to both low- and high-rainfall levels during particular phases of growth and is sensitive to heat during particular phases, as well. Our visit to farmers fields in the Markham Valley along with discussions with experts at GrowPNG in Lae, NARI, and PNG University of Technology in Lae alerted us to the fact that drought was a serious problem for taro, but the model revealed the additional issues with high rainfall and high temperature, along with the additional information about the months that were particularly important for determining the degree of damage that adverse weather might do to yields. National-average yield changes by mid-century under high-emissions scenarios are projected to be small based on the median projections across climate models, however one of the climate models projects a much larger temperature increase relative to the other models, and for that climate model, the yield reduction is around 9% on average.