Global Climate Models (GCMs) are mathematical models to understand and predict the Earth’s climate by projecting the real-world processes over time. These simulation tools help to predict future climate variables that will be useful to develop sustainable long, medium and short-term water resource planning strategies. A new study by a team of scientists - Prof. D. Nagesh Kumar from the Indian Institute of Science, Bangalore and Prof. K. Srinivasa Raju from BITS-Pilani, Hyderabad campus, has analyzed numerous available GCMs to choose the best that would be applicable in the Indian context. Such analysis helps in developing the best resource planning strategies and the best climate models that can be used for localized needs.
Clustering is a method of grouping a set of objects or data points according to similarity, and then extracting information from this data set to a logical structure. The researchers employed various statistical tools to group 36 GCMs from Coupled Model Intercomparison Project Phase 5 (CMIP5) database with reference to maximum temperature (MAXT), minimum temperature (MINT) and the combination of maximum and minimum temperature (COMBT) in India. CMIP5 aims to provide a framework for how to conduct coordinated climate change experiments during the next five years. Assessment of these temperature related variables is essential because they are instrumental in managing the hydrological cycle.
Clustering of GCMs is important because most of them use similar numerical schemes, parameterizations, or are developed by the same agency. “Cluster analysis helped to suggest representative GCM instead of a number of GCMs which are similar,” explains Prof. Raju. Clustering of data is done by K-Means clustering followed by obtaining the optimal clusters of GCMs wherein the GCMs are optimally divided between the clusters. K-means clustering is a statistical method to partition ‘n’ number of observations into ‘k’ number of clusters based on attributes of the observation. Each observation is allotted to a cluster that has a mean value closer to the observation.
The researchers then applied cluster validation methods like Davies–Bouldin Index (DBI) and F-statistic to measure the quality of the chosen group and determine the optimal number of clusters of GCMs. They observed that most of the GCMs are similar and found that the optimum cluster was 3 for MAXT and 2 each for MINT and COMBT. Based on these, an ensemble of GCMs was suggested.
In India, climate studies based on temperature are lesser than studies that rely on precipitation patterns. This study is the first to employ K-Means cluster analysis for choosing an ensemble of Global Climate Models for Indian climate. “The present methodology, in our opinion is the first of its kind where global climate models were clustered seamlessly using twin modeling structures: First modeling structure is based on K-Means cluster analysis to cluster the GCMs whereas second modeling structure is related to Cluster validation techniques such as Davies- Bouldin index and F-Statistic to determine optimal clusters of GCMs,” says Prof. Nagesh Kumar. The ease with which the methodology can be applied and replicated makes this important study a pivot for future endeavors on climate impact assessment studies.
Water scarcity in various parts of the country, especially for clean, drinking water, confirms the need for efficient hydrological modeling applications like rainfall-runoff modeling studies. The inferences drawn from this study will make a remarkable turnover in the current water conservation strategies. “The ensemble of Global Climate Models chosen for different situations can help in efficient reservoir design, drinking water requirements, etc. which helps the public as a whole”, concludes Prof. Nagesh Kumar.