Faculty Publications
Permanent URI for this communityhttps://idr.nitk.ac.in/handle/123456789/18736
Publications by NITK Faculty
Browse
5 results
Search Results
Item Evaluation of ARIMA, Facebook Prophet and a boosting algorithm framework for monthly precipitation prediction of a semi-arid district of north Karnataka, India(Institute of Electrical and Electronics Engineers Inc., 2022) Chowdari, K.K.; Deb Barma, S.D.; Bhat, N.; Girisha, R.; Gouda, K.C.This study evaluates ARIMA, Facebook Prophet and a new boosting algorithm framework known as ThymeBoost for time series prediction of monthly precipitation of Belagavi district (semi-arid) in Karnataka. The dataset was divided into three periods (1901-2002, 1951- 2002, and 1971 - 2002). The first 70% of the data for each period was applied for training while the rest for testing. Also, the datasets were used in two different forms for both training and testing. In the first set, raw data was used as it is, and the second set of data was used after normalizing the time series using the min-max concept (between 0 and 1). However, the normalized data were de-normalized for each period for performance metrics estimation. ThymeBoost is the best model for the first period of raw data and the second period of normalized data. In contrast, Prophet outperforms all other models for the normalized data in terms of all four measures. For the second period of raw data, no model emerged as the best model in terms of all performance metrics. Therefore, all three models performed similarly for the third period of raw and normalized data. © 2022 IEEE.Item EVALUATION OF ERA5 AND IMERG PRECIPITATION DATA FOR RISK ASSESSMENT OF WATER CYCLE VARIABLES OF A LARGE RIVER BASIN IN SOUTH ASIA USING SATELLITE DATA AND ARCHIMEDEAN COPULAS(Zibeline International Publishing Sdn. Bhd., 2022) Deb Barma, S.D.; Uttarwar, S.B.; Barane, P.; Bhat, N.; Mahesha, A.Precipitation as a major water cycle variable influences the occurrences and distribution of terrestrial water storage change (TWSC), evapotranspiration (ET), and river discharge (Q) of a large river basin. However, its relationship with the other water cycle variables using probabilistic dependence structure concept has not been addressed much. Furthermore, precipitation derived from gauge record is plagued by bias due to orography and under-catch. To fill these gaps, bivariate copula and precipitation derived from reanalysis and satellite data were used. In the present study, the basin-wide averages of the precipitation products APHRODITE, ERA5, and IMERG were used as predictors, whereas the areal mean of MOD16 evapotranspiration, GRACE TWSC, and gauge discharge were used as dependent variables (predictants) for the Brahmaputra basin. The bivariate Archimedean copulas were applied to all the pairs of precipitation-TWSC, precipitation-ET and precipitation-Q based on the optimal marginal distributions obtained. Using the best copula for each pair of the variables, the conditional probability was constructed to predict the predictants for different precipitation amounts (5th, 25th, 50th, 75th, and 95th percentiles). The focus of the analysis was on two scenarios of the predictants (i.e.,≤ 5th and ≥ 95th percentiles). The non-exceedance conditional distribution of TWSC, ET, and Q (all predictants ≤ 5th percentile) decreases with precipitation increase. However, the exceedance probability of the predictants (≥ 95th percentile) increases gradually with an increase in precipitation. The results revealed that both ERA5 and IMERG precipitation data could be used to derive probabilistic measures of the water cycle variables in the absence of gauge-based precipitation. © 2022 by the authors.Item Groundwater level modeling using Augmented Artificial Ecosystem Optimization(Elsevier B.V., 2023) Nguyen, N.; Deb Barma, S.D.; van Lam, T.; Kisi, O.; Mahesha, A.Nature-inspired optimization is an active area of research in the artificial intelligence (AI) field and has recently been adopted in hydrology for the calibration (training) of both process-based and statistical models. This study proposes an improved AI model, Augmented Artificial Ecosystem Optimization-based Multi-Layer Perceptron (AAEO-MLP), to build a monthly groundwater level (GWL) forecasting model. AAEO-MLP model is built on the novel Augmented version of Artificial Ecosystem Optimization and traditional MLP network. In AAEO, Levy-flight trajectory and Gaussian random are utilized in exploration and exploitation to improve the optimizing ability. The AAEO-MLP model is tested on two time-series (1989–2012) datasets collected at two wells in India. Various explanatory variables such as monthly cumulative precipitation, mean temperature, tidal height, and previous measurements of GWL were considered for predicting 1-month ahead of GWL. The performance of AAEO-MLP was benchmarked against 17 different models (original AEO, 3 different variants of AEO, and 13 well-known models) in terms of forecasting accuracy based on six metrics (e.g., mean absolute error, root mean square error, Kling–Gupta efficiency, normalized Nash–Sutcliffe efficiency, Pearson's correlation index, a20 index). Furthermore, convergence analysis and model stability are employed to indicate the effectiveness of AAEO-MLP. The compared results express that the AAEO-MLP is superior to other models in terms of prediction accuracy, convergence, and stability. Overall, the results depict that the AAEO is a promising approach for optimizing machine learning models (e.g., MLP) and should be explored for other hydrological forecasting applications (e.g., streamflow, rainfall) to further assess its strengths over existing methods. © 2022 Elsevier B.V.Item Trends of seasonal and annual rainfall of semi-arid districts of Karnataka, India: application of innovative trend analysis approach(Springer, 2023) Chowdari, K.; Deb Barma, S.D.; Bhat, N.; Girisha, R.; Gouda, K.C.; Mahesha, A.Trend analysis of rainfall is often carried out in water resources management to understand its distribution over a given region. The cumulative seasonal and annual rainfall derived from monthly datasets spanning 102 years (1901–2002) for 11 districts of the semi-arid Karnataka, India, was used for the trend analysis. The two-step homogeneous test approach was carried out on all the time series. Then, lag-1 autocorrelation was conducted only on homogeneous time series. Only 78.18 % of the total time series data were detected as homogeneous, and 95.35% of time series data were found to have insignificant autocorrelation. Then, the Innovative Trend Analysis (ITA) method was applied to 43 homogeneous rainfall time series, as well as to 41 time series using the MK and SR tests, and to two time series using the mMK test. The MK and SR tests detected a significant trend in 14.63% of the time series, while the ITA method was able to detect a trend in 93.02% of the total time series data. The MK and SR tests revealed significant trends in winter and post-monsoon season precipitation for two districts, but only for one district in the case of summer and annual rainfall. No trend was identified for monsoon season precipitation. The mMK test showed a positive trend for the post-monsoon season in a district, while the ITA method revealed significant trends for all seasons in most districts. The sub-trend analysis revealed trends that traditional methods were unable to detect. © 2023, The Author(s), under exclusive licence to Springer-Verlag GmbH Austria, part of Springer Nature.Item ANFIS-based soft computing models for forecasting effective drought index over an arid region of India(IWA Publishing, 2023) Kikon, A.; Dodamani, B.M.; Deb Barma, S.D.; Naganna, S.R.Drought is a natural hazard that is characterized by a low amount of precipitation in a region. In order to evaluate the drought-related issues that cause chaos for human well-being, drought indices have become increasingly important. In this study, the monthly precipitation data from 1964 to 2013 (about 50 years) of the Jodhpur district in the drought-prone Rajasthan state of India was used to derive the effective drought index (EDI). The machine learning models hybridized with evolutionary optimizers such as the genetic algorithm adaptive neuro-fuzzy inference system (GA-ANFIS) and particle swarm optimization ANFIS (PSO-ANFIS) were used in addition to the generalized regression neural network (GRNN) to predict the EDI index. Using the partial autocorrelation function (PACF), models for forecasting the monthly EDI were constructed with 2-, 3- and 5-input combinations to evaluate their outcomes based on various performance indices. The results of the different combination models were compared. With reference to 2-input and 3-input combination models, both GA-ANFIS and PSO-ANFIS show better performance results with R2 ¼ 0.75, while among the models with 5-input combination, GA-ANFIS depicts better performance results compared to other models with R2 ¼ 0.78. The results are presented suitably with the aid of scatter plots, Taylor’s diagram and violin plots. Overall, the GA-ANFIS and PSO-ANFIS models outperformed the GRNN model. © 2023 The Authors.