FarmOS have not only defined data structures for many agricultural entities, but made it trivial to expand them in order to develop custom data structure’s using their schema. Agricultural data privacy laws surely have a ways to go in order to protect farmers from untrustworthy institutions. This issue, however, paired with the problem of data being leaked in breeches has led to the extensive research and development of new technologies which put precedence on data security and allow for operations, especially in machine learning, to be performed without compromising security. These new technologies include trusted execution environments , cloud operated machine learning as a service and fully homomorphic encryption amongothers. In this paper I express a collection of architectures built on a combination of some of these technologies as well as others for trusted and secure machine learning model training. A key software utilized is FarmStack. Digital Green is developing FarmStack as a peer-to-peer network protocol which secures data in transit through periodic attestation, network policy enforcement and endpoint application enforcement. Providing agricultural data owners with the means to allow others to use their data securely is the primary goal of the proposed data sharing and model training network architectures.Soil salinity is a known constraint on agricultural production in the Central Valley, particularly in the western San Joaquin Valley , where soils are naturally high in salts due to the marine origin of their Coastal Range alluvium parent material . In such a large region, it is difficult to quantify and map the full extent of soil salinity and its impact on agricultural production and profits. Many geological, meteorological and management factors affect the salinity levels of irrigated soils,raspberry cultivation pot including irrigation water quality, irrigation management, drainage conditions, rainfall and evapotranspiration totals and cultural practices.

Across a region such as WSJV, most of those factors vary at multiple spatial and temporal scales, making it difficult to extrapolate local point measurements of soil salinity to regional scales. Although agricultural salinity is a generally well known issue, communicating the full extent and severity of the problem to policymakers, stakeholders and other nonspecialists is a challenge. Detailed regional maps present the problem visually and can help spur action on planning, management and conservation. Letey argued that long-term sustainable and profitable agriculture in California can be achieved only if regional-scale salt balances can be obtained. Regional-scale salinity maps provide irrigation district managers, water resource specialists and state and federal authorities with timely information that can guide decisions on water allocation needs and groundwater regulation 2015. We qualitatively evaluated the correspondence of remote-sensing high salinity predictions with the presence of salt crusts. To map salt crusts across the WSJV, we used imagery from the 2014 USDA’s National Aerial Imagery Program survey . A supervised classification was used to identify salt crusts. The classification identified NAIP pixels with reflectance properties similar to those observed at locations known to be affected by salt crusts. This analysis identified salt crusts over 0.5% of WSJV farmland. Figure 3A depicts a site near Bakersfield where salt crusts are clearly visible in the NAIP ortho-imagery over fallow land but not in the neighboring corn field. There is excellent correspondence between the high salinity sections of the site as estimated by the remote-sensing map and the location of the salt crusts . To properly compare the NAIP salt pixel classification with figure 1, we aggregated the NAIP classification at the 32.8 × 32.8 yard resolution. Only the 32.8 × 32.8 yard cells that included more than 50% of NAIP salt crusts at the original 1.09 × 1.09 yard resolution were retained for further analysis. A total of 162,829 “salt-crusted” cells were identified. About 94.3% of the salt-crusted pixels were predicted by equation 2 to be ECe > 4 dS/m. In total, the salt crusted pixels had average ECe of 13.6 dS/m, first quartile of 9.7 dS/m, median of 13.5 dS/m and third quartile of 18.2 dS/m, indicating good correspondence between visibly saline soils and predictions of high salinity by equation 2.

Scudiero et al. indicated that remote-sensing estimations at low salinity levels might be imprecise because plants may not be sufficiently osmotically stressed at low salinity to affect crop health. The spatial variability of other soil properties that influence crop yield within a single field could lead to salinity estimation errors at low salinity. Although sub-field variations in soil texture are typically minor in WSJV, some fields exhibit significant variability over short distances. In these cases, soil heterogeneity influences crop performance, introducing uncertainty into the remote-sensing estimations of soil salinity. As an example, consider the remote-sensing salinity predictions for a slightly to moderately saline where NIRWV2 , REDWV2 and BLUEWV2 are the WorldView-2 bands employed in the calculation. The EVI was selected to show that vegetation indices other than CRSI can be used to assess soil salinity, provided they reflect plant status at the target location. The multi-temporal maximum EVI map from the three WorldView-2 images is visually similar to the ground-truth salinity map . The two maps are negatively correlated, with a coefficient of determination of 0.45 . Both maps were resampled to coarser resolutions to study the changes in the strength of their relationship. As shown in figure 5D, the strength of the salinity-EVI relationship increases as block support decreases. In particular, the scaled explained variance and the strength of spatial correlation increase to a maximum at block support of 20 meters , then steadily decrease as the resolution becomes coarser. The strength of the salinity relationship with EVI at the Landsat block support was similar to that at 20 meters, indicating that it could properly represent the salinity spatial patterns at this site, despite being slightly coarser than ideal. Since the early 1950s, irrigation has played an important role in improving the quality of WSJV soils. As an example, the long-term change in soil salinity for western Fresno County is discussed by Schoups et al. . Schoups and colleagues found that long-term irrigation helped reduce soil salinity across western Fresno County throughout the second half of the 20th century. When irrigation stops, there is a risk that these trends will reverse and that salinity will rapidly increase in lands with shallow groundwater, as observed in the long-term study of Corwin . Reduced water allocations have caused farmers to use potentially higher salinity groundwater in place of lower salinity surface water and to fallow fields during the ongoing drought. According to the CropScape database, during the drought, fallow land in WSJV increased from an average of 11.8% during the years 2007 to 2010 to 19.2%, 21.0%, 21.6%, 25.9% and 33.7% through the years 2011 to 2015. Land fallowing could lead to increases in root zone salinity, thereby potentially negatively affecting future crop growth in the WSJV .

When reducing water allocations to farmland, the risks of quick land salinization should be considered. Updated regional-scale inventories of salinity will provide information for better water management decisions to support statewide agriculture and preserve soil productivity, especially in years of drought, when water resources are limited. With water shortages and droughts likely to become longer and more frequent in the future , threats from increasing soil salinity are also likely to become more severe and should, therefore, be given serious consideration by landowners,low round pots water district managers, and federal, state and local agencies. Individual soil salinity maps such as presented in this paper can help landowners and water district managers select land they wish to retire or convert to other uses . But a much greater benefit would be realized if a soil salinity remote-sensing program were established in which maps were created every 5 to 10 years for salinity-affected areas of statewide importance, including the Central and Imperial Valleys. Such a remote-sensing program would allow for the first-time monitoring of soil salinity at regional and state levels, would permit new understandings of drivers and trends in agricultural soil salinity and would aid in the development and assessment of mitigation strategies and management plans. Our primary sample is spread across 12 blocks within 4 districts of the Jhark hand state in eastern India. The blocks were identified as being suitable for a drought-tolerant rice seed variety that we were testing using a randomized controlled trial. We selected a random sample of villages amongst those with 30 to 550 households. Within each village, enumerators located a village leader and asked for names of 35 people from separate households: the 25 largest rice farmers, male individuals that work on other farmer’s fields, and 3 female individuals that also work as casual agricultural laborers. Enumerators carried out a baseline survey with the farmers and workers during the period from late April to early June 2014. Our sample of laborers consists of people that are landless or have small amounts of land. This population makes up a non-trivial share of the people dependent on agriculture in rural India. In contrast to large landowners, these workers generate most of their income from supplying labor to the casual labor market.

Yet, most studies rely on data that aggregates labor market outcomes over a longer time period. This potentially misses short-term movement between occupations. To better measure labor-market outcomes in our context, we collected daily data on wages and employment. We did this by conducting phone surveys that took place during the transplanting and harvesting periods across the 2014, 2015, and 2016 cultivation seasons. Rice is the dominant crop in our sample area and is planted in late July / early August and is harvested in late November. Our phone surveys took place during these times to coincide with the peak periods for agricultural labor demand. During the first year surveyors attempted to contact the 10 laborers in each of the 200 villages. During each call respondents were asked whether they worked on another person’s farm or their own farm, the wage they received, whether the work took place in their own village, and their activity if they did not work in agriculture. This information was collected for the seven days preceding the phone call. We repeated this same process in the 2015 and 2016 seasons with a few important differences. First, we expanded the sample to include 6 female laborers per village. The additional three laborers were selected from a census that had been conducted in all villages on households with casual laborers.Second, starting with the 2015 harvesting survey, we expanded the recall window to 14 days to more easily capture the entire planting or harvesting period for each village. The phone surveys produced a high response rate: an average of 86 percent of the workers in the baseline sample were reached.These data allow us to observe daily employment outcomes for planting and harvesting across three agricultural seasons. In addition, we collected non-agricultural wages in the 2015 planting and both 2016 surveys. These observations consist mostly of casual work for a daily wage — rather than self employment. We observe the daily wage for 82 percent of the non-agricultural work days in these three surveys. This information, along with the individual-level panel on agricultural outcomes, allows us to measure the agricultural wage gap while controlling for unobserved heterogeneity across individuals.Since the people switching sectors give identification, it is useful to compare them to the individuals that work in agriculture for the entire sample period. About 20 percent of the workers from the baseline survey switched sectors. Table 1 shows the differences between these two groups. Switchers are predominantly male and generally poorer in several dimensions. For example, they are less likely to have access to electricity, more likely to be in households using the government’s rural employment guarantee , have larger households, and more likely to belong to lower castes. They are also more likely to have household members that migrate temporarily , but are not more likely to engage in permanent migration. Yet, switchers have no less land. The average laborer household cultivates 0.57 acres during the rainy season and only about 16 percent of households cultivate no land at all.Overall, the people that switch between local agricultural and non-agricultural work are neither the wealthiest or most educated. If anything, the switchers tend to come from poorer households.