Integrated remote sensing and machine learning tools for estimating ecological flow regimes in tropical river reaches

Near real-time flood inundation and hazard mapping of Baitarani River Basin using Google Earth Engine and SAR imagery

Flood inundation mapping and satellite imagery monitoring are critical and effective responses during flood events. Mapping of a flood using optical data is limited due to the unavailability of cloud-free images. Because of its capacity to penetrate clouds and operate in all kinds of weather, synthetic aperture radar is preferred for water inundation mapping. Flood mapping in Eastern India's Baitarani River Basin for 2018, 2019, 2020, 2021, and 2022 was performed in this study using Sentinel-1 imagery and Google Earth Engine with Otsu's algorithm. Different machine-learning algorithms were used to map the LULC of the study region. Dual polarizations VH and VV and their combinations VV×VH, VV+VH, VH-VV, VV-VH, VV/VH, and VH/VV were examined to identify non-water and water bodies. The normalized difference water index (NDWI) map derived from Sentinel-2 data validated the surface water inundation with 80% accuracy. The total inundated areas were identified as 440.3 km2 in 2018, 268.58 km2 in 2019, 178.40 km2 in 2020, 203.79 km2 in 2021, and 321.33 km2 in 2022, respectively. The overlap of flood maps on the LULC map indicated that flooding highly affected agriculture and urban areas in these years. The approach using the near-real-time Sentinel-1 SAR imagery and GEE platform can be operationalized for periodic flood mapping, helps develop flood control measures, and helps enhance flood management. The generated annual flood inundation maps are also useful for policy development, agriculture yield estimation, crop insurance framing, etc.

Calculation and evaluation of suitable ecological flows for eco-environmental recovery of cascade-developed rivers

River cascade development affects the hydrological and habitat characteristics of the region and disrupts the dynamic balance of stable river ecosystems. The most profound impact of river cascade development is on the resident fish species. River ecosystem restoration for maximum river habitat improvement is generally based on water security and environment improvement and effectively embodies the nature-based solutions (NbS) concept of naturalized restoration. Yuanjiang (Y.J.) River is an international river in southwest China seriously affected by cascade development. By determining the response of the river ecosystem and using the key performance indicator method, Yuanjiang carp (Cyprinus carpio rubrofuscus) and red giant catfish (Bagarius rutilus) were identified as the key species in the main stream of the Y.J. River., and the ecological effects of river cascade development on them were studied by applying two-dimensional hydrodynamic physical habitat simulation and multi-objective ecological scheduling models. Based on the calculation results for ecological operation optimization of cascade reservoirs, an improved progressive optimality algorithm was used to calculate the ecological flows required to maintain the stability of the river ecosystem. With the increasing extent of cascade development in the river, important indicators, such as the intra-annual, extreme, high, and low flows have changed significantly, and the hydrological characteristics of the main stream have changed rapidly and comprehensively. Habitat suitability curves were used to determine the appropriate water depth and delineate the weighted usable area required for the spawning, nursing, and growing periods of the key fish species. The suitable ecological flows required for the three life-cycle stages of the C. carpio rubrofuscus accounted for 34, 45, and 62 %, respectively, of the multi-year mean natural water inflow at the Qiaotou (Q.T.) cascade, whereas those required for the three respective periods of B. rutilus accounted for 47, 98, and 27 %, respectively, of the multi-year mean natural water inflow at the Madushan (M.D·S.) cascade. Considering the physiological lifecycle demands of the indicator/key fish species and the upper limit of water resources development and utilization in the key river section, the ecological flow precipitation frequency in the Q.T.-Luodie (Q.T.-L.D.) and M.D·S.-Xinjie (M.D·S.-X.J.) sections (currently at 25, 50, and 75 %, respectively) can be increased to 100 % under optimal operating conditions (cascade hydropower station optimal operation). After implementing the multi-objective ecological operation at the Y.J. River main-stream cascade reservoirs, the suitable habitat area for C. carpio rubrofuscus and B. rutilus increased significantly (>10 % and 15 %, respectively). In general, the NbS-based ecological flow calculation method for cascade-developed rivers has a wide range of applications, which can be useful for the eco-environment restoration of rivers and improving the living habitats of waterway organisms.

MODIS-Landsat fusion-based single-band algorithms for TSS and turbidity estimation in an urban-waste-dominated river reach

Riverine ecosystem management along an urban stretch mostly depends on high-frequent (daily-scale) monitoring of water quality at finer spatial resolutions. However, with the decrease in the number of in-situ monitoring stations owing to their expensive maintenance cost, there is a need to develop the next-generation remote sensing (RS) tools as an alternate approach with better synoptic coverage of river water quality assessment. This study advocates three novel model variants to estimate the total suspended solids (TSS) concentration at daily-scale using the public-domain MODIS and Landsat satellite datasets. The MOD_T model variant uses the 1-day×250 m MODIS public domain datasets, and the FUS_T model is based on the 1-day×30 m MODIS-Landsat fusion datasets, whereas the CFUS_T model integrates the Frank Copula with the FUS_T model. These hierarchical model variants are assessed in the urban-waste-dominated lower Ganges, namely the Hooghly River and the Brahmani River, in eastern India using the measured in-situ TSS datasets at multiple monitoring stations from 2016 to 2019. The results reveal that the CFUS_T is the best TSS estimation model variant that performs with the average coefficient of determination of 0.88-0.93, mean absolute error of 0.17-0.19, and normal root mean square error of 0.05-0.09. Conclusively, the proposed CFUS_T and CFUS_TU stochastic models can be used as potential tools for TSS and turbidity assessment along the dynamic river systems, respectively.