This review explores the comparative utility of machine learning (ML) and deep learning (DL) in land system science (LSS) classification tasks. Through a comprehensive assessment, the study reveals that while DL techniques have emerged with transformative potential, their application in LSS often faces challenges related to data availability, computational demands, model interpretability, and overfitting. In many instances, traditional ML models currently present more effective solutions, as illustrated in our decision-making framework. Integrative opportunities for enhancing classification accuracy include data integration from diverse sources, the development of advanced DL architectures, leveraging unsupervised learning, and infusing domain-specific knowledge. The research also emphasizes the need for regular model evaluation, the creation of diversified training datasets, and fostering interdisciplinary collaborations. Furthermore, while the promise of DL for future advancements in LSS is undeniable, present considerations often tip the balance in favor of ML models for many classification schemes. This review serves as a guide for researchers, emphasizing the importance of choosing the right computational tools in the evolving landscape of LSS, to achieve reliable and nuanced land-use change data.

The maintenance of flood mitigation levees in the U.S. and elsewhere, has often found to be insufficient. Recent U.S. levee safety inspections have discovered substantial deficiencies in many federally monitored levees resulting in some of them being assigned an ‘unacceptable’ safety rating. Most levees in the U.S. were constructed using federal money but then were turned over to local government entities to maintain, thus socioeconomic characteristics of these communities may impact their ability to maintain their levee(s). We used dasymetric mapping to assign socioeconomic parameters to levee protected areas along the lower Illinois River to explore differences in these characteristics between communities with an unacceptable levee safety rating to those with an acceptable rating. Principal components analysis was used to determine which socioeconomic parameters explained the majority of variance between levee protected communities and the Mann-Whitney-Wilcoxon U-Test for significance testing. These analyses revealed that differences in total population , race, average per-capita income, and number of residential homes were influential indicators for explaining differences between comunities with levees with acceptable versus unacceptable ratings. These results suggest populations who inhabit levee systems with unacceptable safety ratings are white and relatively wealthier than communities located within levee systems with acceptable ratings. This finding is counter to research that shows that the poor and poorer minorities live in areas with higher flood risk. This is perhaps wealthier floodplain property owners inhabitating unacceptable levee systems may be foregoing necessary levee maintenance, inastead relying on other government programs to pay for flood damages.

Tropical cyclones are incredibly destructive and deadly, inflicting immense losses to coastal properties and infrastructure. Hurricane-induced coastal floods are often the biggest threat to life and the coastal environment. A quick and accurate estimation of coastal flood extent is urgently required for disaster rescue and emergency response. In this study, a combined Digital Elevation Model (DEM) based water fraction (DWF) method was implemented to simulate coastal floods during Hurricane Harvey on the South Texas coast. Water fraction values were calculated to create a 15 km flood map from multiple channels of the Advanced Technology Microwave Sound dataset. Based on hydrological inundation mechanism and topographic information, the coarse-resolution flood map derived from water fraction values was then downscaled to a high spatial resolution of 10 m. To evaluate the DWF result, Storm Surge Hindcast product and flood-reported high-water-mark observations were used. The results indicated a high overlapping area between the DWF map and buffered flood-reported high-water-marks (HWMs), with a percentage of more than 85%. Furthermore, the correlation coefficient between the DWF map and CERA SSH product was 0.91, which demonstrates a strong linear relationship between these two maps. The DWF model has a promising capacity to create high-resolution flood maps over large areas that can aid in emergency response. The result generated here can also be useful for flood risk management, especially through risk communication.