The National Agriculture Imagery Program (NAIP) acquires aerial imagery during the agricultural growing seasons in the continental U.S. NAIP projects are contracted each year based upon available funding and the imagery acquisition cycle. Beginning in 2003, NAIP was acquired on a 5-year cycle. 2008 was a transition year, and a …

1 - OVERVIEW

This dataset contains overhead images of wind turbines from three regions of the United States – the Eastern Midwest (EM), Northwest (NW), and Southwest (SW). The images come from the National Agricultural Imagery Program and were extracted using Google Earth Engine and wind turbine latitude-longitude coordinates from the U.S. Wind Turbine Database. Overall, there are 2003 NAIP collected images, of which 988 images contain wind turbines and the other 1015 are background images (not containing wind turbines) collected from regions nearby the wind turbines. Labels are provided for all images containing wind turbines. We welcome uses of this dataset for object detection or other research purposes.

2 - DATA DETAILS

Each image is 608 x 608 pixels, with a GSD of 1m. This means each image represents a frame of approximately 608 m x 608m. Because images were collected from overhead the exact wind turbine coordinates, images used to be nearly exactly centered on turbines. To avoid this issue, images were randomly shifted up to 75m in two directions.

We refer to images without turbines as "background images", and further split up the images with turbines into the training and testing set splits. We call the training images with turbines "real images" and the testing images "test images".

Distribution of gathered images by region and type:

Domain

Real

Test

Background

EM

267

100

244

NW

213

100

415

SW

208

100

356

Note that this dataset is part of a larger research project in Duke's 2021-2022 Bass Connections team, Creating Artificial Worlds with AI to Improve Energy Access Data. Our research proposes a technique to synthetically generate images with implanted energy infrastructure objects. We include the synthetic images we generated along with the NAIP collected images above. Generating synthetic images requires a training and testing domain, so for each pair of domains we include 173 synthetically generated images. For a fuller picture on our research, including additional image data from domain adaptation techniques we benchmark our method against, visit our github: https://github.com/energydatalab/closing-the-domain-gap. If you use this dataset, please cite the citation found in our Github README.

3 - NAVIGATING THE DATASET

Once the data is unzipped, you will see that the base level of the dataset contains an image and a labels folder, which have the exact same structure. Here is how the images directory is divided:

| - images

| | - SW

| | | - Background

| | | - Test

| | | - Real

| | - EM

| | | - Background

| | | - Test

| | | - Real

| | - NW

| | | - Background

| | | - Test

| | | - Real

| | - Synthetic

| | | - s_EM_t_NW

| | | - s_SW_t_NW

| | | - s_NW_t_NW

| | | - s_NW_t_EM

| | | - s_SW_t_EM

| | | - s_EM_t_SW

| | | - s_NW_t_SW

| | | - s_EM_t_EM

| | | - s_SW_t_SW

For example images/SW/Real has the 208 .jpg images from the Southwest that contain turbines. The synthetic subdirectory is structured such that for example images/Synthetic/s_EM_t_NW contains synthetic images using a source domain of Eastern Midwest and a target domain of Northwest, meaning the images were stylized to artificially look like Northwest images.

Note that we also provide a domain_overview.json file at the top level to help you navigate the directory. The domain_overview.json file navigates the directory with keys, so if you load the file as f, then f['images']['SW']['Background'] should list all the background photos from the SW. The keys in the domain json are ordered in the order we used the images for our experiments. So if our experiment used 100 SW background images, we used the images corresponding to the first 100 keys.

Naming conventions:

1 - Real and Test images:

{DOMAIN}_{UNIQUE ID}.jpg

For example 'EM_136.jpg' with corresponding label file 'EM_136.txt' refers to an image from the Eastern Midwest with unique ID 136.

2 - Background images:

Background images were collected in 3 waves with the purpose to create a set of images similar visually to real images, just without turbines:

The first wave came from NAIP images from the U.S. Wind Turbine Database coordinates where no wind turbine was present in the snapshot (NAIP images span a relatively large time, thus it is possible that wind turbines might be missing from the images). These images are labeled {DOMAIN}_{UNIQUE ID}.jpg, for example 'EM_1612_background.jpg'.

Using wind turbine coordinates, images were randomly collected either 4000m Southeast or Northwest. These images are labeled {DOMAIN}_{UNIQUE_ID}_{SHIFT DIRECTION (SE or NW)}.jpg. For example 'NW_12750_SE_background.jpg' refers to an image from the Northwest without turbines captured at a shift of 4000m Southeast from a wind turbine with unique ID 12750. Using wind turbine coordinates, images were randomly collected either 6000m Southeast or Northwest. These images are labeled {DOMAIN}_{UNIQUE_ID}_{SHIFT DIRECTION (SE or NW)}_6000.jpg, for example 'NW_12937_NW_6000_background.jpg'.

3 - Synthetic images

Each synthetic image takes in labeled wind turbine examples from the source domain, a background image from the target domain, and a mask. It uses the mask to place wind turbine examples and blends those examples onto the background image using GP-GAN. Thus, the naming conventions for synthetic images are:

{BACKGROUND IMAGE NAME FROM TARGET DOMAIN}_{MASK NUMBER}.jpg.

For example, images/Synthetic/s_NW_t_SW/SW_2246_m15.jpg corresponds to a synthetic image created using labeled wind turbine examples from the Northwest and stylized in the image of the Southwest using Southwest background image SW_2246 and mask 15.

For any remaining questions, please reach out to the author point of contact at caleb.kornfein@gmail.com.

National Agriculture Imagery Program (NAIP)은 미국 본토의 농업 성장 시즌에 항공 이미지를 획득합니다. NAIP 프로젝트는 사용 가능한 자금과 이미지 획득 주기에 따라 매년 계약됩니다. 2003년부터 NAIP는 5년 주기로 획득되었습니다. 2008년은 전환의 해였으며 …

O National Agriculture Imagery Program (NAIP) adquire imagens aéreas durante as estações de cultivo agrícola nos EUA continentais. Os projetos do NAIP são contratados a cada ano com base no financiamento disponível e no ciclo de aquisição de imagens. A partir de 2003, o NAIP foi adquirido em um ciclo de cinco anos. Em 2008, houve uma transição, e um ciclo de três anos começou em 2009. As imagens do NAIP são adquiridas a uma distância de amostragem do solo (GSD, na sigla em inglês) de 1 metro com uma precisão horizontal que corresponde a seis metros de pontos de controle do solo identificáveis por foto, que são usados durante a inspeção de imagens. As imagens mais antigas foram coletadas usando três bandas (vermelho, verde e azul: RGB), mas as imagens mais recentes geralmente são coletadas com uma banda adicional de quase infravermelho (RGBN). Os IDs de recursos RGB começam com "n", os IDs de recursos NRG começam com "c" e os IDs de recursos RGBN começam com "m_". Algumas imagens mais antigas têm GSD de 2 metros.

Abstract:

Saltwater intrusion (SWI) on coastal farmlands can change the soil properties (physical and chemical), rendering it unusable for agricultural purposes. Globally, over a quarter of arable land is negatively impacted by soil salinization, including more than 50% of irrigated land. These salt-impacted lands account for more than 30% of food production worldwide. However, the visible impacts of SWI on coastal ecosystems are challenging to map due to the fine spatial resolution of the salt patches. Here we provide the first mapping of the early visual evidences of SWI impacts on the Delmarva (Delaware, Maryland, Virginia) Peninsula region's farmlands by quantifying and mapping the proportions of the farmlands where the spectral signature of a white salt patch was detected. We focus our effort on fourteen counties on the Delmarva Peninsula. We utilized very high-resolution (1-m) aerial imagery from the National Agriculture Imagery Program (NAIP) and seasonal information derived from the moderate resolution (30-m) Landsat satellite imagery collection. Using a Random Forest algorithm with 100 trees and over 94,240 reference points for training and testing, we developed high-resolution geospatial datasets for the study area for two time-steps: 2011-2013 and 2016-2017. The nine coastal Maryland counties witnessed an average of 79% increase in the salt patches on farmlands. The average increase across the state of Delaware is 81%. Virginia experienced an average of 243% increase in these salt patches. While the expansion rate is alarming, the absolute area with these salt deposits remained rather small even in 2017: about 122 ha in Virginia; 339 ha in Delaware; and 445 ha in Maryland. Visible white salt patches remained a small fraction of total farmlands in each of these counties, ranging between 0.01% and 0.18% in 2011-2013, and between 0.01% and 0.39% in 2016-2017.

This collection of gridded data layers provides the spatial distribution of salt patches along with seven other land cover classes for 14 counties in the Delmarva (Delaware, Maryland and Virginia) Peninsula in the United States of America (USA). We developed high-resolution datasets for the study area for two time-steps: 2011-2013 and 2016-2017. The geospatial datasets are classified images for each time-step and have eight land cover categories as shown below:

Raster value Land cover/use category

1 Forest

2 Marsh

3 Salt patch

4 Built

5 Open water

6 Farmland

7 Bare soil

8 Other vegetation

Input Data:

These geospatial data layers are derived using aerial data from the National Agriculture Imagery Program (NAIP) and satellite data from Landsat 5, 7, and 8. We accessed ortho-rectified NAIP images from June-July 2011 (Maryland), May 2012 (Virginia), September 2013 (Delaware), June 2016 (Virginia), June 2017 (Maryland), and July-August 2017 (Delaware) on the Google Earth Engine (GEE) platform. Cloud-masked top-of-atmosphere (TOA) reflectance images from Landsat 5 (2011, 2012), Landsat 7 (2013), and Landsat 8 (2016, 2017) were obtained using GEE. We derived several spectral indices from the original NAIP and Landsat bands and then used those as inputs into a Random Forest (RF) classifier on GEE.

Methods:

NAIP data contains 4 spectral bands (red, blue, green, and near-infrared) and have a 1 m spatial resolution. Several spectral indices were calculated from the NAIP imagery and used as input into the RF classifier, such as Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), and a Shadow Index (SI). A Principal Component Analysis (PCA) was used to generate four additional bands. In addition, four smoothed NAIP bands were generated using a 3x3 boxcar kernel.

NDVI = (Near-infrared – Red) / (Near-infrared + Red)

NDWI = (Green – Near-infrared) / (Green + Near-infrared)

SI = (256 – Blue) * (256 + Blue)

In order to address limited spectral resolution of the NAIP data and lack of year-long coverage, we incorporated seasonal information from Landsat data. Landsat is a series of satellites launched by the National Aeronautics and Space Administration (NASA) with satellite images distributed through the United States Geological Survey (USGS). Landsat data includes red, green, blue, near-infrared, shortwave infrared, aerosol, cirrus, panchromatic, and thermal bands. All bands are collected at a 30 m resolution except the panchromatic band, which is collected at a 15 m resolution and the thermal bands which are collected at a 100 m resolution. In this work, Landsat 5 was used for 2011 and 2012, Landsat 7 was used for 2013, and Landsat 8 was used for 2016 and 2017. Landsat data (spatial resolution: 30 m) were fused with NAIP data to a resolution of 1 m. An Enhanced Vegetation Index (EVI) was calculated from Landsat bands for each of the four seasons (June-August, September-November, December-February, and March-May) and was then used as input into the RF classifier. Seasonal data was reduced using a median reducer. Landsat thermal bands for each season were also used in the classification. Again, bands were smoothed using a 3x3 boxcar kernel.

EVI = (NIR-Red) / (NIR+6*Red-7.5*Blue+1)

A Random Forest (RF) classifier was used with input data comprised of the four NAIP bands, four PCA bands from NAIP, three indices from NAIP, four smoothed NAIP bands, four smoothed seasonal EVI bands from Landsat, and four smoothed seasonal thermal bands (from Landsat 5) or eight when (for Landsat 7 or 8) – all sampled to a 1 m resolution to match the NAIP input bands.

Due to the high resolution of the input data, there is a considerable 'salt-and-pepper' effects or speckle effects on the classified image, especially for the salt deposit class and its surroundings. As a post-processing step to reduce such speckle effects, we applied a majority filter to the classified image using eight pixel neighbors. For example, any solitary salt patch pixel was reclassified as the majority land cover within the immediate neighborhood. Furthermore, we considered only patches of 10 or more connected 'salt patch' pixels as a valid salt signature. We also used a road mask to minimize the confusion between impervious streets and salt deposits.

Accuracy assessment:

A total of 94,240 reference points were collected from ground surveys and visual interpretation of NAIP imagery from both time periods. 70% of these points were used to train the RF classifier and 30% were used to test accuracy. We calculated user’s accuracy, producer’s accuracy, overall accuracy, kappa statistic, and the F-Score as shown below.

Delaware 2013

Categories

User’s accuracy

Producer’s accuracy

F-score

Overall

Kappa

Forest

88.46%

90.89%

0.90

86.37%

0.83

Marsh

84.03%

80.13%

0.82

Salt patch

97.02%

71.18%

0.82

Built

94.56%

95.06%

0.95

Water

91.01%

96.63%

0.94

Farmland

83.16%

86.19%

0.85

Bare Soil

87.70%

87.30%

0.88

Other Vegetation

82.46%

84.20%

0.83

Delaware 2017

Categories

User’s accuracy

Producer’s accuracy

F-score

Overall

Kappa

Forest

95.07%

90.30%

0.93

91.37%

0.90

Marsh

88.86%

92.44%

0.91

Salt patch

91.82%

85.59%

0.89

Built

87.58%

93.54%

0.90

Water

92.68%

90.48%

0.92

Farmland

91.61%

93.61%

0.93

Bare Soil

95.67%

87.67%

0.91

Other Vegetation

91.16%

90.24%

0.91

Maryland 2011

Categories

User’s accuracy

Producer’s accuracy

F-score

Overall

Kappa

Forest

88.32%

90.97%

0.90

87.20%

0.85

Marsh

87.14%

82.08%

0.85

Salt patch

96.74%

78.76%

0.87

Built

89.02%

88.50%

0.89

Water

92.74%

96.10%

0.94

Farmland

84.31%

89.83%

0.87

Bare Soil

87.53%

90.05%

0.89

Other Vegetation

86.11%

80.57%

0.83

Maryland 2017

Categories

User’s accuracy

Producer’s

This dataset provides a high resolution (1-m) land cover map for Philadelphia, Pennsylvania in the United States of America during the summer of 2017. This dataset was created to differentiate two types of green space in Philadelphia: tree and grass cover. The dataset includes four numerically coded land cover classes.

Input data:

This classification is derived from National Agriculture Imagery Program (NAIP) 1-m aerial imagery captured in the State of Pennsylvania during June of 2017. To improve classification accuracy, NAIP data was stacked with Sentinel-2 level 1C 10-m and 20-m data using the .addBands() function in Google Earth Engine. For the Sentinel-2 data, a median composite was calculated from cloud-masked images collected between April and October of 2017. Sentinel-2 input bands included blue, green, red, red edge 1, red edge 2, red edge 3, near infrared, and shortwave infrared 1. An additional normalized difference vegetation index (NDVI) was calculated from the NAIP and Sentinel-2 bands using the formula:

NDVI = (Near infrared - Red) / (Near infrared + Red)

Classification methods:

We classified the input data using a Random Forest classifier with 200 trees. Data was classified into four coded land cover classes:

1 - Tree

2 - Grass

3 - Human-built structures

4 - Open water

8,961 land cover reference points were collected with 70% used to train and 30% to test the classifier. Results were smoothed using a 3x3 square kernel based on the mode of a pixel’s neighbors.

Accuracy:

Measures of accuracy including overall accuracy and per class user’s (UA) and producer’s accuracy (PA) of the random forest classifier were calculated.

Overall accuracy: 93%

Tree: UA = 89.73% PA = 93.90%

Grass: UA = 93.41% PA = 88.21%

Human-built structures: UA = 98.28% PA = 97.47%

Open water: UA = 93.56% PA = 98.95%

Code link:

The Google Earth Engine code used in this analysis is publicly available.

https://code.earthengine.google.com/32d3a77e70955a6279ec22233778bd8f

Data for download:

Two files are available for download.

1. Philadelphia_classification_points.zip

Contains a shapefile of the 8,961 reference points used to train and test the classifier.

2. Philadelphia_Landcover_2017.zip

Contains a GEOTIFF of the classified image over Philadelphia, Pennsylvania for the summer of 2017.

OverviewThis dataset contains real overhead images of wind turbines in the US collected through the National Agriculture Imagery Plan (NAIP), as well as synthetic overhead images of wind turbines created to be similar to the real images. All of these images are 608x608. For more details on the methodology and data, please read the sections below, or look at our website: Locating Energy Infrastructure with Deep Learning (duke-bc-dl-for-energy-infrastructure.github.io).Real DataThe real data consists of images.zip and labels.zip. There are 1,742 images in images.zip, and for each image in this folder, there is a corresponding label with the same name, but a different extension. Some images do not have labels, meaning there are no wind turbines in those images. Many of these overhead images of wind turbines were collected from Power Plant Satellite Imagery Dataset (figshare.com) and then hand labeled. Others were collected using Google Earth Engine or EarthOnDemand and then labeled. All of the images collected are from the National Agricultural Imagery Program (NAIP), and all are 608x608 pixels. The labels are in YOLOv3 format, meaning each line in the text file corresponds with one wind turbine. Each line is formatted as: class x_center y_center width height. Since there is only one class, class is always zero, and the x, y, width, and height are relative to the size of the image and are between 0-1.The image_locations.csv file contains the latitude and longitude for each image. It also contains the image's geographic domain that we defined. Our data comes from what we defined as four regions - Northeast (NE), Eastern Midwest (EM), Northwest (NW), and Southwest (SW), and these are included in the csv file for each image. These regions are defined by groups of states, so any data in WA, ID, or MT would be in the Northwest region.Synthetic DataThe synthetic data consists of synthetic_images.zip and synthetic_labels.zip. These images and labels were automatically generated using CityEngine. Again, all images are 608x608, and the format of the labels is the same. There are 943 images total, and at least 200 images for each of the four geographic domains that we defined in the US (Northwest, Southwest, Eastern Midwest, Northeast). The generation of these images consisted of the software selecting a background image, then generating 3D models of turbines on top of that background image, and then positioning a simulated camera overhead to capture an image. The background images were collected nearby the locations of the testing images.ExperimentationOur Duke Bass Connections 2020-2021 team performed many experiments using this data to test if the synthetic imagery could help the performance of our object detection model. We designed experiments where we would have a baseline dataset of just real imagery, train and test an object detection model on it, and then add in synthetic imagery into the dataset, train the object detection model on the new dataset, and then compare it's performance with the baseline. For more information on the experiments and methodology, please visit our website here: Locating Energy Infrastructure with Deep Learning (duke-bc-dl-for-energy-infrastructure.github.io).

国家农业影像计划(NAIP) 会在美国大陆的农业生长季获取航空影像。NAIP 项目每年都会根据可用资金和影像获取周期签订合同。自 2003 年起，NAIP 的采购周期为5 年。2008 年是转型之年，也是…

Protecting the future of forests in the United States and other countries depends in part on our ability to monitor and map forest health conditions in a timely fashion to facilitate management of emerging threats and disturbances over a multitude of spatial scales. Remote sensing data and technologies have contributed to our ability to meet these needs, but existing methods relying on supervised classification are often limited to specific areas by the availability of imagery or training data, as well as model transferability. Scaling up and operationalizing these methods for general broadscale monitoring and mapping may be promoted by using simple models that are easily trained and projected across space and time with widely available imagery. Here, we describe a new model that classifies high resolution (~1 m²) 3-band red, green, blue (RGB) imagery from a single point in time into one of four color classes corresponding to tree crown condition or health: green healthy crowns, red damaged or dying crowns, gray damaged or dead crowns, and shadowed crowns where the condition status is unknown. These Tree Crown Health (TCH) models trained on data from the United States (US) Department of Agriculture, National Agriculture Imagery Program (NAIP), for all 48 States in the contiguous US and spanning years 2012 to 2019, exhibited high measures of model performance and transferability when evaluated using randomly withheld testing data (n = 122 NAIP state x year combinations; median overall accuracy 0.89-0.90; median Kappa 0.85-0.86). We present examples of how TCH models can detect and map individual tree mortality resulting from a variety of nationally significant native and invasive forest insects and diseases in the US. We conclude with discussion of opportunities and challenges for extending and implementing TCH models in support of broadscale monitoring and mapping of forest health.

يحصل برنامج "الصور الجوية الزراعية الوطنية" (NAIP) على صور جوية خلال مواسم النمو الزراعية في الولايات المتحدة القارية. ويتم التعاقد على مشاريع NAIP كل عام استنادًا إلى التمويل المتاح ودورة الحصول على الصور. وبدءًا من عام 2003، تم الحصول على NAIP في دورة مدتها 5 سنوات. كان عام 2008 عامًا انتقاليًا، و…

The invasive shrub, Russian olive (Elaeagnus augustifolia), is widely established within riparian areas across the western United States (U.S.). Limited information on its distribution and invasion dynamics in northern regions has hampered understanding and management efforts. Given this lack of spatial and ecological information we worked with local stakeholders and developed two main objectives: 1) map the distribution of Russian olive along the Powder River (Montana and Wyoming, U.S.) with field data and remote sensing; and 2) relate that distribution to environmental variables to understand its habitat suitability and community/invasion dynamics. In the study watershed, field data showed Russian olive has reached near equal canopy cover (18.3%) to native plains cottonwood (Populus deltoides; 19.1%), with higher cover closer to the channel and over a broader range of elevations. At the basin scale, we modeled Russian olive distribution using field surveys, ocular sampling of aerial imagery, and spectral variables from Sentinel-2 MultiSpectral Instrument using a random forest model. A statistical model linking the resulting Russian olive percent cover detection map (RMSE = 15.42, R2 = 0.64) to environmental variables for the entire watershed indicated Russian olive cover increased with flow accumulation and groundwater depth, decreased with elevation, and was associated with poorer soil types. We attribute the success of Russian olive to its broad habitat suitability combined with changing hydrologic conditions favoring it over natives. This study provides a repeatable Russian olive detection methodology due to the use of Sentinel-2 imagery that is available worldwide, and provides insight into its ecological relationships and success with relevance for management across areas with similar environmental conditions. Methods Model Training Data To predict Russian olive percent cover across the Powder River Basin, we created a spectral detection model for the year 2020. The model was trained using two different data collection methods: (1) field data and (2) ocular samples from NAIP 2019 aerial imagery. Field data were collected in June 2021 (Figure 1A). Ten meter radius plots were placed on transects (25 on the east bank and 17 on the west bank) perpendicular to the river and about 50 m apart, for a total of 276 plots (Figure 1A). Within each plot, vegetation cover was estimated for each woody species, including Russian olive, plains cottonwood (Populus deltoides), and tamarisk (Tamarix ramosissima), and height of the tallest woody plant was measured using a survey rod or clinometer. Of the 276 field data plots, 185 contained Russian olive. To increase the dataset size and spatial representation, we conducted randomized ocular image sampling using NAIP 2019 true and false color imagery following a similar sampling procedure as described in (Woodward et al. 2018ab). NAIP 2019 false color imagery was referenced to help with species classification. We used Google Earth Engine (GEE) to collect 10-meter radial plots, matching the size of the field plots. We visually determined the percentage of Russian olive coverage present on a scale of 0-100 %, with 0 % being no Russian olive present and 100 % being full Russian olive cover, within each 10-meter radial plot (Figure 2). Prior to making formal observations, all five observers went through a calibration process to train and reduce bias. Due to the rarity of Russian olive in our random sample, we also opportunistically collected 478 additional plots with Russian olive. Most opportunistic aerial imagery ocular sampling points fell along the Powder River between Clear Creek and Crazy Woman Creek in Wyoming. In preliminary model runs, low to moderate Russian olive cover was unrealistically predicted in cropland areas, such as areas of Barley (Categorization Code “21”), Winter Wheat (“24”), Alfalfa (“36”), and Other Hay (“37”), so we created a simple mask to remove most crops from the final analysis. The mask was created using land cover classifications from the 2020 USDA National Agricultural Statistics Service Cropland Data Layer (NASSCD; 2021). Land cover types where Russian olive is known to occur such as Shrubland (NASSCD attribute code “152”), Grassland/Pasture (“176”), Woody Wetlands (“190”), and Herbaceous Woodlands (“195”) were retained. Table S1.2 contains a detailed list of land cover types that were not masked from the final model. All NASSCD agricultural land cover types from 0-61, 66-77, and 204-254 were excluded from the Russian olive model. The mask was also used to remove the ocular samples to build the model on sampled points that did not fall within agricultural areas. We built our model on 2,160 points (1,407 random ocular samples, 477 opportunistic ocular samples, 276 field samples), 595 of which had Russian olive present (419 ocular samples). Random Forest Model We created a mosaic of 2020 imagery from Copernicus Sentinel-2 MultiSpectral Instrument Level-1C data to cover the Powder River Basin study area, obtained in GEE. We filtered images for those with low cloud cover (<20-30 %), then created a median composite image for each relevant season – spring (2020-04-01 to 2020-05-15), summer (2020-05-16 to 2020-07-31), and fall (2020-08-01 to 2020-09-30) – to account for seasonal phenological variation (Gorelick et al. 2017). Spectral bands and vegetation indices were derived from the images, which included a Normalized Difference Vegetation Index (NDVI), Normalized Difference Moisture Index (NDMI), Normalized Burn Ratio (NBR), Simple Ratio (SR), Tasseled Cap transformation, and others (Table S1.1). The resulting Tasseled Cap brightness, greenness, and wetness (BGW) indices, named for the features they emphasize, improve vegetation classifications because they are sensitive to phenological changes (Crist and Cicone 1984). We also differenced indices between summer and spring and summer and fall to capture seasonal variation of different species to aid Russian olive detection (Evangelista et al. 2009). We modeled Russian olive percent cover and evaluated predictor variable performance using the ‘randomForest’ package in RStudio (Liaw et al. 2002) using spectral bands and vegetation indices from GEE. Our independent variable was Russian olive percent cover and all 61 predictor variables are identified in Table S1.1. The number of trees (ntree) and number of variables randomly sampled at each split (mtry) were set to 1,000 and 3, respectively (Liaw et al. 2002). We valued a model with fewer predictor variables and removed predictors that did not improve the model to achieve better model performance and greater interpretability (Evans et al. 2010). We first ran a model using all predictor variables to evaluate initial out-of-bag model performance using the R2 value and root mean squared error (RMSE), a standard measure of the magnitude of model error. We then evaluated correlations between variables, removing one variable from pairs correlated by greater than 0.7 (Dormann et al. 2013), leaving us with 18 variables after the initial run. With the remaining 18 uncorrelated variables, Wwe ran 1276 additional models using backwards selection to remove the one or two variables with the lowest variable importance as measured by the increase in mean squared error. Variables with partial dependence plots that suggested the variable contributed to over-fitting or had a weak relationship were removed (Friedman 2001). The greater the R2 value and the smaller the RMSE, the better the model performed. The final model had six variables. Finally, we summarized the random forest model results by 5 km hexagon to show trends of Russian olive cover across the study area. Species Composition and Russian Olive Habitat Suitability at the Watershed Scale Plot data collected along transects perpendicular to the Powder River channel allowed for additional insights when paired with lidar and topographic data, particularly because a robust suite of woody species was recorded in addition to Russian olive. Topographic position index (TPI) was derived from 2016 lidar data (Ackerman 2016). TPI is a measure of position by comparing elevation at a given point to the mean elevation in a surrounding window (Weiss 2001). In this case, a 100 cell (100 m) radius was used and can be interpreted as position relative to the detrended channel. We also derived a Canopy Height Model (CHM), calculated as the digital terrain model minus the digital surface model, calculated in OpenTopography.org. We extracted mean TPI, CHM, and distance from channel centerline to field plots to investigate how these varied by species by considering variable distributions (i.e., boxpolots and basic statistical moments and distributions) by dominant plot species. Additionally, we used the complete suite of species and cover data in a k-means grouping analysis that included distance from the channel in ArcGIS Map. The k-means grouping method is an unsupervised classification method where every point is assigned a group based on their similarity (Davies and Bouldin, 1979). The Pseudo F-Statistic was used to determine how many groups to include in the final analysis. This allowed inference regarding the spatial relationships among Russian olive, cottonwood, and tamarisk, which was not possible in the watershed-scale modeling. Previous work (Nagler et al. 2011) describes factors at the continental to reach scale known to influence Russian olive distribution. Robust species cover data for an entire watershed is rare. As such, here we have a unique opportunity to bridge the reach and continental scales (Nagler et al. 2011). At watershed scales, surface and groundwater flow conditions, and their regulation, are known to influence native and invasive riparian species distributions across North America (McShane et al. 2015). Surface flows have declined through time in the Powder

This raster file represents land within the Mountain Home study boundary classified as either “irrigated” with a cell value of 1 or “non-irrigated” with a cell value of 0 at a 10-meter spatial resolution. These classifications were determined at the pixel level by use of Random Forest, a supervised machine learning algorithm. Classification models often employ Random Forest due to its accuracy and efficiency at labeling large spatial datasets. To build a Random Forest model and supervise the learning process, IDWR staff create pre-labeled data, or training points, which are used by the algorithm to construct decision trees that will be later used on unseen data. Model accuracy is determined using a subset of the training points, otherwise known as a validation dataset. Several satellite-based input datasets are made available to the Random Forest model, which aid in distinguishing characteristics of irrigated lands. These characteristics allow patterns to be established by the model, e.g., high NDVI during summer months for cultivated crops, or consistently low ET for dryland areas. Mountain Home Irrigated Lands 2023 employed the following input datasets: US Geological Survey (USGS) products, including Landsat 8/9 and 10-meter 3DEP DEM, and European Space Agency (ESA) Copernicus products, including Harmonized Sentinel-2 and Global 30m Height Above Nearest Drainage (HAND). For the creation of manually labeled training points, IDWR staff accessed the following datasets: NDVI derived from Landsat 8/9, Sentinel-2 CIR imagery, US Department of Agriculture National Agricultural Statistics Service (USDA NASS) Cropland Data Layer, Active Water Rights Place of Use data from IDWR, and USDA’s National Agriculture Imagery Program (NAIP) imagery. All datasets were available for the current year of interest (2023). The published Mountain Home Irrigated Lands 2023 land classification raster was generated after four model runs, where at each iteration, IDWR staff added or removed training points to help improve results. Early model runs showed poor results in riparian areas near the Snake River, concentrated animal feeding operations (CAFOs), and non-irrigated areas at higher elevations. These issues were resolved after several model runs in combination with post-processing masks. Masks used include Fish and Wildlife Service’s National Wetlands Inventory (FWS NWI) data. These data were amended to exclude polygons overlying irrigated areas, and to expand riparian area in specific locations. A manually created mask was primarily used to fill in areas around the Snake River that the model did not uniformly designate as irrigated. Ground-truthing and a thorough review of IDWR’s water rights database provided further insight for class assignments near the town of Mayfield. Lastly, the Majority Filter tool in ArcGIS was applied using a kernel of 8 nearest neighbors to smooth out “speckling” within irrigated fields. The masking datasets and the final iteration of training points are available on request. Information regarding Sentinel and Landsat imagery:All satellite data products used within the Random Forest model were accessed via the Google Earth Engine API. To find more information on Sentinel data used, query the Earth Engine Data Catalog https://developers.google.com/earth-engine/datasets) using “COPERNICUS/S2_SR_HARMONIZED.” Information on Landsat datasets used can be found by querying “LANDSAT/LC08/C02/T1_L2” (for Landsat 8) and “LANDSAT/LC09/C02/T1_L2” (for Landsat 9).Each satellite product has several bands of available data. For our purposes, shortwave infrared 2 (SWIR2), blue, Normalized Difference Vegetation Index (NDVI), and near infrared (NIR) were extracted from both Sentinel and Landsat images. These images were later interpolated to the following dates: 2023-04-15, 2023-05-15, 2023-06-14, 2023-07-14, 2023-08-13, 2023-09-12. Interpolated values were taken from up to 45 days before and after each interpolated date. April-June interpolated Landsat images, as well as the April interpolated Sentinel image, were not used in the model given the extent of cloud cover overlying irrigated area. For more information on the pre-processing of satellite data used in the Random Forest model, please reach out to IDWR at gisinfo@idwr.idaho.gov.

InvestigatorsRay Grizzle, David Burdick, Krystin Ward, Gregg Moore, Lauren White, and Grant McKownOrganizationJackson Estuarine Laboratory, School of Marine Sciences & Ocean Engineering, University of New HampshireContactGrant McKown, james.mckown@usnh.edu or jgrantmck@gmail.comProject DescriptionRockweed macroalgae (Ascophyllum and Fucus sp.) was mapped through automated classification in Google Earth Engine using 2015 Leaf-off Imagery and 2012 - 2016 NAIP imagery. Thresholds were assigned to pre-selected remote sensing indices (Brown Algae Index, Water Index, Red - Blue Ratio, and Elevation). Preliminary classification of rockweed distribution was then refined through boat field surveys to remove erroneously classified salt marshes and overhanging canopy, which is inherent to the classification process. Accuracy assessments were carried out (n = 230 points) across the estuary and an overall accuracy of 90.5% was observed. Field survey sites (n = 25) were extracted from rockweed distribution and surveyed for eastern oyster and mussel use underneath rockweed canopy. Salinity metrics for each region of Great Bay were calculated from SWMP water quality datasondes between 2015 - 2023.Geospatial Dataset DescriptionThe geospatial datasets provided comprise both input, preliminary, and final outputs of the geospatial analysis completed in ArcGIS Pro in 2024. Geospatial assets that were used in the preliminary classification of rockweed distribution can be viewed and downloaded from the Google Earth Cloud Project (https://code.earthengine.google.com/3a3413ab29d03d866cfe6d067b6e969f). The large rasters in the geospatial dataset including aerial imagery and LIDAR are publicly-available as well across NOAA, NH Granit, and USGS EarthExplorer databases. All fields in the attributes were named to be as descriptive as possible. Explainers for the field attributes are not provided.

Narodowy program obrazowania rolnictwa (NAIP) pozyskuje zdjęcia lotnicze w okresie wegetacji w Stanach Zjednoczonych kontynentalnych. Umowy na realizację projektów w ramach NAIP są zawierane co roku na podstawie dostępnych środków i cyklu pozyskiwania zdjęć. Od 2003 r. NAIP jest nabywany w cyklu 5-letnim. Rok 2008 był rokiem przejściowym, a cykl 3-letni rozpoczął się w 2009 r. Zdjęcia NAIP są uzyskiwane z jednometrowego wzorca odległości naziemnej (GSD) z dokładnością poziomą odpowiadającą 6-metrowemu zakresowi punktów kontrolnych naziemnych możliwych do zidentyfikowania na zdjęciu, które są używane podczas kontroli zdjęć. Starsze zdjęcia zostały wykonane w 3 pasmach (czerwonym, zielonym i niebieskim – RGB), ale nowsze zdjęcia są zwykle wykonywane w dodatkowym paśmie podczerwieni (RGBN). Identyfikatory zasobów RGB zaczynają się od „n”, identyfikatory zasobów NRG zaczynają się od „c”, a identyfikatory zasobów RGBN zaczynają się od „m_”. Niektóre starsze obrazy mają GSD wynoszący 2 metry.

Chương trình Hình ảnh nông nghiệp quốc gia (NAIP) thu thập hình ảnh trên không trong mùa sinh trưởng nông nghiệp ở Hoa Kỳ lục địa. Các dự án NAIP được ký hợp đồng mỗi năm dựa trên nguồn tài trợ hiện có và chu kỳ thu thập hình ảnh. Kể từ năm 2003, NAIP được mua theo chu kỳ 5 năm. Năm 2008 là một năm chuyển đổi và …

Ulusal Tarım Görüntüleri Programı (NAIP), ABD'nin ana karasında tarım sezonlarında havadan görüntüler elde eder. NAIP projeleri, mevcut finansmana ve görüntü edinme döngüsüne göre her yıl sözleşmeye bağlanır. NAIP, 2003'ten itibaren 5 yıllık bir döngüde edinildi. 2008 geçiş yılıydı ve 2009'da üç yıllık bir döngü başladı. NAIP görüntüleri, görüntü denetimi sırasında kullanılan fotoğrafla tanımlanabilen yer kontrol noktalarına altı metre mesafede eşleşen yatay doğrulukla bir metrelik yer örnekleme mesafesinde (GSD) elde edilir. Eski görüntüler 3 bant (Kırmızı, Yeşil ve Mavi: RGB) kullanılarak toplanmış olsa da yeni görüntüler genellikle ek bir yakın kızılötesi bant (RGBN) ile toplanır. RGB öğe kimlikleri "n", NRG öğe kimlikleri "c", RGBN öğe kimlikleri "m_" ile başlar. Bazı eski görüntülerde GSD 2 metredir.

國家農業影像計畫(NAIP) 會在美國本土的農業生長期內取得空拍影像。NAIP 每年會根據可用經費和影像取得週期簽約。自 2003 年起，NAIP 的授權期限為5 年。2008 年是轉型之年，也是 …

โปรแกรมภาพถ่ายเกษตรกรรมแห่งชาติ (NAIP) จะเก็บภาพถ่ายทางอากาศในช่วงฤดูปลูกในทวีปอเมริกาเหนือ โปรเจ็กต์ NAIP จะได้รับการว่าจ้างในแต่ละปีตามเงินทุนที่มีอยู่และวงจรการได้มาซึ่งภาพ ตั้งแต่ปี 2003 เป็นต้นไป มีการขอรับ NAIP เป็นรอบ 5 ปี ปี 2008 เป็นปีเปลี่ยนผ่าน และวงจร 3 ปีเริ่มต้นขึ้นในปี 2009 ภาพ NAIP ได้มาจากการถ่ายภาพตัวอย่างภาคพื้นดิน (GSD) ที่ระยะ 1 เมตรที่มีความแม่นยำในแนวนอนซึ่งตรงกับจุดควบคุมภาคพื้นดินที่ระบุได้จากรูปภาพในระยะ 6 เมตร ซึ่งใช้ในการตรวจสอบภาพ ภาพที่เก่ากว่านั้นรวบรวมโดยใช้ 3 ย่านความถี่ (สีแดง เขียว และน้ำเงิน: RGB) แต่ภาพรุ่นใหม่ๆ มักจะรวบรวมโดยใช้ย่านความถี่อินฟราเรดใกล้ (RGBN) เพิ่มเติม รหัสชิ้นงาน RGB จะขึ้นต้นด้วย 'n', รหัสชิ้นงาน NRG จะขึ้นต้นด้วย 'c' และรหัสชิ้นงาน RGBN จะขึ้นต้นด้วย 'm_' รูปภาพเก่าบางรูปมี GSD 2 เมตร

नेशनल एग्रीकल्चर इमेजरी प्रोग्राम (एनएआईपी), अमेरिका के मुख्य हिस्से में फ़सल के सीज़न के दौरान, हवाई तस्वीरें लेता है. उपलब्ध फ़ंड और तस्वीरें लेने के चक्र के आधार पर, एनएआईपी प्रोजेक्ट के लिए हर साल अनुबंध किया जाता है. NAIP को 2003 से, पांच साल के साइकल पर हासिल किया गया था. साल 2008, बदलाव का साल था और …