Estimación de evapotranspiración espacialmente distribuida mediante un enfoque GEOBIA (Estimation of spatially distributed evapotranspiration through of GEOBIA approach).

Abstract

Estimation of the crop water requirement is critical in the optimization of agricultural production process, due yield and costs are directly a ected by this estimation. Important bene ts of a correct estimation of water requirement are the increasing irrigated area, and a high production due a better root condition. Actually in most cases the crop water requirement, represented by the Evapotranspiration (ET), is estimated through local approaches. Some of these approaches use classical and novel techniques, that can estimate ET with high accurate but precluding the spatial crop representation of ET. In this sense, remote sensing appears as a way to ET estimation considering spatial and temporal variability. ET models using satellite images have been developed in the last decades, using in most cases the surface energy balance which has generated good ET representation in di erent study sites. One of these models is METRIC (Mapping Evapotranspiration at high Resolution using Internalized Calibration) that using a simple and physical-empirical basis solving the surface energy balance to estimate ET. The main drawback of METRIC model is the low robustness in the selection of a pair of parameters called anchor pixels, that in the original version of METRIC were selected by an operator, but aiming to standardize this selection and avoid the e ect of di erent operator criteria an automation was proposed. Although this automation standardizes the model response, this requires a selection of an area where to nd these anchor pixels, moreover a high sensibility to di erent anchor pixel candidates still can be observed. In addition, where this automation is implemented selecting di erent areas where anchor pixels are found, important di erences in the ET estimation are generated. In this dissertation an object based image analysis (GEOBIA) is implemented to anchor objects (changing the approach from pix With the GEOBIA approach, segmentation and classi cation processes are used to a correct selection of segments in the image that ful ll the requirements for anchor objects, considering spectral and contextual information. The implementation of this approach aiming to increase the model robustness regarding to anchor objects selection. Image segmentation was done using an adaptation of SLIC algorithm proposed to carry out a hierarchical segmentation, in order to identify the optimal scale segmentation for each parent segment. While, classi cation was divided in two processes: a primary classi cation using Random Forest that was trained considering seasonal and humidity conditions; and a post-classi cation step considering contextual and statistical information. Results showed that the proposed GEOBIA approach allows to improve the robustness in the selection of anchor objects, which was validated through comparison with original selection of anchor pixels in two scenes, moreover, entire set of scenes where the methodology was implemented showed a decrease in the variability of the result. Homogeneity analysis showed that ET objects with classes where crops can be found have low intra-object variability and high inter-object separability. These results are a good indicator of that segmentation process allowed to generate homogeneous segments without loss of spectral characteristics and correctly separated from one another. Validation of the proposed methodology comparing estimated with measured data showed that the proposed methodology generates a slight error greater than original model. This result suggest that some improvements to nal selection of anchor objects can be carried out, in this sense, could be considered di erent metrics to describe the segments, additional mapps, and to ex´plore other relations between segments considering spatial, temporal and hierarchical characteristics.