Improving understanding of climate risks to rice production in South Asia: Historic, present and future linkages between monsoon variability and crop production

Abstract

The Asian summer monsoon is changing due to anthropogenic climate change, threatening the food security and livelihoods of millions of people in South Asia. Crop-climate modelling is a useful tool for understanding how such variability in monsoon climate has impacted - and will continue to impact - crop production. Current modelling approaches use seasonally aggregated weather data, and do not adequately represent nonlinear linkages between weather and crop production. Furthermore, a majority of studies focus on the response of crop production variability to weather impacts, missing a valuable opportunity to assess weather associated with crop failures that have severe and long-lasting consequences for smallholder farmers, regional food security and global market stability. This thesis addresses the limitations of previous work through three main contributions to the literature. In Chapter 3, machine learning models are developed to detect historic linkages between monsoon weather variability and rice production outcomes. This novel application of random forest modelling in the Indian Indo-Gangetic plains is demonstrated to effectively capture crop-climate variability linkages, improving on prior parametric regression modelling studies. Critically, a comparison between yield-response models and cropped area-response models reveals that changes in additional harvested production areas may exacerbate losses resulting from extreme weather events. Downwelling shortwave radiation flux is identified as the most important weather driver of yield variability. The predictor variables monsoon onset and season length included here are novel to statistical crop-climate modelling. Both variables are important in explaining crop yield variability, and their nonlinear responses match biophysical expectations (described in Chapter 2). In Chapter 4, the modelling framework presented in Chapter 3 is further developed to identify drivers of historical crop production failure. Here, the shift from a yield-focus to production-focused modelling was directly driven by the findings in Chapter 3. Moreover, the use of seasonally aggregated weather data was identified as a key limitation of existing approaches, with more work required to understand the consequences of intraseasonal timing of weather extremes during the rice growing season. Machine learning models effectively capture linkages between intraseasonal weather variability and rice production failures across a range of failure severities. In particular, late-season temperature and mid- to late-season precipitation were key drivers of production failures, differing from the results in Chapter 3 that captured weather factors responsible for overall yield and production variability. In Chapter 5, the machine learning models trained on historical data in Chapter 4 are expanded to project future rice production failure probabilities under climate change. Future failure probabilities are projected using down-scaled CMIP6 climate projections, addressing a limitation of coarse spatial scales of climate model data used in existing studies. Similarly, intraseasonal weather predictors are used in these models (as in Chapter 4) to address the limitation of coarse temporal scales of prior work assessing future crop production impacts. The nonlinear relationships built within these historical models reveal that under projected climate change, Indian rice production will experience greater exposure to high-risk temperatures and lesser exposure to high-risk precipitation levels. In addition, all-India modal production failure probability is projected to increase across all SSPs, with a greater occurrence of more extreme failure probabilities. Crucially, the relatively high spatial resolution of this modelling approach reveals spatial heterogeneity in future rice production failure risk across India. Finally, an assessment of future irrigation adaptation strategies demonstrates the spatial heterogeneity of res

Date of Award	1 Aug 2024
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Timothy Foster (Supervisor) & Ben Parkes (Supervisor)