Evaluating observed versus predicted forest biomass: R-squared, index of agreement or maximal information coefficient?

Autor(es)

Valbuena, Ruben ^{[1, 2, 3]} ; Hernando, Ana ^[4] ; Manzanera, Jose Antonio ^[4] ; Gorgens, Eric B. ^[5] ; Almeida, Danilo R. A. ^[6] ; Silva, Carlos A. ^{[7, 8]} ; Garcia-Abril, Antonio ^[4]

Afiliação do(s) autor(es)

^[1] Univ Cambridge, Dept Plant Sci Forest Ecol & Conservat, Cambridge - England

^[2] Univ Eastern Finland, Fac Forest Sci, Joensuu - Finland

^[3] Bangor Univ, Sch Nat Sci, Thoday Bldg, Bangor LL57 2UW, Gwynedd - Wales

^[4] Univ Politecn Madrid, Res Grp SILVANET, Coll Forestry & Nat Environm, Ciudad Univ, Madrid - Spain

^[5] Univ Fed Vales Jequitinhonha & Mucuri, Dept Forestry, Diamantina - Brazil

^[6] Univ Sao Paulo, Luiz de Queiroz Coll Agr, Dept Forest Sci, Piracicaba - Brazil

^[7] Univ Maryland, Dept Geog Sci, College Pk, MD 20742 - USA

^[8] NASA, Biosci Lab, Goddard Space Flight Ctr, Greenbelt, MD - USA

• Tipo de documento: Artigo Científico
• Fonte: EUROPEAN JOURNAL OF REMOTE SENSING; v. 52, n. 1, p. 345-358, 2019.
• Citações Web of Science: 2

Resumo

The accurate prediction of forest above-ground biomass is nowadays key to implementing climate change mitigation policies, such as reducing emissions from deforestation and forest degradation. In this context, the coefficient of determination () is widely used as a means of evaluating the proportion of variance in the dependent variable explained by a model. However, the validity of for comparing observed versus predicted values has been challenged in the presence of bias, for instance in remote sensing predictions of forest biomass. We tested suitable alternatives, e.g. the index of agreement () and the maximal information coefficient (). Our results show that renders systematically higher values than , and may easily lead to regarding as reliable models which included an unrealistic amount of predictors. Results seemed better for , although favoured local clustering of predictions, whether or not they corresponded to the observations. Moreover, was more sensitive to the use of cross-validation than or , and more robust against overfitted models. Therefore, we discourage the use of statistical measures alternative to for evaluating model predictions versus observed values, at least in the context of assessing the reliability of modelled biomass predictions using remote sensing. For those who consider to be conceptually superior to , we suggest using its square , in order to be more analogous to and hence facilitate comparison across studies. (AU)

• Processo FAPESP: 18/21338-3 - Monitoramento da restauração de paisagens florestais usando veículo aéreo não tripulado com sensoriamento remoto Lidar e hiperespectral
• Beneficiário: Danilo Roberti Alves de Almeida
• Modalidade de apoio: Bolsas no Brasil - Pós-Doutorado