Search

The present work serves as a theoretical commentary over the Himalayan Re-gridded and Observational EXperiment (HiROX) and distributes the conceived framework grown under the two-phase experiment. HiROX is a scientific effort aimed at revising the existing precipitation knowledge over the Central Himalayas over the state of Uttarakhand, including the Upper Ganga Basin. The first phase of the study is experimenting with machine learning methods aimed at developing re-gridded precipitation datasets capable of incorporating selective precipitation features and distributing the developed machine learning framework to the research community. The model is designed for easy reproduction of re-gridded precipitation data and the ability to adapt to the available precipitation datasets subject to missing and/or modified inputs. The model is capable of ingesting gridded precipitation datasets at varying spatial resolutions, along with point data values, and can return gridded precipitation information at a spatial resolution of 5 × 5 km2. The second objective of the companion paper under HiROX is the realization of the features of the developed model and exercising it in the development of two high-resolution precipitation datasets at daily timesteps, namely HiROX-1 (~50 yrs) and HiROX-2 (~20 yrs). The present paper limits the discourse to robust development, and the produced datasets are discussed and distributed in the joint paper – Himalayan Re-gridded and Observational Experiment (HiROX): Part II – Application. [ABSTRACT FROM AUTHOR]

This paper releases two regional precipitation products produced in the Himalayan Re-gridded and Observational Experiment (HiROX) conducted over the Central (Uttarakhand) Himalayas including the Upper Ganga Basin. The distributed precipitation products cover a period of about 70 yrs: HiROX-1 ranges from 1972 to 2018 (47 yrs) and is a regridded static daily precipitation product, and HiROX-2 ranges from 2000 to 2021 (22 yrs) and is a regridded dynamic daily precipitation product with continuous upgradation capacity. These datasets are capable of (a) precisely capturing and reproducing refined spatial distribution of precipitation and (b) preserving data accuracy along with other spatiotemporal characteristics over the study region, better than that observed in popular precipitation products covering the study region. The comparative investigation established HiROX-1 as better than some or all the tested datasets in terms of (a) spatial resolution, (b) accuracy, (c) temporal resolution, (d) temporal coverage, and (e) spatial distribution, and presents their application suitability. The second precipitation product, HiROX-2, closely followed HiROX-1, as reported in the ensuing investigation. We employed an open-source precipitation model, HiROX-M1, a flexible machine learning framework capable of processing precipitation information at fine spatial resolution. These regional precipitation products are provided at a spatial resolution of 5 × 5 km2 on daily scale over projected and geographic coordinate systems. HiROX-1 is a station-dependent product and HiROX-2 is a near real-time station-independent product with an updated frequency equal to that of Global Precipitation Measurement (GPM) Integrated Multi-satellitE Retrievals for GPM (IMERG) Final Precipitation L3 1 day 0.1° × 0.1° V06 (GPM_3IMERGDF V06) product. The released products are platform-independent in choice of operating system and geographic information system (GIS) interface. The released data shall assist in closing the spatial discontinuity of ground station information over the Central Himalayan region over Uttarakhand, including the Upper Ganga Basin. [ABSTRACT FROM AUTHOR]

Artificial Intelligence is used to increase the prediction accuracy of the empirical propagation models by considering multivariate parameters in radio path. The mean absolute error between the empirical path loss models and path loss from the measured data at 1800 MHz has been compared. Tuning of the best fit model is done by incorporating correction factors due to local terrain conditions. Lee's model shows the minimum RMSE and is tuned further. Measurement is conducted at the fringe area of Uttarakhand, India, but due to terrain conditions, it is not possible to collect signal strength at mountains, deep valleys, and dense forests. In this paper, tuning and machine learning is used to predict the signal strength at the missed point and thus improves the predicting accuracy and QoS of propagation models. A multivariate machine learning classifier approach is applied, and KNN shows a maximum area under the curve (AUC) of 0.99. [ABSTRACT FROM AUTHOR]