This is a 21 class land use image dataset meant for research purposes.

There are 100 images for each of the following classes:

• agricultural
• airplane
• baseballdiamond
• beach
• buildings
• chaparral
• denseresidential
• forest
• freeway
• golfcourse
• harbor
• intersection
• mediumresidential
• mobilehomepark
• overpass
• parkinglot
• river
• runway
• sparseresidential
• storagetanks
• tenniscourt

Each image measures 256x256 pixels.

The images were manually extracted from large images from the USGS National Map Urban Area Imagery collection for various urban areas around the country. The pixel resolution of this public domain imagery is 1 foot.

For more information about the original UC Merced Land Use dataset, please visit the official dataset page:

http://weegee.vision.ucmerced.edu/datasets/landuse.html

Please refer to the original dataset source for any additional details, citations, or specific usage guidelines provided by the dataset creators.

UC Merced is a 21 class land use remote sensing image dataset, with 100 images per class. The images were manually extracted from large images from the USGS National Map Urban Area Imagery collection for various urban areas around the country. The pixel resolution of this public domain imagery is 0.3 m.

While most images are 256x256 pixels, there are 44 images with different shape.

To use this dataset:

import tensorflow_datasets as tfds

ds = tfds.load('uc_merced', split='train')
for ex in ds.take(4):
 print(ex)

See the guide for more informations on tensorflow_datasets.

https://storage.googleapis.com/tfds-data/visualization/fig/uc_merced-2.0.0.png" alt="Visualization" width="500px">

UC Merced land-use data

This data represents a land use survey of 2012 Merced County conducted by the California Department of Water Resources, South Central Region Office staff. Land use boundaries were digitized, and land use data was gathered by staff of DWR’s South Central Region Office using extensive field visits and aerial photography. Detailed agricultural land uses, and lesser detailed urban and native vegetation land uses were mapped. Landsat 5 imagery was analyzed prior to the field survey by DRA staff to map fields likely to have winter crops. The land use data went through standard quality control procedures before final processing. Quality control procedures were performed jointly by staff at DWR’s DRA headquarters and South Central Region Office. Land use field boundaries were digitized with ArcGIS 9.3 using 2010 NAIP as the base, and Google Earth were used as reference as well. Agricultural fields were delineated by following actual field boundaries instead of using the centerlines of roads to represent the field borders. Field boundaries were not drawn to represent legal parcel (ownership) boundaries and are not meant to be used as parcel boundaries. Field work for land use surveys occurs primarily during the summer and early fall, so it can be difficult to identify fields where winter crops have been produced during the survey year. To improve the mapping of winter crops, we analyzed Landsat 5 imagery to identify fields with high winter vegetative cover. The identification of these fields was based on an analysis of Landsat 5 imagery. Visual inspection of the Landsat scene displayed in false color infrared was used to select fields with high and low vegetative cover. These fields were used to develop spectral signatures using ERDAS Imagine and eCognition Developer software. The Landsat image was classified using a maximum likelihood supervised classification to label each pixel as vegetated or not vegetated, then the zonal attributes of polygons representing agricultural fields were summarized to identify fields vegetated during the winter. Polygons representing these fields were used on laptop taken to the field to highlight the fields which should be checked closely for winter crop residue. Site visits occurred from July through October 2012. Images and land use boundaries were loaded onto laptop computers that, in most cases, were used as the field data collection tools. GPS units connected to the laptop computers were used to confirm surveyor's location with respect to the fields. Staff took these laptop computers into the field and virtually all the areas were visited to positively identify the land use. Land use codes were digitized in the field on laptop computers using ESRI ArcMAP software, version 9.3. Some staff took printed aerial photos into the field and wrote directly onto these photo field sheets. The data from the photo field sheets were digitized back in the office. The primary focus of this land use survey is mapping agricultural fields. Urban residences and other urban areas were delineated using aerial photo interpretation. Some urban areas may have been missed. Rural residential land use was delineated by drawing polygons to surround houses and other buildings along with some of the surrounding land. These footprint areas do not represent the entire footprint of urban land. Water source information was not collected for this land use survey. Therefore, the water source has been designated as Unknown. Before final processing, standard quality control procedures were performed jointly by staff at DWR’s South Central Region, and at DRA's headquarters office under the leadership of Jean Woods, Senior Land and Water Use Supervisor. After quality control procedures were completed, the data was finalized. The positional accuracy of the digital line work, which is based upon the orthorectified NAIP imagery, is approximately 6 meters. The land use attribute accuracy for agricultural fields is high, because almost every delineated field was visited by a surveyor. The accuracy is 95 percent because some errors may have occurred. Possible sources of attribute errors are: a) Human error in the identification of crop types, b) Data entry errors.The associated data are considered DWR enterprise GIS data, which meet all appropriate requirements of the DWR Spatial Data Standards, specifically the “SPATIAL DATA STANDARDS FOR THE CALIFORNIA DEPARTMENT OF WATER RESOURCES” version 3.1, dated September 11, 2019.DWR makes no warranties or guarantees - either expressed or implied - as to the completeness, accuracy, or correctness of the data. DWR neither accepts nor assumes liability arising from or for any incorrect, incomplete, or misleading subject data.Comments, problems, improvements, updates, or suggestions should be forwarded to gis@water.ca.gov.

Dataset README

  1. General Information

Number of Labels: There are a total of 5 labels, namely: Agriculture, Bareland, Forest, Residential, and River. Number of Clients: The dataset consists of 100 clients. Data Volume per Client: Each client contains approximately 350 tif format images.

  2. Data Sources

All the images are collected from 6 different datasets, which are as follows: Eurosat UC Merced Land Use Dataset AID NWPU - RESISC45 WHU-RS19 NaSC-tg2 The data… See the full description on the dataset page: https://huggingface.co/datasets/LiberiOnTerra/FedRB.

Original GIS polygon dataset derived from the Community and Economic Development. The scale of the data is 1:786,633 and covers the entire county of Merced. Data is updated based on CED updates. Information such as Land use, acreage, community type, General plan abbreviations are identified in this layer.

This child data release includes fused topo-bathymetric digital elevation models of the Merced and Tuolumne Rivers in California used to support research on anadromous salmonids. The purpose of this study was to calculate the capacity for reintroduction of salmonids above impassable barriers. Airborne, near-infrared (NIR) LiDAR and hyperspectral imagery were acquired simultaneously in September 2014 from a Cessna Caravan, with the LiDAR data used to map topography of dry land and the imagery used to map water depth in the wetted channel. Topo-bathymetric DEMs of channels and floodplains with 1-m resolution were constructed for the study reaches by using remotely sensed hyperspectral image data to estimate water depths within the below-water portion of the channel and using remotely-sensed LiDAR for the above-water portion of the channel. Water depths were subtracted from water-surface elevations measured by the LiDAR to obtain bed elevations within the wetted channel. The digital elevation model above the water surface was created using the LiDAR data. We used a Leica Airborne Laser Scanner ALS50, with mean point densities >12 points/m2 and reported horizontal and vertical accuracies of 2 cm and 7 cm, respectively. The raw LiDAR point cloud was processed into bare-earth DEMs with 1 m grid cells. The digital elevation model for areas below the water surface was created using the hyperspectral imagery. Hyperspectral imagery was collected using a Compact Airborne Spectographic Imager (CASI) 1500 (ITRES 2014), producing imagery with 48 spectral bands (wavelengths 380 to 1050 nm). Raw image flight strips were geometrically and radiometrically corrected with ITRES software, then atmospherically corrected using ATCOR4 (ReSe 2014). The final images were in units of reflectance for each band, with a spatial resolution of 0.5 m. Water depths were estimated from the imagery using the Optimal Band Ratio Analysis (OBRA) depth retrieval algorithm, a calibration technique that relates field-measured water depths (d) to an image-derived quantity defined as the natural logarithm of the ratio of two spectral bands (Legleiter et al. 2009).

The 2002 Merced County land use survey data set was developed by DWR through its Division of Planning and Local Assistance (DPLA). The data was gathered using aerial photography and extensive field visits, the land use boundaries and attributes were digitized, and the resultant data went through standard quality control procedures before finalizing. The land uses that were gathered were detailed agricultural land uses, and lesser detailed urban and native vegetation land uses. The data was gathered and digitized by staff of DWR’s San Joaquin District. Quality control procedures were performed jointly by staff at DWR’s DPLA headquarters and San Joaquin District. Important Points about Using this Data Set: 1. The land use boundaries were drawn on-screen using orthorectified imagery. They were drawn to depict observable areas of the same land use. They were not drawn to represent legal parcel (ownership) boundaries, or meant to be used as parcel boundaries. 2. If the data is to be brought into a GIS for analysis of cropped (or planted) acreage, two things must be understood: a. The acreage of each field delineated is the gross area of the field. The amount of actual planted and irrigated acreage will always be less than the gross acreage, because of ditches, farm roads, other roads, farmsteads, etc. Thus, a delineated corn field may have a GIS calculated acreage of 40 acres but will have a smaller cropped (or net) acreage, maybe 38 acres. b. Double and multicropping must be taken into account. A delineated field of 40 acres might have been cropped first with grain, then with corn, and coded as such. To estimate actual cropped acres, the two crops are added together (38 acres of grain and 38 acres of corn) which results in a total of 76 acres of net crop (or planted) acres. 3. For this survey, a new special condition attribute was created. It is “C”, for green-chop, and is only used with grain (wheat) that is green-chopped. 4. Water source information was not collected for this survey. 5. Not all land use codes will be represented in the survey.