Optimizing LSTM and Bi-LSTM models for crop yield prediction and comparison of their performance with traditional machine learning techniques
by V. Kiran Kumar, K. V. Ramesh & V. Rakesh
Advance prediction of crop yield is very critical in the context of ensuring food security as the region specific challenges in social and environmental conditions often infringe plan of policy makers. This study presents a generic methodology to configure and fine tune the state-of-the-art Long Short-Term Memory (LSTM) based Deep Learning (DL) model through hyperparameter optimization for prediction of yield (annual crop production) in Wheat, Groundnut and Barely over India based on multiple independent input variables identified using multicollinearity test. The Monte Carlo cross-validation method is used to validate the optimized LSTM models. Results from the LSTM model tuning showed that among the 4 optimizers tested, Adam was found to perform better irrespective of the crop and Bi-LSTM outperformed sLSTM in terms of prediction accuracy. The percentage reduction in error with Bi-LSTM compared to sLSTM in predicting wheat and groundnut crop yield was 39% and 13% respectively while in case of barley crop, error reduction was marginal (0.34%). The performance of optimized Bi-LSTM model is compared with the performance of traditional machine learning (ML) models such as support vector regression (SVR) and SVR polynomial {2nd and 3rd order}, Auto Regressive Integrated Moving Average (ARIMA) and ARIMAX (ARIMA with exogenous variables) and Vector Auto-regression (VAR). The Bi-LSTM model is found to be superior to ML models; the percentage reduction in mean absolute scaled error with the Bi-LSTM compared to the best performing ML model was 94%, 72%, and 71% in predicting wheat, groundnut and barley yield respectively. This study showed that by choosing proper explanatory (independent) variable and hyperparameter optimization, a simple (single layer) structure of deep neural network (LSTM) outperformed traditional ML models in terms of accuracy for crop yield prediction application.
Role of ocean–atmosphere interactions on contrasting characteristics of two cyclones over the Arabian Sea
by P. Ipsita, V. Rakesh & G. N. Mohapatra
The role of ocean–atmosphere interaction in modulating the track and intensity of two cyclonic storms, Nisarga and Nanauk, which originated from almost similar locations in June in the Arabian Sea with completely different tracks, is investigated in this study. Sea surface temperature (SST), steering flow, relative vorticity, latent heat flux (LHF), specific humidity, vertical wind shear (VWS), outgoing longwave radiation (OLR), convective available potential energy (CAPE), and Madden–Julian oscillation (MJO) phases are analyzed to elucidate the causes of contrasting characteristics of these two cyclones. During the progression and intensification of the storm Nanauk, SST decreased drastically (magnitude of ~ 4 °C), while VWS anomaly is found to be increased (~ 8 ms−1), followed by entrainment of dry air leading to a decrease in upward LHF anomaly (~ 3.5 to 4 J m−2) hindering further moisture supply. Moreover, just before the storm initiation, the CAPE anomaly was around − 800 to − 1000 J kg–1 and the MJO condition was also unfavorable for the continuous intensification of the cyclone. All these factors contributed to the quick dissipation of cyclone Nanauk. However, high SST (~ 31 °C) along with other favorable atmospheric conditions contrary to cyclone Nanauk provided a conducive environment for Nisarga to intensify and prevail. A high-pressure anticyclonic circulation at the eastern side of Nisarga over Indian land dragged the storm north-eastwards across the Maharashtra coast. The convective phase of MJO (magnitude > 1) along with the strong lower tropospheric westerly wind at the west side of the convective system also helped cyclone Nisarga to propagate eastward.
Site response analysis beneath the Kashmir basin (NW Himalaya) using ambient noise
by S Vishal Gupta, Imtiyaz A Parvez, and Prosanta K Khan
Deep sedimentary valleys entrap and amplify seismic waves, which is further responsible for site-specific amplification over the allied region. Kashmir valley is a deep and active sedimentary basin in the northwest portion of the Himalayan lap. To characterize the site response for the intermountain Kashmir valley for the first time, we acquire single station microtremor measurements using Lennartz (three components/5s seismometer) at 141 sites with a grid of (5 km × 5 km) and two-dimensional (2D) array measurements using seven units of recording station of different geometry pattern at 38 sites across different lithological setup within the basin. We present here (1) a first-order predominant fundamental frequency (0.21–10.19 Hz) and subsequent sedimentary thickness maps for the entire valley, reflecting deep deposition with spatial heterogeneity in the area using the horizontal-to-vertical spectral ratio (HVSR) technique; (2) spatial classification of generated HVSR curves linking subsurface geomorphology and post strong ground-motion scenario in the valley; (3) extraction of the best dispersion characteristics of Rayleigh waves (fundamental mode) from all recorded vertical signals at each array site using frequency–wavenumber (F-K) method with an emphasis on signal synchronicity; and (4) retrieval of one-dimensional (1D) shear wave velocity (VS) profiles across the valley from the inversion of dispersion curves using the neighborhood algorithm. Distinct model parameterizations were tested for the inversion to achieve the optimal inversion misfit. The collocated 1D VS profiles are consistent with newly drilled borehole logs information. Besides, the presented first-ever ambient noise survey–based site response study meets the objectives of site-specific seismic hazard and risk analysis of the Kashmir valley at the regional scale.
Analysis of heat wave over different physiographical regions in India
by Rani Devi, Krushna Chandra Gouda and Smrutishree Lenka
Heat waves (HW) in the tropical and subtropical region are the deadliest throughout the world. This study represents the HW characteristics over the tropical country India mainly divided in three different landmasses, i.e. hilly, plains and coastal. This work presents the HW characteristic quantification based on the physiography of the regions in India assessed over a period of 70 years (1951–2020) using the Indian Meteorological Department (IMD) observed data. Mostly, the earlier studies are more focused on HW in the plains and coastal regions. HW in hilly region has been excluded by using single threshold for the whole Indian region, so based on the topographical features, three thresholds of daily maximum temperature (DMT), i.e. DMT > 40 °C (plains), DMT > 37 °C (coastal) and DMT > 30 °C (hilly), are being applied. The spatio-temporal analysis of HW shows its occurrence in hilly regions (J&K, Manipur, Mizoram and Tripura) since 1972 and maximum 10-day HW in the northern region. The coupled model intercomparison project phase (CMIP6) model output for 7 models has been assessed to quantify the performance in capturing the HW parameters compared to IMD observations. The ensemble mean of 2 CMIP6 models, i.e. EC Earth3 and MPI LR, has been used to project future changes in HW properties over India under the different emission scenarios, i.e. SSP126 and SSP 585. The key finding of this study shows that for year 2050 onwards, a sharp increase in HW days, HW events and HW duration (average and maximum) will increase in hilly, coastal and plain region. Under SSP 585, the southern part will likely get more new hotspots of HW in India, and there will be significantly more longer, intense and frequent HW occurrence as compared to SSP 126. The results show the need of improvement in understanding of HW at regional scale and especially in hilly region.
Dynamics of Indian summer monsoon in different phases
by Smrutishree Lenka, Krushna Chandra Gouda, Rani Devi & C. M. Joseph
Abstract
Better understanding on the phases of Indian summer monsoon i.e., onset, progress and withdrawal helps in the decision making and proper management. In this paper, date of onset of monsoon (DOM) and date of withdrawal of monsoon (DWM) is estimated using the long-term daily rainfall data observed by the India Meteorological Department (IMD) and cumulative rainfall anomaly concept, as the onset/withdrawal represents a signature of rainfall transition. The results are promising with a better match between the estimated DOM and IMD announced DOM over Kerala. The length of monsoon season variability is also analysed. The concept of onset at grid point (OGP) in India is also evaluated and progress of monsoon is monitored by considering the latitudinal variation, persistent & strength of rainfall in the continental India. The zonal classification of the spread of monsoon in terms of the intensity and time (duration) is presented. Zone wise pentad analysis of onset clearly signifies the progress of monsoon and its dependency on various physical factors like SST, OLR, TTG, ITCZ, Energy budget and wind circulation. The impact of ENSO on the OGP is brought out, which clearly shows that DOM, DWM and progress is almost influenced by the largescale processes. The intensity–duration–frequency analysis quantified the actual ISM length, seasonal rainfall and rainy days at all locations in India. This study will help modellers in better understanding of onset and progress of monsoon dynamics in India and can be integrated with NWP for the accurate and advance prediction of DOM and the progress.
Source: https://doi.org/10.1007/s00382-023-06925-1
- Thermodynamic structure of the atmospheric boundary layer over a coastal station in India for contrasting sky conditions during different seasons
- Dynamical influence of MJO phases on the onset of Indian monsoon
- A case study on multiple self-interactions of MSTID bands: New insights
- Impact of different land use data on WRF model short range forecasts during pre-monsoon and monsoon seasons in India
Page 11 of 27
