Estimating grapevine crop coefficients at high-resolution using open-source satellite data

Abstract

Climate change results in increasing water stress due to co-effects of rising evapotranspiration (ET) and decreased precipitation over the past 65 years (Spinoni et al. 2019). Though mild water deficits can improve fruit and wine quality, severe shortage of water to grapevines negatively influences both grapevine productivity and fruit quality (van Leeuwen et al. 2024). To address these adverse drought effects on grapevines, one direct and effective solution is the application of supplemental irrigation with appropriate schedules (Schlank et al. 2024). For winegrapes, ET-based irrigation scheduling has shown to result in higher yields, bunch numbers, and crop water use efficiency, compared to other methodologies like grower’s experience and soil-based measures (Schlank et al. 2024).

The ET method measures evaporation from soil and transpiration from vegetation. Crop evapotranspiration (ETc) can be computed according to derived formula (Allen et al. 1998):

𝐸𝑇𝑐 = 𝐸𝑇0 × 𝑘𝑐

ET0 is the ‘Reference ET’, which represents the rate of evapotranspiration from a reference surface, often a 10 cm tall grass, and can be calculated via the Penman-Monteith model (Allen et al. 1998), with data commonly obtained from automatic weather stations. A key parameter in the ETc computation is the crop coefficient (kc), a dimensionless parameter related to the canopy size and leaf area, and is highly variable during the growing season due to varying canopy structure, fraction of ground covered by vegetation, training system, and pruning level, amongst others (Pereira et al. 2020).

Compared to other time-consuming, costly, cumbersome and often single-point or small-scale kc measurement methods like lysimeter or flux tower, remote sensing (RS) has high temporal flexibility and broader spatial representativeness by capturing structural, spectral, thermal, or microwave information from land covers (Gautam and Pagay 2020, Gautam et al. 2021). Amongst the mainstream RS platforms including Manned Aircraft System (MAS), Unmanned Aerial Vehicle (UAV), satellites provide low-cost, long-term time-series data with broad spatial coverage, which benefits regional predictive analysis (Gautam and Pagay 2020, Govi et al. 2024). Airborne platforms, by contrast, are more challenging to operate and collect data at such a large scale. Additionally, the similarity between kc and satellite-derived vegetation indices has driven the use of low-cost remote sensing technologies to estimate kc across various spatial and temporal scales (Gautam et al. 2021).

The current limitation of using satellite data is the low resolution of the dataset; typical pixel sizes or ground sampling distances (GSD) are greater than the area occupied by a single grapevine canopy.  Furthermore, vineyard inter-rows are often covered by non-vine vegetation such as cover crops or weeds (Gautam and Pagay 2020), which result in a ‘mixed’ satellite pixel that contain a combination of grapevines, bare soil, weeds, and/or cover crops, all of which add “noise” to the vine spectral signal (Govi et al. 2024). Separation of land cover types to obtain specific VI values can help to improve the accuracy of RS kc predictions (Quintano et al. 2012). To the best of our knowledge, there are no reports on unmixing data of low- to medium-spatial-resolution satellite data for grapevine kc computation. This study aims to fill this gap.