Machine learning estimates of eddy covariance carbon flux in a scrub in the Mexican highland

• Main Authors: A. Guevara-Escobar, E. González-Sosa, M. Cervantes-Jiménez, H. Suzán-Azpiri, M. E. Queijeiro-Bolaños, I. Carrillo-Ángeles, V. H. Cambrón-Sandoval
• Format: Article
• Language: English
• Published: Copernicus Publications 2021-01-01
• Series: Biogeosciences
• Online Access: https://bg.copernicus.org/articles/18/367/2021/bg-18-367-2021.pdf

<p>Arid and semiarid ecosystems contain relatively high species diversity and are subject to intense use, in particular extensive cattle grazing, which has favored the expansion and encroachment of perennial thorny shrubs into the grasslands, thus decreasing the value of the rangeland. However, these environments have been shown to positively impact global carbon dynamics. Machine learning and remote sensing have enhanced our knowledge about carbon dynamics, but they need to be further developed and adapted to particular analysis. We measured the net ecosystem exchange (NEE) of C with the eddy covariance (EC) method and estimated gross primary production (GPP) in a thorny scrub at Bernal in Mexico. We tested the agreement between EC estimates and remotely sensed GPP estimates from the Moderate Resolution Imaging Spectroradiometer (MODIS), and also with two alternative modeling methods: ordinary-least-squares (OLS) regression and ensembles of machine learning algorithms (EMLs). The variables used as predictors were MODIS spectral bands, vegetation indices and products, and gridded environmental variables. The Bernal site was a carbon sink even though it was overgrazed, the average NEE during 15 months of 2017 and 2018 was <span class="inline-formula">−</span>0.78 <span class="inline-formula"><math xmlns="http://www.w3.org/1998/Math/MathML" id="M2" display="inline" overflow="scroll" dspmath="mathml"><mrow class="unit"><mi mathvariant="normal">g</mi><mspace width="0.125em" linebreak="nobreak"/><mi mathvariant="normal">C</mi><mspace width="0.125em" linebreak="nobreak"/><msup><mi mathvariant="normal">m</mi><mrow><mo>-</mo><mn mathvariant="normal">2</mn></mrow></msup><mspace width="0.125em" linebreak="nobreak"/><msup><mi mathvariant="normal">d</mi><mrow><mo>-</mo><mn mathvariant="normal">1</mn></mrow></msup></mrow></math><span><svg:svg xmlns:svg="http://www.w3.org/2000/svg" width="56pt" height="15pt" class="svg-formula" dspmath="mathimg" md5hash="15ac761ab043ccc04915a8227df2339e"><svg:image xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="bg-18-367-2021-ie00001.svg" width="56pt" height="15pt" src="bg-18-367-2021-ie00001.png"/></svg:svg></span></span>, and the flux was negative or neutral during the measured months. The probability of agreement (<span class="inline-formula"><i>θ</i></span>s) represented the agreement between observed and estimated values of GPP across the range of measurement. According to the mean value of <span class="inline-formula"><i>θ</i></span>s, agreement was higher for the EML (0.6) followed by OLS (0.5) and then MODIS (0.24). This graphic metric was more informative than <span class="inline-formula"><i>r</i>²</span> (0.98, 0.67, 0.58, respectively) to evaluate the model performance. This was particularly true for MODIS because the maximum <span class="inline-formula"><i>θ</i></span>s of 4.3 was for measurements of 0.8 <span class="inline-formula"><math xmlns="http://www.w3.org/1998/Math/MathML" id="M7" display="inline" overflow="scroll" dspmath="mathml"><mrow class="unit"><mi mathvariant="normal">g</mi><mspace linebreak="nobreak" width="0.125em"/><mi mathvariant="normal">C</mi><mspace width="0.125em" linebreak="nobreak"/><msup><mi mathvariant="normal">m</mi><mrow><mo>-</mo><mn mathvariant="normal">2</mn></mrow></msup><mspace width="0.125em" linebreak="nobreak"/><msup><mi mathvariant="normal">d</mi><mrow><mo>-</mo><mn mathvariant="normal">1</mn></mrow></msup></mrow></math><span><svg:svg xmlns:svg="http://www.w3.org/2000/svg" width="56pt" height="15pt" class="svg-formula" dspmath="mathimg" md5hash="1c7ec7db4a4b0be66e19c59f4e1acc18"><svg:image xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="bg-18-367-2021-ie00002.svg" width="56pt" height="15pt" src="bg-18-367-2021-ie00002.png"/></svg:svg></span></span> and then decreased steadily below 1 <span class="inline-formula"><i>θ</i></span>s for measurements above 6.5 <span class="inline-formula"><math xmlns="http://www.w3.org/1998/Math/MathML" id="M9" display="inline" overflow="scroll" dspmath="mathml"><mrow class="unit"><mi mathvariant="normal">g</mi><mspace linebreak="nobreak" width="0.125em"/><mi mathvariant="normal">C</mi><mspace width="0.125em" linebreak="nobreak"/><msup><mi mathvariant="normal">m</mi><mrow><mo>-</mo><mn mathvariant="normal">2</mn></mrow></msup><mspace linebreak="nobreak" width="0.125em"/><msup><mi mathvariant="normal">d</mi><mrow><mo>-</mo><mn mathvariant="normal">1</mn></mrow></msup></mrow></math><span><svg:svg xmlns:svg="http://www.w3.org/2000/svg" width="56pt" height="15pt" class="svg-formula" dspmath="mathimg" md5hash="b8e1ecc86fefb11b9db1227eb7813bb1"><svg:image xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="bg-18-367-2021-ie00003.svg" width="56pt" height="15pt" src="bg-18-367-2021-ie00003.png"/></svg:svg></span></span> for this scrub vegetation. In the case of EML and OLS, the <span class="inline-formula"><i>θ</i></span>s was stable across the range of measurement. We used an EML for the Ameriflux site US-SRM, which is similar in vegetation and climate, to predict GPP at Bernal, but <span class="inline-formula"><i>θ</i></span>s was low (0.16), indicating the local specificity of this model. Although cacti were an important component of the vegetation, the nighttime flux was characterized by positive NEE, suggesting that the photosynthetic dark-cycle flux of cacti was lower than ecosystem respiration. The discrepancy between MODIS and EC GPP estimates stresses the need to understand the limitations of both methods.</p>