Remote sensing quantifies evapotranspiration in sugarcane in more detail

The modernization of computational resources and the application of artificial intelligence algorithms have provided advances in studies of crop evapotranspiration using remote sensing. Researchers from the Federal University of Viçosa, Embrapa and the University of Nebraska – Lincoln (USA), used data from the MSI multispectral sensor, on board the Sentinel-2 satellite constellation, associated with machine learning algorithms, in an attempt to estimate crop evapotranspiration of sugar cane irrigated by central pivot in the northern region of the state of Minas Gerais. The developed algorithms were able to estimate crop evapotranspiration in a simpler way than the METRIC reference method – High Resolution Evapotranspiration Mapping with Internalized Calibration. 

According to Robson Argolo, from the Federal University of Viçosa, the results of this research represent an advance in the use of remote sensing for irrigation management, which is still a challenge due to the complexity and temporal resolution when using a single platform. “In this way, explains Argolo, the use of artificial intelligence with data from various remote sensing platforms makes it possible to quantify evapotranspiration in sugarcane on a higher temporal scale, thus acquiring greater information about evapotranspiration in this crop”. 

Furthermore, he mentions that this study is part of two more in development. The other two represent further advances, one being the use of remote sensing with machine learning to quantify evapotranspiration in sugarcane without the need for a meteorological station in the field, while the other has the same purpose, but uses remote sensing data for radar that has the ability to cross clouds and interact with targets on the Earth's surface, which is not possible with optical remote sensing - Sentinel, Landsat, MODIS, among others.

Vinicius Bufon, researcher at Embrapa Meio Ambiente, explains that the objective was to develop more simplified models for estimating evapotranspiration through remote sensing, but with acceptable prediction capacity and reliability. Such algorithms, based on artificial intelligence, have a robust structure that allows the identification of relationship patterns between the variables to be modeled and the so-called (independent) predictor variables. Machine learning is an interdisciplinary field based on computer science, statistics, mathematics and optimization, among several other areas. 

Evapotranspiration is the main measure of water use by agricultural crops and native vegetation. In addition to being a reference for plant health, vigor and productive potential, it allows the water balance to be carried out – an accounting of water inflows and outflows from the environment where the plants are located. Through the water balance, the level of water satisfaction of plants is understood, including in the context of climate change.

In addition to the MSI sensor, embarked on the Sentinel-2 satellite mission, through specific calibrations, the methodology can also be applied to other sensors, embarked on other satellites, as a way to expand the source of information and increase the spatial and temporal resolution of evapotranspiration estimates. . Despite being extensive, the methodology is easy to apply and satellite images can be accessed free of charge. 

Alternative methods of understanding evapotranspiration, according to the researchers, present challenges. Direct measurement of evapotranspiration using a lysimeter is quite reliable, but expensive and does not allow area coverage on the scale that satellite images do. Energy and mass flow methods require high-cost meteorological instrumentation and present the same challenge as covering and representing large areas – the problem of spatial resolution.

In itself, the analysis of multispectral images from satellite sensors such as Sentinel, Landsat, among others, already allows estimation by applying models such as METRIC, SEBAL, among others. However, the combination of remote sensing with the application of artificial intelligence and machine learning models offers the possibility of improving evapotranspiration estimates in an even more simplified way, being able to integrate multiple platforms, that is, not just using Sentinel or Landsat, but both with the addition of other satellites to make up the multiplatform.

The complete work by Robson Argolo, Everardo Mantovani, Elpídio Inácio Fernandes-Filho, Roberto Filgueiras, Rodrigo Lourenço, Universidade Federal de Viçosa (UFV), Vinícius Bufon, Embrapa Meio Ambiente and Christopher Neale, University of Nebraska, USA, is available here.