Downscaling precipitation to river basin in India for IPCC SRES scenarios using support vector machine

Abstract

This paper presents a methodology to downscale monthly precipitation to river basin scale in Indian context for special report of emission scenarios (SRES) using Support Vector Machine (SVM). In the methodology presented, probable predictor variables are extracted from (1) the National Center for Environmental Prediction (NCEP) reanalysis data set for the period 1971-2000 and (2) the simulations from the third generation Canadian general circulation model (CGCM3) for SRES emission scenarios A1B, A2, B1 and COMMIT for the period 1971-2100. These variables include both the thermodynamic and dynamic parameters and those which have a physically meaningful relationship with the precipitation. The NCEP variables which are realistically simulated by CGCM3 are chosen as potential predictors for seasonal stratification. The seasonal stratification involves identification of (1) the past wet and dry seasons through classification of the NCEP data on potential predictors into two clusters by the use of K-means clustering algorithm and (2) the future wet and dry seasons through classification of the CGCM3 data on potential predictors into two clusters by the use of nearest neighbour rule. Subsequently, a separate downscaling model is developed for each season to capture the relationship between the predictor variables and the predictand. For downscaling precipitation, the predictand is chosen as monthly Thiessen weighted precipitation for the river basin, whereas potential predictors are chosen as the NCEP variables which are correlated to the precipitation and are also realistically simulated by CGCM3. Implementation of the methodology presented is demonstrated by application to Malaprabha reservoir catchment in India which is considered to be a climatically sensitive region. The CGCM3 simulations are run through the calibrated and validated SVM downscaling model to obtain future projections of predictand for each of the four emission scenarios considered. The results show that the precipitation is projected to increase in future for almost all the scenarios considered. The projected increase in precipitation is high for A2 scenario, whereas it is least for COMMIT scenario.

Evaluating a modified point-based method to downscale cell-based climate variable data to high-resolution grids

“…We have already shown HRCD's downscaling success, and expect that aggregating accurate fine-scale values would give more accurate estimates of coarse-scale values than estimates based on less fine-scale data (Riley et al, 2009;Di Vittorio and Miller, in review). Furthermore, coarse-scale model outputs generally have biases often resulting from lack of finer-scale information, which has led to a variety of bias-correction techniques for statistical downscaling of precipitation (Anandhi et al 2008;Boé et al 2007). HRCD uses fine-scale topographic information to compensate for such biases by reducing scale mismatch between coarse-scale input and fine-scale output.…”

“…Many applications and research disciplines, including ecosystem modeling (Di Vittorio et al 2010; Kucharik et al 2000;Miguez et al 2009), hydrologic assessment (Anandhi et al 2008), land conservation (Bayliss et al 2005;Cabeza et al 2010;Galatowitsch et al 2009), and regional planning and decision-making (Girvetz et al 2008), require high spatial resolution, gridded input climate data with daily or hourly frequency. Ecosystem models generally have daily (Di Vittorio et al 2010;Running and Coughlan 1988;Thornton and Rosenbloom 2005) or sub-daily (Kucharik et al 2000;Miguez et al 2009) time steps, and many regional applications need high spatial resolution with length scales less than 1 km.…”

“…RCMs are more robust than statistical methods, but to date they do not give significantly better results for temperature and precipitation and are considered too expensive for operational use (Salathe et al 2007;Spak et al 2007). Statistical methods are computationally inexpensive, assume stationarity in predictor-predictand relationships, require robust relationships among sufficient data, are vulnerable to data error and omitted dependencies, and generally focus on precipitation and/or temperature (Anandhi et al 2008;Boé et al 2007;Charles et al 1999;Cheng et al 2008;Landman and Tennant 2000;Maurer et al 2010;Schubert 1998;Trigo and Palutikof 2001;Wilby et al 1998;Zhang 2005). These statistical methods rely on relationships between coarsely gridded reanalysis or Global Circulation Model (GCM) data, and point or fine-scale gridded observations.…”

Effects of variance adjustment techniques and time-invariant transfer functions on heat wave duration indices and other metrics derived from downscaled time-series. Study case: Montreal, Canada

“…Anandhi et al 2008;Ahrens 2003;Chen et al 2011), biologists (e.g. Cossarini et al 2008Bucklin et al 2012;Saba et al 2012), agronomists, economists (e.g.…”

Spatial Downscaling of Grid Precipitation Using Support Vector Machine Regression

“…연구지역의 고도 가 크게 높지 않은 경우 강우에 대한 지형적 특성의 고려는 배재된 경우가 많으며, 이는 습도 및 대기 중 가능강수 수 분량 등 강우와 보다 직접적 상관관계를 가지는 인자들에 대한 고려로 보완되었다 (Beckmann and Buishand, 2002;Charles et al,. 1999;Cavazos, 1999 (Bardossy, 1997;Anandhi et al, 2008;Chu et al, 2008;Olsson et al, 2001). 특히 SVM 회귀 방법을 다른 회귀 방법과 함께 적용하여 비교한 연구의 경우 공통 적으로 SVM을 사용한 예측 값이 다른 방법에 비해 전반적 으로 우수한 것으로 나타났다.…”