Automated Geospatial Models of Varying Complexities for Pine Forest Evapotranspiration Estimation with Advanced Data Mining

Sudhanshu Panda,Rhett Jackson,Devendra Amatya,Asko Noormets,Ge Sun

doi:10.3390/w10111687

Abstract

The study goal was to develop automated user-friendly remote-sensing based evapotranspiration (ET) estimation tools: (i) artificial neural network (ANN) based models, (ii) ArcGIS-based automated geospatial model, and (iii) executable software to predict pine forest daily ET flux on a pixel- or plot average-scale. Study site has had long-term eddy-flux towers for ET measurements since 2006. Cloud-free Landsat images of 2006−2014 were processed using advanced data mining to obtain Principal Component bands to correlate with ET data. The regression model’s r2 was 0.58. The backpropagation neural network (BPNN) and radial basis function network (RBFN) models provided a testing/validation average absolute error of 0.18 and 0.15 Wm−2 and average accuracy of 81% and 85%, respectively. ANN models though robust, require special ANN software and skill to operate; therefore, automated geospatial model (toolbox) was developed on ArcGIS ModelBuilder as user-friendly alternative. ET flux map developed with model tool provided consistent ET patterns for landuses. The software was developed for lay-users for ET estimation.

Highlights

Evapotranspiration (ET) processes at the leaf-to-landscape scales in multiple land uses have important controls and feedback for the local, regional and global climate and water resource systems [1,2,3]
Researchers developed three different types of satellite image-based ET models varying from a complex to a more simple and user-friendly level as a software package for users to apply based on their knowledge and the availability of data for pine forests
As forest ET is governed by soil moisture, canopy temperature, leaf area index, stomatal conductance, and vegetation vigor besides other environmental factors, the relevant spectral bands from Landsat were processed using principal component analysis algorithm

Summary

Introduction

Evapotranspiration (ET) processes at the leaf-to-landscape scales in multiple land uses have important controls and feedback for the local, regional and global climate and water resource systems [1,2,3]. ET comprises a large part of the hydrologic budget for forests and serves as an indicator of forest growth and ecosystem productivity [2,3]. By altering forest species composition, tree age distributions, and tree densities, land management directly affects ET [8] and watershed hydrology. CO2 concentrations in atmosphere decreases ET due to decreasing stomatal conductance is certain species but the increase in forest biomass is the cause of increased forest ET [9]. Accurate estimation of ET across the landscape is needed for many hydrologic analyses, the assessment of

Objectives

Methods

Results

Conclusion