This study developed the first 10-meter resolution Hawaiian Cropland Data Layers (HCDL) for 2023 using gap-filled Sentinel-Landsat multi-sensor 10-day composites and advanced algorithms. Evaluating both Random Forest and CNN models, the research found Random Forest to be more effective for operational purposes, achieving an average accuracy of 93.6%. The HCDL provides a critical tool for Hawaii's agricultural management, bridging data gaps and supporting informed decision-making.

The transition from 30m to 10m resolution in the annual Cropland Data Layers (CDL) is imperative for enhanced national-scale assessments. This paper addresses the computational challenges posed by this transition, emphasizing the need for increased efficiency in data preparation, image processing and improved crop classification accuracy. Leveraging Google Earth Engine (GEE) cloud-based platform and advanced machine learning algorithms, this paper introduces a novel operational workflow to swiftly generate a 10-meter resolution CDL for 2022 across the conterminous United States (CONUS). The workflow includes the development and use of Sentinel-2 and Landsat 8/9 derived multi-sensor gap-filled 10-day image composites, additional ancillary variables, tile-based localized analysis, and stratified random sampling strategy, leading to improved accuracy for major and specialty crops. This workflow not only reduces labor requirements but also enhances decision-making processes for agriculture and policymaking stakeholders.

Agricultural administrative data, which include information collected primarily for administrative purposes by governments and other organizations, are commonly used to improve agricultural statistics. However, one limitation of administrative data is the issue of undercoverage, or the proportion of the target population data that is not represented. If the extent and type of the undercoverage is well understood, adjustments can be made to account for it in the estimation process. The objective of this study is to develop a new method utilizing remote sensing and geospatial data to estimate the extent and type of crop specific administrative data undercoverage in two states in the United States. Preliminary results are reasonable and consistent over a five-year period from 2018 - 2022. The proposed method for estimating the extent and type of crop-specific administrative data undercoverage provides a new tool to improve crop acreage estimation in the U.S.

High-Order Markov Chains (HOMC) are conventional models, based on transition probabilities, that are used by the United States Department of Agriculture (USDA) National Agricultural Statistics Service (NASS) to study crop-rotation patterns over time. However, HOMCs routinely suffer from sparsity and identifiability issues because the categorical data are represented as indicator (or dummy) variables. In fact, the dimension of the parametric space increases exponentially with the order of HOMCs required for analysis. While parsimonious representations reduce the number of parameters, as has been shown in the literature, they often result in less accurate predictions. Most parsimonious models are trained on big data structures, which can be compressed and efficiently processed using alternative algorithms. Consequently, a thorough evaluation and comparison of the prediction results obtain using a new HOMC algorithm and different types of Deep Neural Networks (DNN) across a range of agricultural conditions is warranted to determine which model is most appropriate for operational crop specific land cover prediction of United States (US) agriculture. In this paper, six neural network models are applied to crop rotation data between 2011 and 2021 from six agriculturally intensive counties, which reflect the range of major crops grown and a variety of crop rotation patterns in the Midwest and southern US. The six counties include: Renville, North Dakota; Perkins, Nebraska; Hale, Texas; Livingston, Illinois; McLean, Illinois; and Shelby, Ohio. Results show the DNN models achieve higher overall prediction accuracy for all counties in 2021. The proposed DNN models allow for the ingestion of long time series data, and robustly achieve higher accuracy values than a new HOMC algorithm considered for predicting crop specific land cover in the US.

A novel approach for crop-specific prediction of future crop planting and the development of corresponding uncertainty measures for all predictions is proposed. Using transition probabilities, predictive crop categories are first developed to predict crop-specific planting in the pilot study state of Illinois. Corresponding entropy layers are developed concurrently and can be used to flag survey sample units based on the level of uncertainty associated with the crop predictions. This allows survey units to be prioritized for imputation or for data collection based on whether the predicted value is sufficiently reliable. Further, the predicted acreage is assigned only to those survey units for which the prediction has the potential to be the sufficiently accurate. This approach can provide a solid methodology for reducing survey costs and farmer response burdens without introducing estimation bias or incurring severe losses of statistical efficiency. This has far-reaching implications for the sample design, quality, and timeliness of results for future surveys.

Early season crop identification is important for food security and economic stability. The USDA NASS uses optical data to provide acreage estimates, each June, to the NASS Agricultural Statistics Board. However, early season crop identification is difficult using optical data alone, because imagery is frequently cloudy during the spring. The purpose of this study is to determine whether using SAR and SAR texture can improve early season winter wheat identification compared to optical data alone. Study areas in the Missouri "Bootheel" (2017 growing season) and Northwest Texas (2018 growing season), United States (U.S.) are selected. The SAR data used in this study are Sentinel-1. Optical data include: Landsat 8, Disaster Monitoring Constellation, and Sentinel-2. Study results show that optical data with SAR achieved the highest winter wheat accuracies, 7.7% higher than optical data alone, in Missouri. Optical with SAR and SAR texture resulted in improved accuracies over optical alone, but only marginally, in Texas. These results indicate that optical and SAR, used together, can potentially improve early season crop identification.

Agricultural flood monitoring is important for food security and economic stability. Synthetic Aperture Radar (SAR) has the advantage over optical data by operating at wavelengths not impeded by cloud cover or a lack of illumination. This characteristic makes SAR a potential alternative to optical sensors for agricultural flood monitoring during disasters. The purpose of this study is to assess the effectiveness of using freely available Copernicus Sentinel-1 SAR data for operational agricultural flood monitoring in the United States (U.S.). The operational detection of flood inundation was tested during Hurricane Harvey in 2017, which resulted in significant flooding over Texas and Louisiana, U.S. This paper presents 1) the agricultural flood monitoring method that utilizes Sentinel-1 SAR, the NASS 2016 Cultivated Layer, and the NASS 2016 and 2017 Cropland Data Layers; 2) flood detection validation results and 3) inundated cropland and pasture acreage estimates. The study shows that Sentinel-1 SAR is an effective and valuable data source for operational disaster assessment of agriculture.

Area Sampling Frames are used for surveys including crop acreage and yield, forests, and natural resource inventories and are the foundation of the statistical program of the USDA National Agricultural Statistics Service (NASS) and many statistical survey programs around the world. An automated area frame stratification method was recently implemented into NASS operations, which is based on the objective calculation of percent cultivation derived from the NASS geospatial Cropland Data Layers (CDLs). While autostratification consistently outperforms manual stratification in cultivated areas, we found that CDL-based pixel counting estimation consistently underestimated crop acreage. Previous research indicates that CDL classification accuracy is affected by training data pixel level buffering. We hypothesize that training data pixel level buffering will also affect the CDL based auto-stratification results and crop acreage estimation. This paper evaluates the impact of training data buffering on area frame stratification results and crop estimates. Preliminary results indicate that the crop acreage underestimation can be directly attributed to the training data pixel level buffering procedure.

Area Sampling Frames (ASFs) are the basis of many statistical programs around the world. To improve the accuracy, objectivity and efficiency of crop survey estimates, an automated stratification method based on geospatial crop planting frequency and cultivation data is proposed. This paper investigates using 2008–2013 geospatial corn, soybean and wheat planting frequency data layers to create three corresponding single crop specific and one multi-crop specific South Dakota (SD) U.S. ASF stratifications. Corn, soybeans and wheat are three major crops in South Dakota. The crop specific ASF stratifications are developed based on crop frequency statistics derived at the primary sampling unit (PSU) level based on the Crop Frequency Data Layers. The SD corn, soybean and wheat mean planting frequency strata of the single crop stratifications are substratified by percent cultivation based on the 2013 Cultivation Layer. The three newly derived ASF stratifications provide more crop specific information when compared to the current National Agricultural Statistics Service (NASS) ASF based on percent cultivation alone. Further, a multi-crop stratification is developed based on the individual corn, soybean and wheat planting frequency data layers. It is observed that all four crop frequency based ASF stratifications consistently predict corn, soybean and wheat planting patterns well as verified by the 2014 Farm Service Agency (FSA) Common Land Unit (CLU) and 578 administrative data. This demonstrates that the new stratifications based on crop planting frequency and cultivation are crop type independent and applicable to all major crops. Further, these results indicate that the new crop specific ASF stratifications have great potential to improve ASF accuracy, efficiency and crop estimates.

This paper proposes a novel method for land cover area frame stratification based on corn planting frequency and percent cultivation. South Dakota U.S. geospatial crop frequency (2008-2013) and cultivation (2013) data layers created from NASS Cropland Data Layers are utilized to develop a novel area sampling frame (ASF) stratification design. Eight corn planting frequency strata are derived using a k-means clustering method based on mean corn planting frequency calculated at the NASS ASF primary sampling unit level. The corn planting frequency strata are then sub stratified based on percent cultivation, which, together, provide more crop specific information than the current NASS ASF based on percent cultivation alone. Using 2014 Farm Service Agency Common Land Unit Data as in situ validation, it is found that this novel ASF design predicts crop specific planting patterns well. These results indicate that the new stratification method has potential to improve ASF accuracy, efficiency and crop estimates.

Information on future crop specific planting is valuable for improving agricultural survey estimates. This information is critical for agricultural production planning, agricultural product commodity inventory control, natural resource allocation and conservation, etc. However, future crop planting details are generally unavailable. This paper proposes to use crop specific planting frequency data as indicators to indirectly provide information regarding future crop planting. A methodology to derive crop planting frequency data layers based on 2008-2013 Cropland Data Layers has been presented in this paper. Crop frequency layers for corn, soybeans, wheat and cotton were successfully built at the national level and for two states including Indiana and Mississippi, USA. Multi-year (2008-2013) Farm Service Agency (FSA) Common Land Unit (CLU) data were utilized to assess the accuracy of the derived crop frequency data layers. The accuracies of the national scale crop frequency data layers are 91.00%, 90.13%, 87.67% and 85.96% for corn, cotton, soybean and wheat respectively.

Multispectral satellite images have been utilized in the National Agricultural Statistics Service (NASS) for crop cover classification and crop acreage estimation since the 1970's. Though ancillary data is utilized to enhance the classification accuracy, there are few applications that maximize the utilization of the feature information of the given multispectral images. Every multispectral image band directly provides the specific spectral response to a given land cover category. The different combinations of band ratios or vegetation indices enhance spectral characteristics of some crops while suppressing others. Therefore, various vegetation indices and image ratios of Landsat images have been extensively studied and applied to identify various land cover and land use characteristics in the past. However, NASS began using the ResourceSat-1 AWIFS sensor for operational crop classification and acreage estimation in 2006. The AWIFS’ bands are different from those of Landsat, and there is sparse literature published about research and applications of the spectral characteristics of AWIFS image band ratio and vegetation indices. In this paper, the impact of using band ratio and vegetation indices of the AWIFS images to the crop classification accuracy is empirically investigated via supervised classification. The classification results with respect to the additional vegetation index and band ratio are presented and compared in terms of the overall and crop only classification accuracy. The research indicates that appropriately used vegetation indices and image ratios can potentially improve crop classification accuracy though the gain may not be huge. It is concluded that further research is needed.

The ongoing drought in California substantially reduced surface water supplies for millions of acres of irrigated farmland in California's Central Valley. Rapid assessment of drought impacts on agricultural production can aid water managers in assessing mitigation options, and guide decision making with respect to mitigation of drought impacts. Satellite remote sensing offers an efficient way to provide quantitative assessments of drought impacts on agricultural production and increases in fallow acreage associated with reductions in water supply. A key advantage of satellite-based assessments is that they can provide a measure of land fallowing that is consistent across both space and time. We describe an approach for monthly and seasonal mapping of uncultivated agricultural acreage developed as part of a joint effort by USGS, USDA, NASA, and the California Department of Water Resources to provide timely assessments of land fallowing during drought events. This effort has used the Central Valley of California as a pilot region for development and testing of an operational approach. To provide quantitative measures of uncultivated agricultural acreage from satellite data early in the season, we developed a decision tree algorithm and applied it to timeseries of data from Landsat TM, ETM+, OLI, and MODIS. Our effort has been focused on development of indicators of drought impacts in the March August timeframe based on https://ntrs.nasa.gov/search.jsp?R=2016001276