IUP Publications Online
Home About IUP Magazines Journals Books Archives
     
Recommend    |    Subscriber Services    |    Feedback    |     Subscribe Online
 
The IUP Journal of Computer Sciences :
Assessing Climate Changes in California Using Support Vector Machine in Statistical Downscaling
:
:
:
:
:
:
:
:
:
 
 
 
 
 
 
 

General Circulation Model (GCM) is used for assessing the climate change for a large river basin by dividing the whole earth into grids and solving partial differential equations which is computationally difficult. For assessing smaller river basin hydrology, statistical downscaling is applied which is based on obtaining a relationship between GCM’s simulated climatic variables and the streamflow over the region (Ghosh and Mujumdar, 2008). This paper presents a methodology for assessing the climatic changes over California Region in the winter months (October to March) using various models such as linear regression, Artificial Neural Networks (ANN) and Support Vector Machines (SVM) and choosing the most appropriate model for predicting the future streamflow of the winter months and predicting the changes in flooding patterns of California at various stations. The National Center for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) data is used as the predictor variables and the result obtained is compared with the actual values of streamflow from the Hydro-Climatic Data Network (HCDN) data collected over the years 1948-1988, which is the historical period of the station of our interest.

A decreasing trend is observed for the future period of 2020-2050 than the historical patterns, which reveals the decrease patterns of flooding’s over California. These predicted results using SVM model are compared with the Variable Infiltration Capacity (VIC) model (uses land variables such as runoff and soil properties along with atmospheric variables as the predictor variables) by plotting a CDF graph, and the accuracy of the model is compared with other models based on that, and if a more computationally easier and efficient model could be used for predicting better results is assessed.

 
 
 

Assessing climate changes over a region involves using the atmospheric variables simulated by the General Circulation Model (GCM) and predicting the streamflow, but this method works well on a coarser scale and a larger spatial scale thus able to simulate the time series of the climatic variables globally, but performs poorly on a smaller scale (<103 km). The poor performance of these GCM models on a smaller scale has led to the development of Limited Area Model (LAM) which involves dynamic downscaling where the fine smaller grids of a smaller region are nested within the coarser grid of the GCM (Ghosh and Mujumdar, 2008)# and the corners of those grid are served as the boundary conditions for solving the partial differential equations to obtain its solution and getting the predictands.
The method is computationally effective but involves a lot of computational cost and time and complicated design. Another approach to downscaling is the statistical downscaling method which converts the large-scale climatic variables simulated by the GCM into a smaller regional climatic variables and thus establishing a statistical relationship among the predictor climatic variables and the predictand by training our model to predict certain dependent variables such as precipitation or streamflow for a smaller region (Ghosh and Mujumdar, 2006; and Dettinger Michael, 2011). There are three implicit assumptions involved in statistical downscaling. Firstly, the predictors are variables of relevance and are realistically modeled by the host GCM. Secondly, the empirical relationship is valid also under altered climatic conditions. Thirdly, the predictors employed fully represent the climate change signal (G-M).
The most popular approach of downscaling is the use of transfer function which is a regression-based downscaling method that relies on direct quantitative relationship between the local scale climate variable (predictand) and the variables containing the large-scale climate information (predictors) through some form of regression (G-M). Linear and nonlinear regression, Artificial Neural Network (ANN) etc. have been used to derive predictor-predictand relationship. Among them, ANN-based downscaling techniques have gained wide recognition owing to their ability to capture nonlinear relationships between predictors and predictand.

 
 
 

Computer Sciences IUP Journal ,General Circulation Model (GCM), Statistical downscaling, Support Vector Machine (SVM), Streamflow, Artificial Neural Network (ANN), Variable Infiltration Capacity (VIC)