Search for: All records

Non-crop vegetation, such as hedgerows and cover crops, are important on-farm diversification practices that support biodiversity and ecosystem services; however, information about their rates and patterns of adoption are scarce. We used satellite and aerial imagery coupled with machine learning classification to map the use of hedgerows/windbreaks and winter cover crops in California's Central Coast, a globally important agricultural area of intensive fresh produce production. We expected that adoption of both practices would be relatively low and unevenly distributed across the landscape, with higher levels of adoption found in marginal farmland and in less intensively cultivated areas where the pressure to remove non-crop vegetation may be lower. Our remote sensing classification revealed that only ~6% of farmland had winter cover crops in 2021 and 0.26% of farmland had hedgerows or windbreaks in 2018. Thirty-seven percent of ranch parcels had cover crops on at least 5% of the ranch while 22% of ranches had at least one hedgerow/windbreak. Nearly 16% of farmland had other annual winter crops, some of which could provide services similar to cover crops; however, 60% of farmland had bare soil over the winter study period, with the remainder of farmland classified as perennial crops or strawberries. Hotspot analysis showed significant areas of adoption of both practices in the hillier regions of all counties. Finally, qualitative interviews revealed that adoption patterns were likely driven by interrelated effects of topography, land values, and farming models, with organic, diversified farms implementing these practices in less ideal, lower-value farmland. This study demonstrates how remote sensing coupled with qualitative research can be used to map and interpret patterns of important diversification practices, with implications for tracking policy interventions and targeting resources to assist farmers motivated to expand adoption. 

Based on the established task of identifying boosted, hadronicallydecaying top quarks, we compare a wide range of modern machine learningapproaches. Unlike most established methods they rely on low-levelinput, for instance calorimeter output. While their networkarchitectures are vastly different, their performance is comparativelysimilar. In general, we find that these new approaches are extremelypowerful and great fun.