Given that our study area is surrounded by areas of intensive agriculture , we examined associations with surrogates for potential sources of airborne endotoxin [agricultural land cover and animal-feeding operations , urban parks, and schools] to assess whether proximity to these sources would have predictive power for spatial mapping. Specifically, we investigated associations of daily endotoxin levels with the area of various land covers within 10-km and 20-km radius buffers of the endotoxin measurement locations. Spatially resolved land cover data for cropland, grasslands, and pastureland in and around Fresno were obtained from the 2001 National Land Cover Data database . The locations of confined animal-feeding operations were obtained from the California Department of Conservation . Urban parks and schools were included because the presence of dogs and dog waste are associated with elevated endotoxin levels, and owners frequently walk dogs at these types of facilities.Regression analyses were conducted to assess whether proximity to these potential sources of outdoor endotoxin have predictive power for spatial mapping. These regressions did not prove useful for mapping. The endotoxin concentrations tended to be higher in the outlying areas than in the urban core; therefore, hydroponic shelf system models developed from the urban core data always underestimated the higher values in the outlying areas.
We used simple spatial mapping to estimate the average spatial patterns of pollutants [see Supplemental Material ]. Spatial mapping was performed before temporal averaging because of the limited amount of temporal data in the FACES data set. Separate maps were generated on each of the 22 dry-season days. Concentration estimates were made for a grid of points with 0.25-km spacing over a 50 × 60 km domain centered on Fresno. Average dry-season maps were constructed by time averaging the daily mapped concentrations. The accuracy of the mapping method was evaluated by removing one data point at a time.Because of the potential importance of endotoxin in the pathogenesis of asthma , we have characterized the factors that influence its temporal and spatial variability as part of a study of the natural history of asthma in children who live in an urban area surrounded by large areas of agricultural land and whose air quality is influenced by two heavily trafficked highways that pass through it. Daily variability in endotoxin concentrations could be characterized by a common set of physical variables and variable specifications at two different locations approximately 5 km apart and at different distances from major agricultural areas that surround the study area [Figure 1; Supplemental Material, Table S2 ]. We did not have daily data on potential source emissions, which explains, in part, why our model accounted for only about 50% of each day’s variability. Furthermore, the model parameters and the form of the variables based on First Street data provided a better fit to the Fremont data than interpolation of First Street data to the Fremont site. The larger intercept for the Fremont model reflects the year-round higher levels at this location than at First Street.
Decreased relative humidity and greater recirculation of air masses with low wind speeds were associated with increased endotoxin concentrations. These conditions often coincide with summertime air inversions in the San Joaquin Valley characterized by high ozone concentrations . It is well known that components of the bioaerosol that increase during the ozone season in Southern California can increase the occurrence of wheeze in children with asthma . The specific importance of endotoxin in this setting has not yet been evaluated in any detail. High endotoxin concentrations were measured on days with both high and low PM2.5and PMc concentrations. Moreover, high endotoxin concentrations were most frequently observed on days when the concentrations of PM2.5 were below the current daily national standard of 35 µg/m3 . Given the proinflammatory properties of endotoxins , ambient endotoxin concentrations likely play a role in respiratory outcomes associated with PM. In our study area, endotoxin concentrations are highest during dry seasons. However, in other study areas with different climatic conditions and likely sources of endotoxin, the highest concentrations may be at other times of the year . The ambient concentrations we observed are higher than airborne concentrations found indoors in other studies: Airborne endotoxin concentrations have been associated with respiratory illness in children in the first 2 years of life , and house dust endotoxin concentrations have been associated with wheezing . Endotoxin concentrations in Fresno generally are higher than those reported for 13 Southern California locations that included desert, coastal, and inland areas . Based on 8 sampling days spread over 1 year, that study reported a geometric mean concentration was 0.34 EU/m3 across all sites and 1.85 EU/m3 at the highest site .
In Fresno, the geometric mean of year-round daily samples was 1.44 EU/m3 at Fremont and ranged from 0.98 to 1.38 EU/m3 at First Street., The maximum daily concentration observed in Southern California was 5.5 EU/m3 compared with 9.4, 12.4, and 16 EU/m3 observed at First Street, Fremont, and a Fresno residence, respectively. A June–September study in Palo Alto, California, a suburban area south of San Francisco that is not surrounded by large tracts of agricultural land, reported a geometric mean outdoor concentration of endotoxin in the PM10 fraction of 0.7 EU/m3 , which is considerably lower than similar months over the 3 years of our study . The levels we observed are one to two orders of magnitude lower than those reported in proximity to specific sources found in our study area, such as a large dairy farms and other forms of animal husbandry . Our analysis shows that the spatial patterns of endotoxin, PM2.5, PMc, and EC are distinct and that the spatial pattern of endotoxin concentrations does not mirror any other conventionally measured pollutants in Fresno. The only similarities across pollutants are the tendency for lower concentrations in the north or northeastern areas, which are bounded by native vegetation and far from Highway 99, and higher levels along and southwest of Highway 99, which is close to large areas of agriculture that include CAFOs. The differences in spatial patterns suggest differences in the locations and strength of the emission sources for different pollutants and, in the case of PM2.5, the influence of secondary aerosol. The daily endotoxin values at homes and schools were not reliably predicted from measurements at First Street alone or from First Street and the four schools in the urban core. These urban core measurements tend to be lower than those collected in the outlying areas , which may limit their usefulness for predicting the broader pattern. Understanding of the strengths and locations of the endotoxin emission sources is quite limited; these data suggest the sources are more likely outside rather than inside the urban core. One potential major source for which we have no data relates to patterns of dog ownership and walking patterns.
In indoor environments, dogs are an important source of airborne endotoxin. For example, Park et al. reported that presence of a dog in homes accounted for 15% of the variance of airborne endotoxin, more than twice as much as any other factor. And studies have found that the presence of a dog is a major contributor to house dust and indoor air endotoxin concentrations . Although not based on a random sample of Fresno,cannabis drying racks commercial subjects in the southwest quadrant had the lowest reported prevalence of dog ownership , which suggests that dogs are not likely to be a major source of ambient endotoxin, at least in this quadrant. An indirect assessment of agricultural source contribution is evident in that during all dry months, endotoxin concentrations were higher at Fremont than at First Street [median over study period: Fremont = 4.5 ; First Street = 2.5 ]. Relative to First Street, the Fremont site is 4.1, 4.0, and 4.3 km closer to the western, southwestern, and southern boundaries of the urban core that abut on large tracts of agricultural land, respectively, with the First Street site being 6.1, 4.5, and 6 km from these boundaries, respectively. Moreover, the ratio of the quadrant-specific endotoxin concentrations relative to concentrations at First Street were greater in the southwest quadrant during both warm and dry seasons, but particularly during the dry season . A summertime study of the dispersion of endotoxin at various downwind distances from a 10,000-cow dairy in Idaho found that endotoxin concentrations decreased exponentially and reached upwind concentrations at about 1,390 m downwind . Wind speeds were similar to those observed in our study . Although the Fremont and First Street sites are downwind of the dominant flows, both are substantially farther downwind of sources than in the Idaho study. Thus, in our study the endotoxin concentrations would be expected to be lower than those in that study, but higher at Fremont than at First Street. However, concentrations of endotoxin measured at 10 homes closest to the west edge of the urban boundary , on average, did not have a average higher than 47 other homes further away . Potential limitations of our spatial analysis relate to use of only 22 days in one dry season and some measurements made at different locations on different days. However, the pattern on these days is fairly robust. We found, using a leave-one-out approach, that the average spatial pattern was insensitive to the data for any specific day . The representativeness of the spatial patterns for other time periods remains uncertain because of the lack of independent data to compare.Precision livestock farming, which collects detailed measurements of animals through sensors or cameras, has become an important research focus with the rising demand of animal production . Monitoring animal activity can facilitate the management of animal production, and it is conventionally conducted by frequently visiting the farms or manually reviewing recorded videos . However, these approaches can be subjective and laborious. Alternatively, the technology of computer vision , which is inspired by human vision that can intuitively focus on objects of interest and exclude noisy signals, can automate monitoring animal activity solely based on information obtained from video recordings. Automation is fulfilled by deriving imagery features from a series of computational tasks, such as video segmentation and edge detection, guiding the CV models to recognize animals’ body contours, positions, and behavioral categories. With the CV technology, many studies have shown promising results in practicing precision livestock farming. For example, it used to be costly to manage cattle in large-scale pasture lands. Coupling with unmanned aerial vehicles, CV is possible to automate cattle counting in real-time with labor costs substantially reduced . In smaller-scale indoor farms, CV systems were also used to detect body cleanness , entirety , structure , and behaviors for animal production. This technology is particularly beneficial to the swine industry, as pigs are usually group-housed in indoors settings. By deploying one top-view RGB colors camera, producers can track pig activity by capturing their positions and identities in a high-throughput manner . To assess complicated traits that are labor intensive to be measured, the deployment of multiple cameras or RGB-D depth sensing cameras can provide extra dimensions of information to enhance the CV system. Many successful applications have also demonstrated automation in assessing body weight , feeding behaviors , and more precise measurement of real-time pig positions in recent years. Nevertheless, challenges still exist in the current CV systems and make CV difficult to be widely implemented in most farming environments. First, the performance of CV systems is sensitive to the imagery features, which are derived based on the observed imagery patterns under certain environmental conditions . When the CV system is deployed in a new environment, its performance may decrease as the features are not generalized enough to be associated with pig morphological patterns under different illumination conditions . Second, most CV systems are established by supervised learning, in which intensive effort in labeling ground truths for the training process is required. The insufficient size or quality of training datasets may be a harmful factor for the model robustness. Lastly, many successful CV systems are built by deep learning models, in which tens of millions of unknown parameters are estimated . Such nature makes CV systems challenging to implement, limiting a wider deployment of high-throughput monitoring in animal industry.