Mississippi Land Cover Training Dataset

This dataset provides 1-meter resolution land cover image-label pairs for training machine learning models to classify land cover in Mississippi. Consisting of 1,000 image-label pairs, the dataset is split into 4 splits according to a specialized stratified sampling method that relies on Annual NLCD 2023 Land Cover data to ensure that each split is representative of land cover found across the state of Mississippi. More information will be provided in the accompanying paper for which this dataset supports --- for more information, please contact Dakota Hester. The imagery used in this dataset is derived from images acquired through the USDA National Agriculture Imagery Program (NAIP) during the year 2023, and the labels are hand-annotated by the dataset authors with assistance from a pre-trained Segment Anything 2.0 model hosted by CVAT.ai to expedite the annotation process. Tiles are visually inspected to ensure quality and consistency, though this dataset is provided "as is" without any guarantees of accuracy or completeness, nor with any warranties of merchantability or fitness for a particular purpose. Imagery and Label Format Within each split, the folders `input` and `target` contain the 1-meter resolution imagery and corresponding labels, respectively. Each file is named according to a unique 7-digit identifier, (e.g., `0096708.tif`). All files are in GeoTIFF format and are projected in the Mississippi Transverse Mercator coordinate system. The input imagery is in 4-band (RGBN) format with 8-bit unsigned integer values with size 256x256. Target labels are provided as a single band, with each pixel value corresponding to a specific land cover class as follows: Pixel Value Land Cover Class Description 0 No Data N/A 1 Open Water Water bodies such as lakes, rivers, ponds, streams, pools, etc. 2 Impervious Structures Man-made structures that have elevated surfaces, such as buildings, docks, large vehicles, etc. 3 Impervious Surfaces Man-made surfaces that are not elevated, such as roads, parking lots, sidewalks, etc. 4 Barren Land Areas absent of vegetation, structures, or water, such as bare soil, sand, gravel, rock, etc. 5 Tree Canopy/Woody Vegetation Areas with visible tree canopy or moderate-to-large woody shrubs with substantial density and/or structure. 6 Herbaceous/Low Vegetation Low-growing vegetation such as grasses, aquatic plants, pastures, and small shrubs with low density and/or sparse structure. 7 Cultivated Crops Agricultural fields with visible crop growth identifiable by high near-infrared reflectance and row crop patterns. 8 Unclassified Regions obscured by shadows or other image artifacts that prevent reliable classification. The structure of each split is as follows: split_1/ ├── input/ │ ├── 0096708.tif │ └── ... └── target/ ├── 0096708.tif └── ... Acknowledgments This dataset was created with support from the Mississippi Agricultural and Forestry Experiment Station under a strategic research initiative in an effort to build machine learning systems that enhance stakeholder understanding of the distribution of natural resources in the state of Mississippi. This dataset is provided under the Creative Commons Attribution 4.0 International License, which allows for free use, distribution, and modification of the dataset with proper attribution to the authors. When using or citing this dataset, please reference the following citation: @misc{hester2025mslc, author = {Dakota Hester and Vitor Souza Martins}, title = {Mississippi Land Cover Training Dataset}, year = {2025}, publisher = {Mississippi State University}, doi = {10.5281/zenodo.15670824}, }