Figure 2 : Machine learning workflow
Reference Models used Timescale / Study Study area
Tongal and Booij ( 2018 ) |
SVR , RF |
Daily streamflow |
North Fork River , Chehalis River , Carson River , Sacramento River ( United States ) |
Fox and Magoulick ( 2019 ) |
RF |
Monthly and annual hydrologic disturbance of stream |
Arkansas , Missouri , eastern Oklahoma ( United States )
Li et al . ( 2019 ) ELM , RF , BPNN , SVR Daily streamflow Wei River Basin ( China )
Sahoo et al . ( 2019 ) |
SVR , ANN-ELM , GPR |
Monthly low flow |
|
|
time series |
Mahanadi River Basin ( India )
Hussain & Khan ( 2020 ) |
MLP , SVR , RF |
Monthly streamflow |
Hunza River Basin |
|
|
|
( Pakistan ) |
Konapala et al . ( 2020 ) LSTM , PB , LSTM-PB Daily streamflow United States
Li et al . ( 2020 ) ENR , SVR , RF , XGB , MSES
Pham et al . ( 2021 ) |
RF , MLR , Naïve |
Short-term daily |
|
|
streamflow |
Monthly streamflow Yangtze River Basin ( China )
Watersheds in Pacific Northwest ( United States )
Adib & Harun ( 2022 ) |
SVR , RF |
Mothly streamflow |
Kurau River Basin |
|
|
|
( Malaysia ) |
Saravanan et al . ( 2023 ) SVR , RF , M5P , MLP , LR
Daily streamflow Godavari River ( India )
Nasir et al . ( 2024 ) |
SVR |
Monthly streamflow |
Kurau River Basin |
|
|
|
( Malaysia ) |
Table 2 : ML-based models application for hydrological processes
illustrated in Figure 2 . The ability to autonomously learn from past occurrences , deal with nonlinear physical processes , and address their complexity to make precise predictions using minimal data and mathematical equations without being explicitly programmed to do so has recently garnered significant attention from hydrologists ( Xu & Liang , 2021 ). Physical-based models like SWAT ( Arnold et al ., 2012 ), HEC-HMS ( USACE-HEC , 2000 ), and MIKE SHE ( Refshaard & Storm , 1995 ), as well as conceptual models like IHACRES and TOPMODEL ( Abdulkareem et al ., 2018 ), demand substantial input data comprising topography , soil moisture content , initial water depth , topology , and other physical catchment characteristics , rendering them resource-intensive , time-consuming and expensive . As water scarcity and extreme weather events are giving alarming signals , the application of ML in modelling future irrigation availability from river basins and rivers holds great potential
27