Open Collections

Sidhu, Balsher Singh

University of British Columbia

Predicting crop yield response to climate change is a topic of active research. A popular method involves building statistical models using historical climate and agricultural data, and then applying them on future climate projections for predicting crop yields. Using India as a case study, this dissertation examines these statistical models along two dimensions: the type of climate variables included, and the statistical techniques used. We also employ these models for predicting climate change impact on Indian crop yields till 2100. First, we examine the role of seasonal (e.g. total seasonal precipitation) versus subseasonal (e.g. precipitation over each crop growing stage) climate variables in explaining crop yields. We observe that even though adding extra climate variables does not always improve overall model accuracy, the proportion of yield variability explained by climate (versus non-climatic variables like geography and time) can increase significantly. This underscores the importance of combining physiological and statistical knowledge while choosing climate variables for statistical crop models. Second, we compare the well-known statistical method of OLS linear regression (LR) to a popular machine learning method called boosted regression trees (BRTs). While LR models were simpler to interpret, BRTs could uncover unexpected non-linear relationships and exhibited better yield prediction accuracy. Compared to LR, BRTs sometimes showed lower sensitivity to temperature variation. Higher flexibility of BRTs allowed them to identify obscure interactions between variables that could be missed by LR. We then use different climate variables and statistical techniques for building statistical models to predict climate change impact on India’s future crop yields. We found that nationally-averaged rice, wheat, and pearl millet yields could reduce by up to 3.4, 4.3, and 5.5 percent (respectively) by 2050 under the intermediate emissions scenario. Some parts of India may benefit from climate change, while other regions could face yield losses of up to 20 percent. Depending on the climate variables or statistical technique employed, we observe high variability in yield change predictions. We therefore suggest combining multiple models for estimating climate change impact on crop yields.

Text

2021-11-03

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/80145

Resources, Environment and Sustainability

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/