This session showcases fresh, original perspectives that tackle unique challenges in geospatial technology. Explore innovative approaches to gaining access to historical and archival imagery and discover new ways to modernize GIS for today’s needs.
4:30 PM – 4:45 PM – Satellite-Based Crop Classification for Navajo Nation: Integrating High-Confidence Cropland Data Layer Pixels with Sentinel-2 Data
The agricultural industry is the primary source of food, goods, and commodity services. Despite advancements in technology, predicting crop yield and productivity remains a significant challenge. Accurate predictions largely depend on effective crop monitoring and optimization of resources such as fertilizers, water, and pesticides. In the United States, crop classification primarily relies on open-source data from the United States Department of Agriculture (USDA), specifically the Crop Data Layer (CDL). This study focuses on the Navajo Nation, where agriculture is deeply tied to both livelihoods and cultural preservation, and the region remains highly vulnerable to drought impacts. To support sustainable agricultural practices, this research demonstrates crop classification using high-confidence pixels, which have high certainty based on strong agreement with reference data. These reliable pixels are derived from the USDA Cropland Data Layer (CDL) for the years 2017 and 2022. These years are selected to align with the Census of Agriculture statistics and supplemented with Sentinel-2 satellite data. The Random Forest machine learning algorithm is used to label crop data and spectral features from Sentinel-2 imagery and classify land cover types. To ensure robust model performance, the dataset is split into 80% for training and 20% for validation. This approach allows the Random Forest model to accurately distinguish between crop types and assess classification reliability, reinforcing the effectiveness of high-confidence pixels in regional crop mapping. Spectral indices, including the Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Green Chlorophyll Vegetation Index (GCVI), and Land Surface Water Index (LSWI) are used to improve the classification accuracy of crops derived from Random Forest model. The presentation will discuss the classification accuracy of the data with and without the use of high-confidence pixels for 2017 and 2022 respectively and their significance in decision making, A more accurate crop classification will likely enhance the effectiveness of crop monitoring and resource optimization by facilitating timely data sharing and knowledge exchange among the Navajo Nation farming community. This will also support coordinated responses to threats such as weeds, diseases, insects, pests, and weather, ultimately ensuring farm profitability, with potential for scaling up to larger geographic areas.
Varatharajaperumal Thangavel, Florida Atlantic University
4:45 PM – 5:00 PM – Above Ground Biomass, Carbon Emission, and Removal Estimation Using Machine Learning and Multi-Source Remote Sensing Data in the Living Konso Cultural Landscape World Heritage Site, Ethiopia
Worldwide cultural landscapes are sites of key biodiversity hotspots, and as such, they demonstrate high potential for carbon sequestration. Despite their substantial contribution to climate change mitigation, the focus in terms of research remained predominantly on natural forests. One major challenge in this regard is the lack of accurate estimation of biomass from cultural landscapes, which still remains a critical challenge due to the heterogenous nature of cultural landscapes. Our study’s objective is to estimate carbon storage dynamics in the Konso cultural landscape in Ethiopia. In our study, we benefited from the use of multiple remote sensing and environmental datasets, such as Global Ecosystem Dynamics Investigation (GEDI) Level 4A (L4A) Version 2 data, Global Above and Below Ground Biomass Carbon Density data for the year 2010 (developed by the United Nations Environment Programme World Conservation Monitoring Centre (UNEP-WCMC)), Sentinel 1 GRD (Ground Range Detection), Sentinel-2 (S2) SR (Surface Reflectance), Landsat 5 NDVI, Landsat 9, NICFI (Norway’s International Climate and Forest Initiative), slope, elevation, and aspect, to accurately estimate AGB for heterogenous cultural landscapes in Ethiopia. We have employed a random forest regression model to estimate aboveground biomass (AGB) carbon storage, illustrating relatively strong performance on the training data (R-squared: 0.833 for 2022, 0.74 for 2011; RMSE: 0.11 and 0.25, respectively). However, a significant drop in R-squared on validation data (0.24 for 2022, 0.21 for 2011) and increased RMSE values (0.33 and 0.47) indicate model overfitting.
We estimated the mean AGB carbon storage for the Konso cultural landscape of 32.2 tC/ha for 2011 and 37.1 tC/ha for 2022. The amount we estimated appears to correspond to the national estimate of 32.7 tC/ha for the Combretum-Terminalia forest type. The findings demonstrate that the Konso cultural landscape possesses considerable potential for carbon sequestration, which is crucial for achieving the objective of maintaining global warming below 1.5°C. Our study gives a new perspective into the estimation of AGB from heterogenous cultural landscapes through the integration of multiple datasets using a machine learning approach.
Abiyot Kura, Dilla University
5:00 PM – 5:15 PM – A Deep Learning-Powered Super-Resolution Algorithm for Sentinel-2 Images to Bridge the Resolution Gap in Flood Mapping
Timely and accurate geospatial data is necessary for flood mapping, yet publicly accessible Sentinel-2 imagery with a resolution of 10 m frequently falls short in capturing small-scale flood dynamics. This study introduces STCN-SR, a novel hybrid deep learning architecture that improves Sentinel-2 imagery to a resolution of 5 m by merging transformers and convolutional neural networks. The model is trained and assessed on the DeepFlood dataset, and it produces visually clear, spatially aligned outputs that are on par with high-resolution UAV imagery. It also obtains superior PSNR and SSIM scores. STCN-SR produces georeferenced, super-resolved tiles that can be used directly in GIS workflows, in contrast to conventional techniques. This study shows that super-resolution can be used to improve flood boundary delineation by aligning outputs with UAV ground truth. This is particularly useful in situations where data is scarce or emergency response is required.
Jagrati Talreja, North Carolina A&T State University
Leila Hashemi Beni, North Carolina A&T State University
