Modeling network of research sites for monitoring carbon flows by Random Forest method

DOI: 10.35595/2414-9179-2022-1-28-645-658

View or download the article (Rus)

About the Authors

Valentina A. Dobryakova

Tyumen State University,
Volodarskiy str., 6, 625003, Tyumen, Russia;
E-mail: v.a.dobryakova@utmn.ru

Natalya N. Moskvina

Tyumen State University,
Volodarskiy str., 6, 625003, Tyumen, Russia;
E-mail: n.n.moskvina@utmn.ru

Andrey B. Dobryakov

The Ural head department of the Central bank of the Russian Federation,
Volodarskiy str., 48, 625000, Tyumen, Russia;
E-mail: dobryakov_andrey@mail.ru

Lilia F. Zhegalina

Immanuel Kant Baltic Federal University,
Proletarskaya Str. 131, 236029, Kaliningrad, Russia;
E-mail: lzhegalina@kantiana.ru

Abstract

Environmental observing networks provide information for understanding and predicting the spatial and temporal dynamics of Earth biophysical processes. The optimization of resources for large-scale environmental monitoring activities is required. The paper describes and then tests spatial structure of Tyumen region research sites network. The network is based on principles of landscape approach, taking into account cost minimization. At the baseline of research, two testing sets of 40 and 105 points were determined. Proposed locations were evaluated using Random Forest (RF) method. The study accomplished in two stages for each test set. At the first stage, the model was trained; its capacity and indicators of additional diagnostics were studied. At the second stage, the trained model was used to predict the points formed of regular grid covering entire territory of this region (544 points). In conclusion, the obtained results were compared with similar point sets of the same volume but generated randomly. Primary Productivity Gross (GPP) was chosen as predictable variable because it is one of the major complex environmental indicators associated with carbon production in this area. The ability of an area to absorb or produce carbon is one of the main parameters that determine climate processes. As independent variables characterizing geosystemic processes, a set of indicators associated with climate, terrain parameters, and variability of soil resources has been selected. The problem was solved using Forest-Based Classification and Regression tool from Spatial Statistics—Modeling Spatial Relationships toolkit of ArcGIS Pro software package. As the result of the study, a high forecast accuracy and reliability for both approaches to research sites locations was obtained. The study was based on open source data.

Keywords

carbon flux monitoring, random forest, gross primary productivity (GPP), landscape approach

References

  1. Atlas of Tyumen region. Issue 1. Moscow–Tyumen: General Directorate of Geodesy and Cartography, 1971. (in Russian)
  2. Aubinet M., Vesala T., Papale D. Eddy Covariance: A Practical Guide to Measurement and Data Analysis. New York: Springer, 2012. 450 p. DOI: 10.1007/978-94-007-2351-1.
  3. Baccini A., Goetz S.J., Walker W.S., Laporte N.T., Sun M. et al. Estimated carbon dioxide emissions from tropical deforestation improved by carbon-density maps. Nature Climate Change, 2012. V. 2 (3). P. 182–185. DOI: 10.1038/nclimate1354.
  4. Breiman L. Random Forests. Machine Learning, 2001. V. 45. P. 5–32. DOI: 10.1023/A:1010933404324.
  5. Clewley D., Whitcomb J., Akbar R., Silva A.R., Berg A., Adams J.R. et al. A method for upscaling in situ soil moisture measurements to satellite footprint scale using random forests. IEE Journal of Selected Topics in Applied Earth Observation and Removed Sensing, 2017. V. 10. No. 6. P. 2663–2673. DOI: 10.1109/JSTARS.2017.2690220.
  6. Cutler D.R., Edwards T.C., Beard K.H., Cutler A., Hess K.T. Random forests for classification in ecology. Ecology, 2007. V. 88. No. 11. P. 2783–2792. DOI: 10.1890/07-0539.1.
  7. Fick S.E., Hijmans R.J. WorldClim 2: new 1-km spatial resolution climate surfaces for global land areas. International Journal of Climatology, 2017. V. 37. No. 12. P. 4302–4315. DOI: 10.1002/joc.5086.
  8. Friedl M., Sulla-Menashe D. MCD12C1 MODIS/Terra+Aqua Land Cover Type Yearly L3 Global 0.05Deg CMG V0 06. 2015. Web resource: https://lpdaac.usgs.gov/products/mcd12c1v006/ DOI: 10.5067/MODIS/MCD12C1.006.
  9. Gomes L.C., Faria R.M., de Souza E., Veloso G.V., Schaefer E.G.R., Filho E.I.F. Modelling and mapping soil organic carbon stocks in Brazil. Geoderma, 2019. V. 340. P. 337–350. DOI: 10.1016/j.geoderma.2019.01.007.
  10. Han H., Wan R., Li B. Estimating Forest Aboveground Biomass Using Gaofen-1 Images, Sentinel-1 Images and Machine Learning Algorithms: A Case Study of the Dabie Mountain Region. Remote Sensing, 2022. V. 14. No. 1. P. 176. DOI: 10.3390/rs14010176.
  11. Joiner J., Yoshida Y., Vasilkov A.P., Schaefer K., Jung M., Guanter L., Zhang Y., Garrity S., Middleton E.M., Huemmrich K.F., Gu L., Belelli Marchesini L. The seasonal cycle of satellite chlorophyll fluorescence observations and its relationship to vegetation phenology and ecosystem atmosphere carbon exchange. Remote Sensing of Environment, 2014. V. 152. P. 375–391. DOI: 10.1016/j.rse.2014.06.022.
  12. Karra K., Kontgis C. Statman-Weil Z., Mazzariello J.C., Mathis M., Brumby S.P. Global land use/land cover with Sentinel-2 and deep learning. 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS, 2021. P. 4704–4707 DOI: 10.1109/IGARSS47720.2021.9553499.
  13. Kirschbaum M.U.F., Eamus D., Gifford R.M., Roxburgh S.H., Sands P.J. Definitions of Some Ecological Terms Commonly Used in Carbon Accounting. Net Ecosystem Exchange, 2001. P. 2–5.
  14. Kozin V.V. Nature management in the North-West of Siberia: experience in solving problems. Tyumen: Tyumen State University, 1996. 167 p. (in Russian)
  15. Mascaro J, Asner G.P., Knapp D.E., Kennedy-Bowdoin T., Martin R.E., Anderson C. et al. A Tale of Two “Forests”: Random Forest Machine Learning Aids Tropical Forest Carbon Mapping. PLoS ONE, 2014. V. 9 (1). P. 1–9. DOI: 10.1371/journal.pone.0085993.
  16. Pastorello G., Trotta C., Canfora E. et al. The FLUXNET2015 dataset and the ONEFlux processing pipeline for eddy covariance data. Scientific Data, 2020. V. 7 (225). DOI: 10.1038/s41597-020-0534-3.
  17. Rodriguez-Galiano V.F., Abarca-Hernandez F., Ghimire B., Chica-Olmo M., Atkinson P.M., Jeganathan C. Incorporating Spatial Variability Measures in Land-cover Classification using Random Forest. Procedia Environmental Sciences, 2011. V. 3. P. 44–49. DOI: 10.1016/j.proenv.2011.02.009.
  18. Turner D.P., Ritts W.D., Cohen W.B., Gower S.T., Zhao M., Running S.W. et al. Scaling Gross Primary Production (GPP) over boreal and deciduous forest landscapes in support of MODIS GPP product validation. Remote Sensing of Environment, 2003. V. 88. P. 256–270. DOI: 10.1016/j.rse.2003.06.005.
  19. Villarreal S., Guevara M., Alcaraz-Segura D., Vargas R. Optimizing an environmental observatory network design using publicly available data. Journal of Geophysical Research: Biogeosciences, 2019. V. 124. No. 7. P. 1812–1826. DOI: 10.1029/2018jg004714.
  20. Villarreal S., Vargas R. Representativeness of FLUXNET sites across Latin America. Journal of Geophysical Research: Biogeosciences, 2021. V. 126. No. 3. DOI: 10.1029/2020JG006090.
  21. Wang X., Liu C., Lv G., Xu J., Cui G. Integrating Multi-Source Remote Sensing to Assess Forest Aboveground Biomass in the Khingan Mountains of North-Eastern China Using Machine-Learning Algorithms. Remote Sensing, 2022. V. 14. No. 4. DOI: 10.3390/rs14041039.

For citation: Dobryakova V.A., Moskvina N.N., Dobryakov A.B., Zhegalina L.F. Modeling network of research sites for monitoring carbon flows by Random Forest method. InterCarto. InterGIS. GI support of sustainable development of territories: Proceedings of the International conference. Moscow: MSU, Faculty of Geography, 2022. V. 28. Part 1. P. 645–658. DOI: 10.35595/2414-9179-2022-1-28-645-658 (in Russian)