As field size increases, the length of time required to move a packet of air from one side of the field to the other will increase, decreasing the probability that wind speed and direction will remain relatively constant. Furthermore, as the moisture content increases down wind, this would decrease vapor pressure deficit, potentially reducing rates of ET downwind. Another explanation, suggested by that fact that some crops showed a positive correlation between LSTand slope, is that rather than advection of plant-transpired moisture downwind over individual fields, there is instead an accumulation of water vapor over the field. This idea will be explored further in Section 4.1.3. Second, we did not find positive correlations between GV fraction and water vapor slope as postulated in Hypothesis E. If green vegetation is transpiring and adding to the water vapor above a field, we would expect higher fractions of GV to contribute more water vapor, and thus increase the size of the gradient. We found no correlation between water vapor slope and the GV fraction, even when results were segmented by field size and GV fraction. We used 50% GV as the cutoff to demarcate sparsely vegetated fields from highly vegetated fields, as is consistent with previous studies. However, we found that the average fractional GV coverage of fields that showed good alignment between wind direction and water vapor directionality was around 45%. Therefore, future studies may want to consider a lower GV threshold or a segmentation of fields into multiple GV classes. Finally, we did not find an inverse correlation between water vapor slope and LST in support of Hypothesis G. Either no correlation was found, drainage collection pot or highest water vapor slopes were found with higher temperature crops.Water vapor patterns were as expected at the field level, in response to wind.

However, water vapor patterns were not as expected in response to the surface properties of field size, GV fraction, and ET rate as expressed by field-scale LST. We had hypothesized that field-level water vapor slopes can be used to infer crop transpiration, but did not find evidence supporting that hypothesis. Rather, our results suggested that water vapor accumulation from transpiration was more dominant than the advection signal at the field level. The rate of ET has been found to remain constant with downwind distance across a field, even if warm, dry air is being advected toward a vegetated field. If plants are transpiring at a constant rate and winds are not strong enough or stable enough in directionality to evenly advect the moisture, the concentration of water vapor above the field would increase relatively evenly throughout the field, leading to a diminished slope. Crops are also more aerodynamically rough than an empty soil field, and the resultant turbulence caused by vegetation creates eddies and atmospheric mixing that may muddle signals of field-level advection discernable above smoother landscapes. The hypothesis of water vapor accumulation is supported by results that found a positive relationship between LST and slope for some crops, a negative relationship between field size and slope, and a weak positive correlation between water vapor intercept and GV fraction in 2013 and 2015. Therefore, the results of this study lead us to new conceptual understanding that the magnitude of water vapor as assessed though the intercept of a fitted plane may be better indicator of ET than the slope. However, underlying heterogeneity of the landscape and scaling issues, as discussed below, prohibited isolated analyses of intercepts in this study area.There is error within all water vapor estimates regardless of which retrieval method is used, and the estimates vary significantly from model to model. However, Ben-Dor et al. found that, of six different water vapor retrievals, ACORN estimated water content with acceptable accuracy and, importantly for our study, it was one of only two models that accurately discriminated water vapor from liquid water in plants.

Therefore, the positive correlations found in years 2013 and 2015 between water vapor and vegetation fraction are assumed to be a product of coupling between the landscape and the atmosphere, rather than an artifact of the retrieval.Wind direction and magnitude can change significantly within a small period of time, making estimations of wind within the study scene at the time of the flight particularly difficult. Furthermore, a sparse network of meteorological stations, may not accurately capture more local variation in wind between the stations. Thus the IDW wind field we used in this study may not adequately characterize fine spatial or temporal variability in winds at the field scale.Unlike Ogunjemiyo et al. who studied water vapor over a relatively homogeneous area of transpiring poplar trees, this study evaluated water vapor as it varies across a very diverse agricultural landscape with many different crop species, green vegetation cover, and irrigation regimes. As such, Ogunjemiyo’s conceptual model illustrated an ideal relationship between water vapor and vegetation at the field-scale that may not hold in our complex study area. First, interactions between water vapor occurring over two diverse, adjacent fields may alter the vapor deficit and stomatal response of a single crop field and result in water vapor trends that do not follow Ogunjemiyo’s model. The schematic in Fig 15A illustrates one possible interaction in which a transpiring field is upwind of a non-transpiring field. While the transpiring field will act as hypothesized with the slope and direction of a fitted plane in line with the wind direction, a plane fitted to the fallow field downwind will likely show a slope that is opposite in direction to the wind. The wind carries moist air from the vegetated field onto the fallow field, leading the upwind edge of the fallow field to have higher water vapor concentrations than the edge that is downwind. In the case of the downwind area being another highly transpiring field , the moist, advected air from the upwind field may reduce the transpiration rate of the downwind field at the boundary by decreasing the vapor pressure deficit.

This may lead to an exaggerated water vapor slope over the downwind field. The accumulation of water vapor from one field can therefore lead to shifts in vegetation response that are difficult to account for. Fig 15C illustrates the scenario where a dry, fallow field is upwind of a transpiring field. If the area upwind of a vegetated field is fallow, we would expect the saturation deficit of the dry advecting air to increase the evaporation rate at the boundary unless the vapor pressure deficit is high enough to initialize stomatal closure. A higher ET rate at the upwind side of the field will lessen the expected, observable trend of advection across the field. The transpiration response will be species-dependent. Second, not all fields will interact with the atmosphere in the same ways, due to differences in aerodynamic roughness, affected by row spacing, plant height, plant size, orientation, and composition. The aerodynamic roughness of a field will influence how effectively and at what height the transpired water vapor will mix with the atmosphere. Agricultural fields may differ strongly in aerodynamic roughness, and these differences will lead to deviations from the hypothesized water vapor slope and intercept patterns as they vary with crop type. Therefore, we would not expect all fields to show the same relationships between water vapor, wind, and estimated transpiration rates. We would expect aerodynamically rougher surfaces, such as orchards,snap clamps for greenhouse to generate greater turbulence, generate mixing higher up in the atmosphere, and show greater coupling with the wind than row crops . Depending on the wind speed, orchards may show higher or lower slopes than row crops if their vapor patterns are more tied to wind patterns. In contrast, shorter and smoother row crops such as alfalfa will be less coupled to the atmosphere . Because crops such as orchards are more closely coupled to the atmosphere, they may be more appropriate to study with water vapor imagery. Therefore, isolating the effects of neighboring fields would be beneficial for field-level water vapor analyses, but this was not logistically possible in our study. The study area is a high-producing agricultural area where most fields are bordered by multiple neighbors of varying GVcover, crop type, size, physical characteristics that influence roughness, and ET rate. Further, without LiDAR data from which physical characteristics such as orientation, height and structure could be obtained, it was not possible to model field-scale differences in aerodynamic roughness in this study. This work has aimed to enhance understanding of the impact of GV fraction, field size, crop type and water use on patterns of water vapor.

Positive findings include the presence of significant vapor gradients over most fields, and regional patterns in water vapor that are consistent with advection. High water use crops also showed a disproportionally higher level of agreement between interpolated wind direction and the direction of water vapor gradients. Field size impacted water vapor slope, although slopes were higher in smaller fields than larger fields, in contrast to expectations. We suspect improved knowledge of winds at the field scale, would improve our ability to interpret water vapor gradients. For example, given that a majority of the fields showed statistically significant water vapor slopes, an alternative hypothesis may be that those gradients better represent winds at the field scale, than interpolated winds from a sparse network of stations. Finally, we found the intercept of the best-fit surface for water vapor over a field to be more significant than the slope, suggesting that water vapor is accumulating over fields, rather than advecting.Water vapor imagery shows patterns of vapor that are highly variable through space and time and that hold valuable information about land-atmosphere interactions. We suggest there is considerable potential for this imagery and explored some of this potential here.To further scientific understanding of water vapor imagery analysis, further studies are necessary to refine observation and quantification of land-surface interactions as the signal is highly complex and is affected by many factors. While water vapor imagery could potentially be used to parameterize models of land-surface interactions, additional studies in a diversity of landscapes are necessary to define the conditions and scales at which this imagery can be used. Almost 4,000 AVIRIS images have been collected since 2006 and are available for public download. With such a large repository of data collected at different time points, under varied atmospheric conditions, and over diverse surfaces, future research could tease out the conditions under which interactions can best be observed in a more comprehensive way than this study of three snapshots in time could. Further, with future remote sensing missions such as SBG, which will collect hyperspectral imagery at moderate spatial resolutions and enable column water vapor estimates globally, these data streams can be exploited for comparisons of water vapor over large agricultural areas worldwide. These large archives of water vapor observations can also act as a compliment to models that estimate water vapor and plant water use by providing validation data. In addition to increasing analysis of similarly complex scenes, future studies would benefit from additional data sources that could isolate the signal of water vapor and validate its link to the surface. Such controls include on-site continuous wind measurements, flux tower measurements of ET, and/or more spatially comprehensive wind data. On-site wind data and ET measurements at a high temporal resolution would both validate trends seen in the water vapor imagery and assist in pinpointing the appropriate temporal scale and time of day for which this analysis is best suited. A mesoscale weather model such as the Weather Research and Forecasting Model , might also provide a more accurate fine scale representation of wind fields than simple IDW of weather stations as used here. A finer network of weather stations, and or controlled experiments with meteorological equipment deployed in advance of a flight at specific fields would also be of benefit. Although more work is needed in order to refine understanding of the water vapor signal in a complex agricultural environment, the results suggest that this technique could be of use for crop water analyses in agricultural areas that experience less variation in crop type, wind, and field size than the Central Valley of California.