Deriving Regional IDF Curves for Data-Sparse Areas in Climate Change Scenarios using Merged Satellite and Ground-based Precipitation and GCMs

Rainfall Intensity-Duration-Frequency (IDF) curves are widely used in studies related to planning, design, and operation of various water control (e.g., barrages, dams, levees) and conveyance structures (e.g., culverts, spillways, storm sewers) for mitigating risk associated with floods attributable to extreme precipitation. In many parts of the globe, precipitation data are limited, and the network of gauges is sparsely distributed. Therefore, the use of only at-site data for the construction of IDF curves could have large uncertainties. To overcome this impediment, regional IDF relationships could be developed by regional frequency analyses (RFA) which uses information pooled from several meteorologically similar sites. Recently, there is growth in the use of fine spatial scale remote-sensing precipitation products to arrive at IDF relationships for ungauged locations, as the spatial coverage of these products is exhaustive. However, recent studies indicate that most of the remote sensing products underestimate the precipitation intensities corresponding to different durations and return periods and also perform worse at shorter time scales (e.g., daily and sub-daily). Although remote sensing products can be corrected for biases before use in developing the IDF relationships, there is ambiguity in the choice of bias correction methods. Furthermore, in sparsely gauged locations, the availability of only a limited number of ground observation stations for bias correction enhances uncertainty in the developed IDF relationships. In addition, relying on only one satellite product may not be meaningful, as the skill of different satellite products varies across the globe. Also, the conventional practice of developing IDF curves considering the stationary assumption may lead to large biases in estimates of precipitation extremes in a changing climate. To address these issues, this study proposes a novel methodology to develop non-stationary regional IDF relationships for use in climate change scenarios. The methodology involves nonstationary RFA utilizing fine grid-scale daily precipitation derived by merging multiple satellite-based precipitation products and ground-based precipitation products for homogenous extreme precipitation regions (EPRs). The merging of different products is achieved using a novel random forest-based regression method. Effectiveness of the proposed methodology is demonstrated through a case study on Karnataka state in India, which extends over approximately 0.2 million square kilometers. The homogenous EPRs are delineated in the study area using ensemble cluster analysis of the relevant predictor variables/covariates. Non-stationary regional IDF curves are developed using the proposed methodology corresponding to different CMIP6 climate change scenarios, considering an ensemble mean precipitation derived from eleven GCMs (General Circulation Models). The curves are compared with those obtained using conventional stationary methods considering block-maxima and partial duration series of extreme precipitation.