Details

Autor(en) / Beteiligte

• Cortés‐Molino, Álvaro
• Valdés‐Uribe, Alejandra
• Ellsäßer, Florian
• Bulusu, Medha
• Ahongshangbam, Joyson
• Hendrayanto
• Hölscher, Dirk
• Röll, Alexander

Titel

Combining UAV thermography, point cloud analysis and machine learning for assessing small‐scale evapotranspiration patterns in a tropical rainforest

Ist Teil von

• Ecohydrology, 2024-01, Vol.17 (1), p.n/a

Ort / Verlag

Oxford: Wiley Subscription Services, Inc

Erscheinungsjahr

2024

Link zum Volltext

Quelle

Wiley Blackwell Single Titles

Beschreibungen/Notizen

• Microclimate and vegetation structure control evapotranspiration (ET) from land surfaces at stand and landscape scales. Tropical rainforests are among the most diverse and complex terrestrial ecosystems, harbouring vast plant and animal species throughout their dense multistory canopy. They contribute substantially to global precipitation through their high ET. However, there is little information about ET influences at very small spatial scales under given climatic conditions. In a tropical rainforest on Sumatra, we studied the relationship between pixel‐level ET as derived from high‐resolution (~10 cm), near‐surface thermography from an unmanned aerial vehicle (UAV) and canopy structure as derived from red–green–blue (RGB) image and three‐dimensional (3D) point cloud analyses. The 16 derived potential predictors encompassed vegetation height, height variability, vegetation density and reflectance variables. Using regression models, several of the studied variables had a significant linear relationship with ET, but the explained variance was only marginal. However, applying a random forest algorithm including forward feature selection and target oriented cross validation explained substantial parts of the pixel‐level variance in ET (R2 = 0.56–0.65), thus indicating multiple non‐linear relationships with interactions among predictor variables. Therein, green leaf index, leaf area density and vegetation height were often the most important variables for the prediction outcome, but their sequence varied among the four study plots. Overall, combining canopy structure variables derived from RGB photogrammetry explained relatively large parts of spatial ET variations. Our study thus indicates the large potential of combining UAV‐based thermography and photogrammetry techniques with machine learning approaches to better understand ET but also suggests that more work remains to be done in explaining ET patterns at very small spatial scales.