Growth and yield prediction and projection for eucalyptus plantations using medium spatial resolution multispectral imagery
Name: JEANGELIS SILVA SANTOS
Publication date: 11/02/2020
Advisor:
Name |
Role |
|---|---|
| ADRIANO RIBEIRO DE MENDONÇA | Advisor * |
Examining board:
| Name |
Role |
|---|---|
| ADRIANO RIBEIRO DE MENDONÇA | Advisor * |
| ANDRÉ QUINTÃO DA ALMEIDA | External Examiner * |
| GILSON FERNANDES DA SILVA | Internal Examiner * |
Summary: SANTOS, J. S. Growth and yield prediction and projection for eucalyptus plantations using medium spatial resolution multispectral imagery. 2020. Thesis (Doctorate in Forest Sciences) Federal University of Espírito Santo, Jerônimo Monteiro, ES. Advisor: Prof. Dr. Adriano Ribeiro de Mendonça. Co-Advisors: Dr. Fábio Guimarães Gonçalves and Prof. Dr. Samuel de Pádua Chaves e Carvalho.
The efficient management and planning of forest areas depends directly on the acquisition of accurate information about the stands. Information about the development of forests can be previously obtained by growth and yield models. However, the adjustment of these models requires data from continuous forest inventories, which are complex and costly activities. One of the alternatives that can reduce the costs of the forest inventory is the use of remote sensing tools. Therefore, the objective of this work was to propose a methodology for using medium spatial resolution multiespectral data for the prediction and projection of growth and yield and to determine the technical age of harvesting of eucalyptus forests, aiming at reducing the number of plots measured in the forest inventory. For this purpose, two databases were used: one containing information on age and volume per hectare of 40 permanent plots measured between 2006 and 2011, with ages varying from two to seven years, and other containing time series of Tasseled Cap (TC) metrics extracted from ETM+/Landsat 7 imagery, smoothed by the Savitzky-Golay filter. To assess the possibility of reducing the number of plots measured in the continuous forest inventory when using remote sensing data, three scenarios were proposed, with different sampling intensities: 1) one plot every 28 ha; 2) one plot every 42 ha, and; 3) one plot every 83 ha. The estimation was performed by artificial neural networks and, in the prediction, the input variables were the age of the stand and the metrics of the Tasseled Cap transformation (brightness, greenness and wetness). For the projection, the variables were the current and future age and the current volume, obtained by the prediction for the first year of the continuous forest inventory. The prediction and projection were applied wall-to-wall, and the projection maps were used to calculate the mean and current annual increment and to determine the technical age of harvest. In the wall-to-wall prediction, the RMSE values ranged from 7.92% in scenario 1 to 10.67% in scenario 3. As for the projection, the RMSE varied from 9.68% in scenario 2 to 11.75% in scenario 3. In general, there was no major discrepancy between the accuracy measures in the three scenarios. In addition, all the scenarios analyzed for prediction and projection presented estimated values within the confidence interval of the forest inventory. The mean and current monthly increment values projected by the different scenarios analyzed did not differ from that obtained by the continuous forest inventory, with the growth curve inflection and forest maturity points being very close. Therefore, it can be concluded that the use of remotely sensed data allowed to accurately estimate the prediction and projection of growth and production of eucalyptus forests. In addition, by applying the methodology presented here, it is possible to significantly reduce the sampling intensity by up to one plot every 83 ha, with accuracy compatible with the methodology traditionally used in the continuous forest inventory.
Keywords: Forest Management, Tasseled Cap, Artificial Neural Networks, Google Earth Engine, Enhanced Forest Inventory
