Potential of unmanned aerial vehicle-based remote sensing to detect water stress and estimate yield in pecan

Abstract

Three potential uses for UAV remote sensing in pecan were tested in this study over two seasons, the 2020/21 and 2021/22 seasons. The use of vegetation indices (VIs) to detect water stress, the use of remotely sensed canopy temperature to detect water stress, and the use of remote sensing to estimate yield were all tested. Vegetation indices were of absolute importance in the processing of raw images by allowing the separation of pecan canopy from background soil and vegetation The Simple Ratio Index (SRI) was found to be the best suited of all the VIs tested for this purpose, due to ease of calculation and the large range of values. Vegetation indices were found to have a weak relationship with water stress. The best relationship found between a VI and midday stem water potential (ψmidday) during the 2020/21 season was an R2 of 0.122 with normalised difference vegetation index (NDVI). Most other VIs tested had an R2 an order of magnitude smaller. During the 2021/22 season, the Green NDVI (GNDVI) had the best relationship of all the VIs with ψmidday (R2 = 0.183). GNDVI also had the best relationship (R2 = 0.248) of all the VIs tested with stomatal conductance (gs). However, these relationships were far too weak to conclude that VIs can be used to detect or quantify water stress. It is suspected that variability in the conditions during different remote data collection flights contributed to the poor relationships between VIs and water stress, due to variability in both intercepted and reflected radiation from the canopy. When data from a single flight was tested against ψmidday, to eliminate variability in conditions, the relationship did not improve (R2 < 0.1), this suggests that VIs are inherently poor at detecting water stress. The stress degree day (Tc-Ta) and the crop water stress index (CWSI) were the thermal indices tested to allow water stress detection using remotely sensed canopy temperature (Tc), while adjusting for the effects of air temperature (Ta) and vapour pressure deficit (VPD). A weak relationship was found between Tc and ψmidday (R2 = 0.186), adjusting for Ta using the stress degree day did not improve the relationship (R2 = 0.16). This proved the necessity of adjusting for VPD as well, through the CWSI. The baselines of the CWSI were calculated using the non-water-stressed baseline (NWSB), reference surfaces and the warmest and coldest pixels of the orchard canopy-only thermal image. Destructive measurement of the non-transpiring baseline was attempted, but the resulting data was never used due to extreme variability observed in leaf-scale measurements. The equation of the NWSB was (Tc-Ta) = -0.8086VPD + 0.509 for the 2020/21 season, and (Tc-Ta) = -0.7312VPD – 0.3315 for the 2021/22 season. The differences in intercept are the result of the prevailing conditions during each season, with regards to factors other than Ta and VPD, and include radiation and windspeed. The combined NWSB using data from both seasons was (Tc-Ta) = -0,7549VPD + 0,0482. Water stress data was regressed against the CWSI from each season’s own NWSB and the combined NWSB. The importance of using Tc data from a full canopy only during the hours either side of midday was also shown. Data collected during the early morning and late evening, and from a porous canopy yielded a NWSB that differed greatly in both slope and intercept from one collected using the accepted methodology. these also differed greatly from any published NWSB for pecan. All the methods of obtaining the CWSI baselines yielded a CWSI that did not correlate well with ψmidday (NWSB and 6˚C constant upper limit: R2 = 0.157, wet reference surface and 6˚C constant upper limit: R2 = 0.0026, warmest and coolest pixels: R2 = 0.07). Unexpectedly the NWSB method performed the best, while the methods that relied on the extraction of the CWSI limits from the thermal image performed exceptionally poorly. This is evidence that the fault lies not in the method, but in some aspect of the thermal data itself, or the extraction of the reference data from the thermal data. The poor performance of the NWSB method relative to examples in the literature may have been as a result of inaccurate Tc from the thermal camera used. More work will need to be done, with accurate equipment, before the CWSI can be used to quantify water stress in pecans. Yield estimation was performed by finding the relationship between yield and canopy fractional cover, change in canopy size over the season and % change in canopy size over the season. Canopy fractional cover gave a relationship strong enough to estimate yield, but only in an “on”, or heavy bearing, year of an alternate bearing cycle, and performed better in a high yielding cultivar in the area (‘Western Schley’ R2 = 0.603) than a low yielding one (‘Wichita’ R2 = 0.497). Both of these relationships were observed at the beginning of March of the season when each cultivar had an “on” year. A good relationship between yield and canopy fractional cover was also observed early in the second measurement season (November) when ‘Wichita’ trees were in an “on” year (R2 = 0.535), no data was available for the first half of the first measurement season. Both the absolute change in fractional canopy cover and the % change in canopy cover performed poorly (R2 < 0.3) relative to fractional canopy cover on its own for both cultivars during the second season, when data was available for both the start and the end of the season, enabling the calculation of change in canopy cover. Reasonable relationships between VIs and yield were primarily observed in the last two months of the season (March and April) during the 2021/22 season, while only data from these months was available for the 2020/21 season. The VIs that performed best were RDVI, MCARI and OSAVI. All three VIs correlated best (0.35 <R2< 0.5) with the yield of the cultivar that was “on” during each season. It is likely that these VIs are sensitive to differences in some aspect of canopy structure that is related to yield. Specific VIs seem to be best able to estimate yield for specific crop conditions at specific times of the year. Large datasets will be required to determine the exact relationships present and the best time to use them, before VIs can be used to estimate yield in pecans.