Assessing wetland habitat vulnerability in moribund Ganges delta using bivariate models and machine learning algorithms

Pal, Swades; Paul, Satyajit

doi:10.1016/j.ecolind.2020.106866

“…More than 70% of wetland area has been converted to land, and existing wetlands have undergone into hydrological inconsistency, diminishing water depth, unpredictable water supply, and rising temperatures and all of these are detrimental to the aquatic environment in general and fish habitat in particular (Whitney et al 2020;Pandey et al 2021;Thaman 2021). Wetland loss and hydro-ecological alteration in existing wetland are caused by land use change, agricultural encroachment, built-up extension, fragmentation of wetland, loss of tie channels, and river flow reduction due to damming Islam, et al 2021;Pal & Paul 2020). Wetland is an excellent fishing spot, and a large number of people depend on it for their livelihood (Gosling et al 2017;.…”

Section: Discussionmentioning

confidence: 99%

Integration of satellite image–derived temperature and water depth for assessing fish habitability in dam controlled flood plain wetland

Kundu

¹

,

Pal

²

,

Talukdar

³

et al. 2022

Environ Sci Pollut Res

Self Cite

6

0

View full text Add to dashboard Cite

The present study attempted to investigate the changes in temperature conducive to fish habitability during the summer months in a hydrologically modified wetland following damming over a river. Satellite image-driven temperature and depth data calibrated with field data were used to analyse fish habitability and the presence of thermally optimum habitable zones in some fishes such as Labeo Rohita, Cirrhinus mrigala, Tilapia fish, Small shrimp, and Cat fishes. The study was conducted both at the water's surface and at the optimum depth of survival. It is very obvious from the analysis that a larger part of wetland has become an area that destroyed aquatic habitat during the post-dam period and existing wetlands have suffered significant shallowing of water depth. This has resulted in a shrinking of the thermally optimum area of fish survival in relation to surface water temperature (from 100.09 km2 to 74.24 km2 before the dam to 93.97 km2 to 0 km2 after the dam) and an improvement in the optimum habitable condition in the comfortable depth niche of survival. In the post-dam period, it increased from 75.49 % to 99.765%. Since the damming effect causes a 30.53 to 100% depletion of the optimum depth niche, improving the thermal environment has no effect on fish habitability. More water must be released from dams for restoration. Image-driven depth and temperature data calibrated with field information has been successfully applied in data sparse conditions, and it is further recommended in future work.

show abstract

“…RF reduces the dimensionality in the database by using a built-in feature selection system that can regulate multiple parameters without removing some of them (Tella et al, 2021). By estimating the increase in prediction error in the dataset, RF can compute the variable importance scores of each constituent tree and also the entire database (Elavarasan and Vincent, 2021;Pal and Paul, 2020). Studies like Chan and Paelinckx, (2008) and Pal and Mather, (2003) reported that RF (uses bagging) is more sensitive to noise than the algorithms based on boosting technique.…”

Section: Random Forest Algorithm Uses Bootstrap Aggregation Technique While Growing Multiple Decisionmentioning

confidence: 99%

Point and pixel inclusive machine learning models for exploring gully erosion susceptibility

Pal

¹

,

Debanshi

²

,

Paul³

2022

Preprint

Self Cite

0

View full text Add to dashboard Cite

Sample point based spatial model derived from Machine Learning (ML) algorithms often generalizes the spatial pattern of event. The present study has tried to highlight how far it is acceptable and can it replace pixel based modelling? The present study has presented a comparative view of pixel and sample point based modelling of gully erosion susceptibility of the upper Mayurakshi basin to assess the predictabilities. Random forest (RF), Support Vector Machine (SVM) and (ADB) were applied for developing pixel and point based models in WEKA and Python software environments respectively. From the models, it is found that 14-25% area located mainly in the upper parts of the study unit is very highly susceptible to gully erosion. Based on the accuracy and performance level using Area under curve (AUC) of Receiver operating curve, sensitivity, precision, F1 score, MCC pixel based ensemble models are superior to point based modelling. Often, the point based models have a very poor agreement with training and testing data. So, pixel based models could not be replaced with point based models. RF is found as the best representative model. The study, therefore, recommends using pixel based modelling for this or a similar purpose. Since the models have figured out the gully erosion susceptible areas, it would be a useful tool for related planning processes.

show abstract

“…RF reduces the dimensionality in the database by using a built-in feature selection system that can regulate multiple parameters without removing some of them (Tella et al, 2021). By estimating the increase in prediction error in the dataset, RF can compute the variable importance scores of each constituent tree and also the entire database (Elavarasan and Vincent, 2021;Pal and Paul, 2020). Studies like Chan and Paelinckx, (2008) and Pal and Mather, (2003) reported that RF (uses bagging) is more sensitive to noise than the algorithms based on boosting technique.…”

Section: Random Forest Algorithm Uses Bootstrap Aggregation Technique...mentioning

confidence: 99%

“…ML algorithms have the potential to identify the non-linear complex relationship (Mosavi et al, 2018) and eventually produce the spatial model of gully erosion susceptibility based on associated predisposing factors of gully erosion (Saha et al, 2020). Presently the integration of remote sensing and GIS with ML and ensemble algorithms has made the mapping of any environmental issue including gully erosion easier and precise (Saha et al, 2020;Pal and Paul, 2020). For example, Arabameri et al, (2018); used R programming language and GIS technique for assessing gully erosion susceptibility and reported greater performance accuracy and preciseness over the conventional techniques.…”

Section: Introductionmentioning

confidence: 99%

Point and pixel inclusive machine learning models for exploring gully erosion susceptibility

Pal

¹

,

Debanshi

²

,

Paul³

2022

Geocarto International

Self Cite

1

0

View full text Add to dashboard Cite

Sample point based spatial model derived from Machine Learning (ML) algorithms often generalizes the spatial pattern of event. The present study has tried to highlight how far it is acceptable and can it replace pixel based modelling? The present study has presented a comparative view of pixel and sample point based modelling of gully erosion susceptibility of the upper Mayurakshi basin to assess the predictabilities. Random forest (RF), Support Vector Machine (SVM) and (ADB) were applied for developing pixel and point based models in WEKA and Python software environments respectively. From the models, it is found that 14-25% area located mainly in the upper parts of the study unit is very highly susceptible to gully erosion. Based on the accuracy and performance level using Area under curve (AUC) of Receiver operating curve, sensitivity, precision, F1 score, MCC pixel based ensemble models are superior to point based modelling. Often, the point based models have a very poor agreement with training and testing data. So, pixel based models could not be replaced with point based models. RF is found as the best representative model. The study, therefore, recommends using pixel based modelling for this or a similar purpose. Since the models have figured out the gully erosion susceptible areas, it would be a useful tool for related planning processes.

show abstract

Assessing wetland habitat vulnerability in moribund Ganges delta using bivariate models and machine learning algorithms

Cited by 42 publications

References 122 publications

Integration of satellite image–derived temperature and water depth for assessing fish habitability in dam controlled flood plain wetland

Integration of satellite image–derived temperature and water depth for assessing fish habitability in dam controlled flood plain wetland

Point and pixel inclusive machine learning models for exploring gully erosion susceptibility

Point and pixel inclusive machine learning models for exploring gully erosion susceptibility

Contact Info

Product

Resources

About