Mapping canopy conductance and CWSI in olive orchards using high resolution thermal remote sensing imagery

Abstract

The characterization of field properties related to water in heterogeneous open canopies is limited by the lack of spatial resolution of satellite-based imagery and by the physical constraints of point observations on the ground. We apply here models based on canopy temperature estimated from high resolution airborne imagery to calculate tree canopy conductance (G_c) and the crop water stress index (CWSI) of heterogeneous olive orchards. The G_c model requires the simulation of net radiation (R_n) and the aerodynamic resistance (r_a) as a function of windspeed and canopy structure. In both cases, the R_n and r_a models were tested against measurements and published data for olive orchards. Modeled values of G_c of trees varying in water status correlated well with G_c estimates obtained from stomatal conductance measurements in the same trees. The model used to calculate the Crop Water Stress Index (CWSI) took into account, not only the vapor pressure deficit but the R_n and the windspeed as well, parameters known to affect the temperature differences between the air and the tree canopy. The calculated CWSI for water deficit and well irrigated olive trees correlated with the water potential measured on the same trees. The methodology applied in this manuscript was used to validate the estimation of theoretical baselines needed for the CWSI calculations, comparing against traditional empirical baseline determination. High resolution thermal imagery obtained with the Airborne Hyperspectral Scanner (AHS), and from an Unmanned Aerial Vehicle (UAV) for two years was used to map G_c and CWSI of an olive orchard where different irrigation treatments were applied. The methodology developed here enables the spatial analysis of water use within heterogeneous orchards, and the field characterization of water stress, leading to potential applications in the improvement of orchard irrigation management using high resolution thermal remote sensing imagery.

Introduction

The established method for detecting water stress of crops in the field uses a pressure chamber to measure the xylem water potential of individual leaves in selected plants (Hsiao, 1990, Shackel et al., 1997). Another water stress indicator is based on measuring stomatal conductance with leaf diffusion porometers. Both methods provide point observations which are not free of measuring errors (Hsiao, 1990) and are very time consuming, limiting the number of individuals that can be monitored to accurately characterize a field. Water stress induced stomatal closure reduces transpiration rate, thus reducing evaporative cooling and increasing leaf temperature that could be tracked by thermal infrared thermometers and imagers. This approach at detecting water stress became very popular in the 1970's and 80's with the advent of hand-held thermometers (Tanner, 1963, Fuchs and Tanner, 1966, Idso et al., 1978, Idso et al., 1981, Jackson et al., 1977a, Jackson et al., 1977b, Jackson et al., 1981, Jackson, 1982) and led to the development of a normalized index to overcome the effects of other environmental parameters affecting the relation between stress and plant temperature. The index was termed the Crop Water Stress Index (CWSI) (Idso et al., 1981, Jackson et al., 1981), and consisted in relating the actual difference between canopy and air temperatures (T_c and T_a, respectively) to the difference between the T_c − T_a values of a non-water stressed baseline (NWSB), and an upper T_c − T_a limit, both being a function of the atmospheric vapor pressure deficit (VPD) (Idso et al., 1981). A CWSI ranging from 0 to 1 is thus obtained which was found to be proportional to the stress level in many crops, provided that NWSB values are known for the crop and local conditions. Many NWSB equations were published in the past for different crops (Idso, 1982, Nakayama and Bucks, 1983, Glenn et al., 1989, Wanjura et al., 1990, Sepaskhah and Kashefipour, 1994, Yazar et al., 1999, Testi et al., 2008). A major reason for the interest in the CWSI was the possibility of measuring it remotely, thus avoiding time consuming techniques used for detecting stress at the field or farm levels.

However, the use of CWSI as a stress indicator has not been widely adopted for two main reasons (Cohen et al., 2005): (i) temperature remotely sensed from readily available satellite platforms or airborne sensors lacks the necessary spatial resolution for the accurate separation of canopy temperature from the sunlit and shaded soil background; (ii) the different equations of NWSB published are site dependent, since the VPD normalization procedure used for obtaining the CWSI does not account for differences in net radiation and aerodynamic resistance which are known to affect the index (Hipps et al., 1985, Jackson et al., 1988, Jones, 1999). Avoiding the first issue is not easy, since currently available satellite thermal imagery is limited to Landsat TM and ASTER scanners, yielding 120 m and 90 m, respectively, and MODIS or AVHRR, with 1 km pixel size. The medium-low spatial resolution of such satellite thermal scanners makes mapping water stress only potentially suitable for regional scales if successfully accounting for canopy heterogeneity (Moran et al., 1994, Norman et al., 1995). Alternatively, airborne thermal imagery has been proved suitable for mapping water stress on discontinuous canopies (Sepulcre-Canto et al., 2006, Sepulcre-Canto et al., 2007), provided that the spatial resolution allows the detection of isolated tree crowns (< 2 m pixel size). However, the cost and operational complexity of airborne platforms make its extensive use in agriculture very limited (Berni et al., in press). New thermal imaging sensors onboard unmanned aerial platforms provide sub-meter spatial resolution (Herwitz et al., 2004, Sugiura et al., 2005, Berni et al., 2009) enabling the retrieval of pure canopy temperature, thus minimizing soil thermal effects. High resolution thermal imagery would make possible the retrieval of energy fluxes from pure vegetation on open canopies, such as tree orchards, where most remote sensing based methodologies do not perform well. Nevertheless, atmospheric effects and atmospheric transmittance should be considered even for low altitude platforms aimed at keeping temperature measurement errors below 1 K (Berni et al., 2009).

If high resolution canopy temperature can be accurately monitored then the extensive use of CWSI may be a practical option. New theoretical and practical approaches have been proposed to overcome the need for the empirical retrieval of the NWSB needed in the CWSI calculation. The use of theoretical equations of CWSI based on the energy balance equation (Jackson et al., 1988) is limited by the need to estimate net radiation and aerodynamic resistance, but it allows the calculation of canopy conductance (Smith, 1988, Leinonen et al., 2006, Lhomme and Monteny, 2000). Most recent works overcome this problem by using dry and wet references that account for the CWSI upper and lower limits, respectively, allowing the estimation of CWSI with a minimum of meteorological measurements (Jones et al., 2002, Cohen et al., 2005, Grant et al., 2007, Möller et al., 2007). However the use of such reference surfaces is a clear limitation for the practical and extensive use of this methodology.

This manuscript validates a methodology to map the spatial distribution of CWSI and the canopy conductance of a field from very high spatial resolution thermal imagery and in situ atmospheric variables. This approach is particularly suitable for monitoring areas of medium size (in the order of hundred of hectares) using unmanned aircrafts that could provide frequent visits and short turnaround times to detect water stress for irrigation scheduling. The methodology presented here does not require the use of reference surfaces and relies on physical models to estimate all input variables of the energy balance equations.