Direct observation of feed bunk behaviors would ultimately be needed to positively determine the underlying mechanism for this behavioral pattern among this subset of animals.In order to account for this informational redundancy, age in days at start of trial and the 95th quantile of daily milk yield were here scaled to uniform variance and jointly encoded using the ecodePlot utility . Visualizations of this encoding can be found in Supplemental Materials. Bivariate tree tests, however, returned no significant associations between this age-yield encoding and any of the tensor plot encodings of daily time budgets. This result differed slightly from analyses of overall time budgets, wherein encodings that accounted for heterogeneity in sensor error and plasticity in daily response also returned no significant association, but encodings that accounted only for sensor noise found that heifers were significantly under represented among the most moderate time time budgets . Collectively, these results suggest that plasticity in daily time budgets might be obscuring links to age and productivity, but that this relationship cannot be brought into resolutions by accounting for systematic responses to environmental factors alone. This could occur if there were heterogeneity in the lag time between an environmental stressor and the behavioral response, or if there are more complex interactions occurring between transient environmental variables and more persistent internal biological states. If this in fact the case, then this result may suggest that an encoding strategy capturing only the variability in the response, trim tray screens and not systematic fluctuations, may in fact be a more effective means of pinning down this bivariate association.
Visualizations of these associations are provided in Supplemental Materials. As with overall time budgets, these resultsrevealed that cows with no health complications were over represented among the most moderate time budgets while sick cows were over represented among cows with low time spent eating. Overall, however, these associations were stronger with encodings of overall time budgets than with daily time budgets. Bivariate associations were also evaluated against health records disaggregated into five broad categories: healthy, mammary infection and injury, hoof and leg lameness, infections of the reproductive tract, and digestive and metabolic diseases. Significant associations were recovered for all three dissimilarity estimators used to encode daily time budgets. Visualizations of these associations can again be found in Supplemental Materials. Again, these results mirror the bivariate patterns recovered for overall time budgets, with no clear evidence that temporal dynamics notably modified these associations. Figure 4 displays the strongest association recovered for the ensemble-weighted encoding, wherein the optimal metaparameter combination for the tensor mechanics encoding employed no re-weighting along the temporal axis. As with overall time budgets, mammary infections were over-represented among cows with consistently extremely high time spent eating, while digestive and metabolic diseases were over-represented among low eating time budgets . Infections of the reproductive tract were also again over-represented among the most moderate time budgets, but in this encoding with daily time budgets, it is more visually apparent that these same time budgets are also the most variable over time. As these observations occurred fairly early in the lactation period, it seems unlikely that temporary perturbations in behavioral patterns attributable to estrus behaviors could fully explain so much variability in eating times .Finally, significant bivariate associations were recovered between patterns in milking order and encodings of daily time budget using all three dissimilarity estimators.
Visualizations of these bivariate patterns are provided for all encodings in Supplemental Materials. As a whole, the bivariate associations recovered against encodings created using overall time budget were stronger than for encodings using daily time budgets . Results for the unweighted Euclidean distance and the KLD distance largely mirrored patterns recovered using overall time budgets, with extremely low eating times being under represented among cows entering at the end of the queue and over represented among animals just ahead of these caboose animals. Visualizations of the bivariate association recovered against the ensemble-weighted Euclidean distance encoding are provided in Figure 5. Here the pattern between time budgets and cows that enter nearer the rear of the milking queue are brought into slightly higher resolution. This visualization shows that it is cows that consistently enter at the end of the queue, but not the last handful of cows, that are over represented among time budgets with moderate-to-high time spent eating. Time budgets with consistently low time spent eating, on the other hand, are significantly over represented among cows that routinely jumped between the middle and rear of the milking queue. Collectively, these results confirm that a later milking position does not preclude a cow from investing a significant amount of time at the feed bunk, though it should be noted that these results may not extrapolate to larger milking groups with longer waiting times in the milk parlor . As systematic temporal patterns in daily time budgets do not, however, appear to play a significant role in further distinguishing these bivariate patterns, future work might consider if these asynchronous perturbations in queuing patterns and time budgets might be more directly linked within individual, and consider latent biological causes of these isolated behavioral responses.Drawing complete and holistic quantitative inferences is always challenging when working with multivariate data streams. When a temporal component is also present, information compression via data aggregation can simplify the analytical pipeline, but a considerable amount of behavioral complexity can be lost.
In analyses of Cow Manager rear tag accelerometer records available in the Organilac dataset, tensor mechanics analyses revealed that cows in this experimental herd were remarkably consistent in their overall time budgets. Never the less, these algorithms were still able to capture subtle shifts in the tradeoffs between the eating and nonactive axes and the eating and highly active axes as cows progressed in their lactations, even though such temporal patterns were only found amongst a subset of the herd. While the resulting encodings did not differ substantially from overall time budgets, these analyses helped to better visualize differences between animals in the relative plasticity of their behavioral responses. While tensor mechanics algorithms are designed to extract systematic changes over time by leveraging collective responses of animals to a shared environment, these results have served to highlight the need to look beyond the mean value and more fully explore the richness of PLF datasets, which can provide repeated measurements on individuals at a scale never before possible .Future work should consider how to further leverage these approaches to not only explore patters in such plastic responses that can be extracted by leveraging group level information, but also variability attributable only to the individual, and subsequently any complex interactions that may occur between these systematic and fleeting perturbations in behavioral responses. Rice is the most widely consumed staple food for a large part of the human population, especially in Asia, trimming tray with screen providing more than one fifth of the calories consumed by humans worldwide . In many Asian countries, rice accounts for more than 70% of human caloric intake. China is the world’s largest rice producer, accounting for 30% of the total world production, followed by India , Indonesia , and Bangladesh . The traditional method of rice cultivation in the world used to be transplanting rice , which ensured a steady yield during the long history of mankind . However, in all climatic zones, human labor represents more than 50% of the cost of TPR farming , followed by the cost of other inputs such as water and fertilizers . With the development of rice cultivation science and the requirements of different climatic zones, many other methods of rice cultivation gradually emerged, such as dry or wet direct rice seeding . Direct-seeded rice , which is cultivated by directly broadcasting seeds onto the topsoil of paddy fields without needing to raise and transplant seedlings, provides an opportunity to save both labor and time .
This farming method also enables earlier crop establishment, providing an opportunity to make better use of early season rainfall, while increasing crop intensification in some rice-based systems . Moreover, the development of early-maturing varieties and improved nutrient management techniques, along with increased availability of chemical weed control methods, have encouraged many farmers in Asia to switch from transplanted to direct-seeded rice culture . During the last two decades, the change in the method of crop establishment from manual transplanting of seedlings to directseeding has occurred in many Asian countries in response to rising production costs, especially those of labor and water . In the Taihu Lake Basin of east China, many farmers have accepted the cultivation of DSR, although the average yield of DSR is not as stable and is still slightly lower than that of TPR . The area with DSR cultivation has rapidly increased and has already exceeded 50% of the total farmland in many TLB regions . In accordance with this development trend, the direct-seeding approach will likely continue to remain popular in the TLB. Water flow in paddy fields with cultivated rice involves the interaction of very complex processes, and their observation and evaluation under field conditions is relatively difficult, costly, and time consuming. Therefore, a large number of scientists increasingly use computer models to study the complex processes in the soil and to provide management and planning guidance. Hydrus-1D and Hydrus are numerical models that have often been used by many researchers to simulate water flow in agricultural fields with different crops and various irrigation schemes , including TPR fields . However, the Hydrus-1D model has not yet been used for simulating water flow in DSR fields. Compared to traditional TPR, DSR requires different water management, which provides rice with a different growth environment, particularly during its seedling stage . During the first two weeks after seeding, rather than being flooded as with TPR, the top soil only needs to keep sufficient moisture to allow for seed germination . As a result, the root mass of DSR is distributed shallower than that of traditional TPR, which consequently produces different vertical profiles of the water content. Furthermore, compared to TPR, DSR prefers an alternative drying and wetting soil environment during the middle late season when multiple smaller irrigations can benefit both the plant growth and deeper root growth . This water management produces distinctly different characteristics of the water flow regime and water losses from DSR fields compared to TPR fields. In this study, field observations in a DSR field in the TLB during two consecutive rice-growing seasons are evaluated using Hydrus-1D, and the main characteristics of the water flow regime and water losses are discussed.The agricultural land in the Taihu Lake Basin is used for very intensive production of the rice crop. The basin area, which is located south of the Yangtze River, is approximately 36,900 km2, with rice fields accounting for about 34.8% of this area. Rotations of rice with either wheat or rape are the most popular cultivating modes in this region. The basin has a subtropical monsoon climate with average annual rainfall of 1181 mm, 60% of which occurs from May through September. The annual PAN evaporation from the water surface is approximately 822 mm, and the average annual air temperature is 15–17 ◦C. The study site is in the Dangyang region , upstream of Taihu Lake . The dominant soil type in this region is classified as a hydromorphic paddy soil, and the parent material is a lacustrine deposit. The physical and chemical properties of the soil at the site are listed in Table 1.In our experiments, the variety of rice used for the DSR cultivation was Wuxiangjing 14 , a type of Japonica rice thatis predominantly cultivated in the Dangyang region. The observations were carried out in the same field during two growth seasons. After the mechanical field preparation, the seeds were evenly broadcasted by hand on the soil surface at 75 kg/ha, without prior soaking, on June 8th in 2008 and June 11th in 2009. After seeding, the fields were irrigated until the surface soil was saturated . The harvest dates were on November 1st in 2008 and November 5th in 2009. The total growing periods during these two years were thus 147 and 149 days, respectively. The water management in the DSR field followed instructions from the local Agricultural Technical Guidance Station and drew from the farmers’ own experience.