Extending SC-PDSI-PM with neural network regression using GLDAS data and Permutation Feature Importance. (November 2022)

• Record Type: Journal Article
• Authors: Ramirez, Saul G.; Hales, Riley Chad; Williams, Gustavious P.; Jones, Norman L.
• Is Part Of: Environmental modelling & software. Volume 157(2022)
• Publication Date: 2022-11
• Subjects: PDSI -- Drought -- Machine learning -- GLDAS -- Time series regression -- Data extension; Environmental monitoring -- Computer programs -- Periodicals; Ecology -- Computer simulation -- Periodicals; Digital computer simulation -- Periodicals; Computer software -- Periodicals; Environmental Monitoring -- Periodicals; Computer Simulation -- Periodicals; Environnement -- Surveillance -- Logiciels -- Périodiques; Écologie -- Simulation, Méthodes de -- Périodiques; Simulation par ordinateur -- Périodiques; Logiciels -- Périodiques; Computer software; Digital computer simulation; Ecology -- Computer simulation; Environmental monitoring -- Computer programs; Periodicals; Electronic journals; 363.70015118
• DOI: 10.1016/j.envsoft.2022.105475
• Languages: English
• ISSNs: 1364-8152

Abstract

Abstract: The Palmer Drought Severity Index (PDSI) ranges from −10 to 10 and is used for monitoring drought extent and severity. PDSI is a monthly global gridded data set with partial global coverage from 1850 through 1947 and full global coverage from 1948 through 2018. PDSI updates are infrequent. We present a method to extend PDSI using Global Land Data Assimilation System (GLDAS) data. We provide an updated dataset and code for the method. We discuss the accuracy and bias of the method for various regions. We have high accuracy, with 99.5% of the globe exhibiting RMSE values less than 1. Globally our method is unbiased with an average ME of approximately 0. Some regions have slight biases with dryer and wetter regions showing a slight negative and positive biases, respectively. Prediction errors exhibits spatial trends with the highest errors in areas with extreme climate. Editor highlights: PDSI data are infrequently updated. We develop a machine learning method to regress the GLDAS model data to impute missing PDSI data. We use regression, rather than using interpolation or extrapolation methods, which allows imputation of large gaps. RMSE is low and global mean error is approximately 0 indicating no global bias. Errors exhibit spatial trends with the highest errors in areas with extreme climate.