USGS Science⌕ Search
Home/Archive/USGS

USGS · 70259229

Crop type classification, trends, and patterns of central California agricultural fields from 2005 to 2020

Britt Windsor Smith·Christopher E. Soulard·Jessica J. Walker
View original source ↗Export metadata ↓

Abstract

California produces many key agricultural products in the United States. Current geospatial agricultural datasets are limited in mapping accuracy, spatial context, or observation period. This study uses machine learning and high-resolution imagery to produce a time series of crop maps to assess crop type trends and patterns across central California from 2005 to 2020. National Agriculture Imagery Program and Landsat imagery were used to classify nine crop types that are common in the study region: grain crops, field crops, rice, citrus and subtropical, deciduous fruit and nut, vineyard, berry and vegetable, pasture, and fallow/young perennial crop types. To create labeled data, we sampled 1253 fields and manually identified crop types for each examined year using high-resolution imagery and Landsat normalized difference vegetation index time series. We applied a random forest machine learning algorithm in Google Earth Engine. Results show that the mean overall classification accuracy of the nine-class map was 93.1%, with individual accuracies ranging from 99.3% (rice) to 89.5% (fallow/young perennial). Mann–Kendall trend tests showed significant ( p less than 0.05) declines in field crop and pasture area during the study period, while deciduous fruit and nut, citrus and subtropical, and fallow/young perennial crop types experienced significant increases. At an aggregate level, there was a general shift from annual crop types to perennial crop types. These data provide a 16-year time span of spatially explicit crop type classifications, trends, and patterns in central California that can be used to aid managers and decision makers for resource planning or hazard mitigation.

Explore related subjects

Agrosystems, Geosciences & EnvironmentArticle
90° N90° S · 180° W ← longitude → 180° E
Source-reported bounding extent: 34.420043865180986° to 40.45021878284146° latitude; -124.11418123355082° to -118.14248794924036° longitude. This indicates report coverage, not an exact sampling location. View area on OpenStreetMap.

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Britt Windsor Smith, Christopher E. Soulard, Jessica J. Walker. 2024-07-25. Crop type classification, trends, and patterns of central California agricultural fields from 2005 to 2020. https://doi.org/10.1002/agg2.20553

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related USGS reports

USGS2025

Patterns of water-extractable soil organic matter in the US Great Plains: Insights from the Haas Soil Archive

Novel approaches that are fast and sensitive are needed to evaluate soil change and integrate soil ecosystem properties. Carbon (C) and nitrogen (N) extracted from soil with water are associated with plant nutrients and microbial activity but information about change over time in the US Great Plains is sparse. We used cool (20°C) and hot (80°C) water extracts from historic (1947) and contemporary (2018) soil samples collected at Moccasin, MT; Akron, CO; and Big Spring, TX; to examine changes to labile C and N and optical properties after 71 years of dryland cropping. Concentrations of C and N extracted with cool water decreased between 1947 and 2018 in surface (0–15.2 cm) samples from Moccasin, by 52% and 35%, and Big Spring, by 37% and 32%, but remained unchanged at Akron. Conversely, net (hot−cool) extractable C did not change at Moccasin or Big Spring but increased at Akron by 26%. Net extractable N decreased at Moccasin by 22% but did not change elsewhere. Sequential principal component analysis and stepwise discriminant analysis identified three important optical properties. Values of SUVA 254 (where SUVA 254 is the specific ultraviolet absorbance at 254 nm) in extracts did not change at Moccasin between 1947 and 2018 but increased at Akron, indicating increased aromaticity. Conversely, SUVA 254 decreased at Big Spring. Values for Sag 350–400 (where Sag 350–400 is the slope from a nonlinear fit of an exponential function to the absorption spectrum over the wavelength range from 350 to 400 nm), inversely related to extract molecular weight and aromaticity, decreased at Moccasin but not elsewhere. The proportion of recalcitrant to labile compounds, C:T (where C:T is the ratio of fluorescence intensity from Peak C [ex340/em440] to Peak T [ex275/em340]), increased in extracts from all sites but especially at Akron. Together, these methods provided insights into soil change while conserving samples.

Colorado, Kansas, Montana, Nebraska, New Mexico, N↗