Water, Energy, and Carbon with Artificial Neural Networks (WECANN): A statistically-based estimate of global surface turbulent fluxes and gross primary productivity using solar-induced fluorescence

Abstract

A new global estimate of surface turbulent fluxes, latent heat flux (LE) and sensible heat flux (H), and gross primary production (GPP) is developed using a machine learning approach informed by novel remotely sensed Solar-Induced Fluorescence (SIF) and other radiative and meteorological variables. This is the first study to jointly retrieve LE, H and GPP using SIF observations. The approach uses an artificial neural network (ANN) with a target dataset generated from three independent data sources, weighted based on triple collocation (TC) algorithm. The new retrieval, named Water, Energy, and Carbon with Artificial Neural Networks (WECANN), provides estimates of LE, H and GPP from 2007 to 2015 at 1° × 1° spatial resolution and on monthly time resolution. The quality of ANN training is assessed using the target data, and the WECANN retrievals are evaluated using eddy covariance tower estimates from FLUXNET network across various climates and conditions. When compared to eddy covariance estimates, WECANN typically outperforms other products, particularly for sensible and latent heat fluxes. Analysing WECANN retrievals across three extreme drought and heatwave events demonstrates the capability of the retrievals in capturing the extent of these events. Uncertainty estimates of the retrievals are analysed and the inter-annual variability in average global and regional fluxes show the impact of distinct climatic events - such as the 2015 El Niño - on surface turbulent fluxes and GPP.

Figure 1

Architecture of the ANN layers. Input layer provides the matrix P of the inputs to the Hidden layer. Hidden layer has a matrix W of weights and b of biases for the neurons, and the f1 transfer function. The output of the Hidden layer (a = f₁(WP +b)) is an input to the Output layer that applies the transfer function f₂ to the estimates and generates final outputs O.

Figure 2

Left column: Annual average retrievals in 2011 for (a) LE, (b) H, and (c) GPP. Right column: Density scatterplot between estimates of ANN and target data for (d) LE, (e) H, and (f) GPP during the validation period (2011). The density of scatter points is represented by the shading color. The diagonal black line depicts the 1:1 relationship.

Figure 3

Global patterns of seasonal average LE from WECANN in 2011, (a) December – February, (b) March – May, (c) June – August, and (d) September – November.

Figure 4

Similar to Figure 3 but for H instead of LE

Figure 5

Similar to Figure 3 but for GPP instead of LE

Figure 6

Correlation coefficient (R²) between WECANN retrievals and FLUXNET tower estimates categorized across different plant functional types for (a) LE, (b) H, and (c) GPP. Markers show mean, and whiskers show one standard deviation intervals. (CRO=Croplands, DBF=Deciduous Broadleaf Forests, EBF=Evergreen Broadleaf Forests, ENF=Evergreen Needleleaf Forests, GRA=Grasslands, MF=Mixed Forests, SAV=Savannas, and WET=Permanent Wetlands)

Figure 7

Comparison of the retrievals with eddy covariance observations of LE, H and GPP across 5 sites (a) US-ARM site, USA, (b) AT-Neu site, Austria, (c) BE-Bra site, Belgium, (d) FI-Hyy site, Finland, (e) US-SRG, USA, and (f) ZA-Kru, South Africa

Figure 8

Difference between annual mean LE retrieved by WECANN and the three target datasets (a–c). Scatter plots of LE retrieved from WECANN vs. from each of the target datasets (d–f). Data used are from 2011.

Figure 9

Similar to Figure 8 but for H instead of LE

Figure 10

Similar to Figure 8 but for GPP instead of LE

Figure 11

Mean monthly anomalies (in percentage with respect to mean value) for three extreme heatwave events.

Figure 12

Relative absolute difference between ET estimates of WECANN compared to modeled ET from basin scale water budget closure. Markers show mean, and whiskers show one standard deviation intervals.

Figure 13

Annual mean estimates and uncertainty bounds of LE (top row), H (middle row) and GPP (bottom row) retrievals at global (left column) and regional (four right columns) scales between 2007 and 2015. The central line in each box indicates the mean, the edges of the box are 25th and 75th percentiles, and the whiskers show the most extreme values.