Faculty Publications
Permanent URI for this communityhttps://idr.nitk.ac.in/handle/123456789/18736
Publications by NITK Faculty
Browse
4 results
Search Results
Item An extreme learning machine approach for modeling evapotranspiration using extrinsic inputs(Elsevier B.V., 2016) Patil, A.P.; Deka, P.C.Precise estimation of evapotranspiration is crucial for accurate crop-water estimation. Recently machine learning (ML) techniques like artificial neural network (ANN) are being widely used for modeling the process of evapotranspiration. However, ANN faces issues like trapping in local minima, slow learning and tuning of meta-parameters. In this study an improved extreme learning machine (ELM) algorithm was used to estimate weekly reference crop evapotranspiration (ETo). The study was carried out for Jodhpur and Pali meteorological weather stations located in the Thar Desert, India. The study evaluated the performance of three different input combinations. The first input combination used locally available maximum and minimum air temperature data while the second and third combination used ETo values from another station (extrinsic inputs) along with the locally available temperature data as inputs. The performance of ELM models was compared with the empirical Hargreaves equation, ANN and least-square support vector machine (LS-SVM) models. Root mean squared error (RMSE), Nash-Sutcliffe model efficiency coefficient (NSE) and threshold statistics (TS) were used for comparing the performance of the models. The performance of ELM model was found to be better than the Hargreaves and ANN model. The LS-SVM and ELM displayed similar performance. ELM3 models, with 36 and 33 neurons in hidden layer were found to be the best models (RMSE of 0.43 for Jodhpur and 0.33 for Pali station) for estimating weekly ETo at Jodhpur and Pali stations respectively. The results showed that ELM is a simple yet efficient algorithm which exhibited good performance; hence, can be recommended for estimating weekly ETo. Furthermore, it was also found that use of ETo values from another station can help in improving the efficiency of ML models in limited data scenario. © 2016 Elsevier B.V.Item Hybrid wavelet packet machine learning approaches for drought modeling(Springer, 2020) Das, P.; Naganna, S.R.; Deka, P.C.; Pushparaj, J.Among all the natural disasters, drought has the most catastrophic encroachment on the surrounding and environment. Gulbarga, one of the semi-arid districts of Karnataka state, India receives about 700 mm of average annual rainfall and is drought inclined. In this study, the forecasting of drought for the district has been carried out for a lead time of 1 month and 6 months. The multi-temporal Standardized Precipitation Index (SPI) has been used as the drought quantifying parameter due to the fact that it is calculated on the basis of one simplest parameter, i.e., rainfall and additionally due to its ease of use. The fine resolution daily gridded precipitation data (0.25º × 0.25º) procured from Indian Meteorological Department (IMD) of 21 grid locations within the study area have been used for the analysis. Forecasting of drought plays a significant role in drought preparedness and mitigation plans. With the advent of machine learning (ML) techniques over the past few decades, forecasting of any hydrologic event has become easier and more accurate. However, the use of these techniques for drought forecasting is still obscure. In this study, Artificial Neural Network (ANN) and Support Vector Regression (SVR) techniques have been employed to examine their accuracy in drought forecasting over shorter and longer lead times. Furthermore, two hybrid approaches have been formulated by coupling a data transformation method with each of the aforementioned ML approaches. At the outset, pre-processing of input data (i.e., SPI) has been carried out using Wavelet Packet Transform (WPT) and then used as inputs to ANN and SVR models to induce hybrid WP-ANN and WP-SVR models. The performance of the hybrid models has been evaluated based on the statistical indices such as R2 (co-efficient of determination), RMSE (Root Mean Square Error), and MAE (Mean Absolute Error). The results showed that the hybrid techniques have better forecast performance than the standalone machine learning approaches. Hybrid WP-ANN model performed relatively better than WP-SVR model for most of the grid locations. Also, the forecasting results deteriorated as the lead time increased from 1 to 6 months. © 2020, Springer-Verlag GmbH Germany, part of Springer Nature.Item Spatial and temporal variations in river water quality of the Middle Ganga Basin using unsupervised machine learning techniques(Springer Science and Business Media Deutschland GmbH, 2020) Krishnaraj, A.; Deka, P.C.In this study, cluster analysis (CA), principal component analysis (PCA) and correlation were applied to access the river water quality status and to understand spatiotemporal patterns in the Ganga River Basin, Uttara Pradesh. The study was carried out using data collected over 12 years (2005–2017) regarding 20 water quality parameters (WQPs) covering spatially from upstream to downstream Ankinghat to Chopan, respectively (20 stations under CWC Middle Ganga Basin). The temporal variations of river water quality were established using the Spearman non-parametric correlation coefficient test (Spearman R). The highest Spearman R (?0.866) was observed for temperature with the season and a very significant p value of (0.0000). The parameters EC, pH, TDS, T, Ca, Cl, HCO3, Mg, NO2 + NO3, SiO2 and DO had a significant correlation with the season (p < 0. 05). K-means clustering algorithm grouped the stations into four different clusters in dry and wet seasons. Based on these clusters, box and whisker plots were generated to study individual clusters in different seasons. The spatial patterns of river WQ on both seasons were examined. PCA was applied to screen out the most significant water quality parameters due to spatial and seasonal variations out of a large data set. It is a data reduction process and a more conventional way of speeding up any machine learning algorithms. A reduced number of three principal components (PCs) were drawn for 20 WQPs with an explained total variance of 75.84% and 80.57% is observed in the dry and wet season, respectively. The parameters DO, EC_ Gen, P-Tot, SO4 are the most dominating parameters with PC score more than 0.8 in the dry season; similarly, TDS, K, COD, Cl, Na, SiO2 in the wet season. The different components of water quality monitoring, such as spatiotemporal patterns, scrutinize the most relevant water quality parameters and monitoring stations are well addressed in this study and could be used for the better management of the Ganga River Basin. © 2020, Springer Nature Switzerland AG.Item Performance enhancement of SVM model using discrete wavelet transform for daily streamflow forecasting(Springer Science and Business Media Deutschland GmbH, 2021) Kambalimath S, S.; Deka, P.C.Streamflow modeling becomes a vital task in any hydrological study for an improved planning and management of water resources. Soft computing and machine learning techniques are becoming popular day by day for their predictive capability when limited input data are available. In the present study, Support Vector Machine (SVM) technique is applied to forecast 1-day, 3-day, and 5-day ahead streamflow using daily streamflow time-series of Khanapur, Cholachguda, and Navalgund gauging stations in Malaprabha sub-basin located in the Karnataka state of India. Furthermore, Discrete Wavelet Transform is used as a data pre-processing method to evaluate the performance enhancement of SVM model, for which four different mother wavelet functions are used and tested separately, namely, Haar, Daubechies, Coiflets, and Symlets. Models are evaluated using coefficient of determination (R2), root-mean-square error, and Nash–Sutcliffe efficiency. The study indicates that the performance of SVM model improves considerably when wavelet method is coupled. It is found that the R2 values for Khanapur station using SVM are 0.91, 0.66, and 0.46 for 1-day, 3-day, and 5-day lead-time forecasts, respectively. However, when wavelet method is coupled with SVM model, the R2 is improved to 0.99, 0.73, and 0.68 for 1-day, 3-day, and 5-day lead-time forecasts, respectively. © 2021, The Author(s), under exclusive licence to Springer-Verlag GmbH, DE part of Springer Nature.