Avaliação de mapas anuais de estoque de carbono orgânico do solo brasileiro (1985-2023)

Abstract

The soil is one of the main regulators of life on Earth, essential for human survival and well-being. It plays a crucial role in the biogeochemical carbon cycle and can act as a storage reservoir for this element. However, land-use changes can affect soil carbon stocks, leading to the loss of this element into the atmosphere. Therefore, understanding and monitoring soil organic carbon (SOC) stocks promotes sustainable natural resource management. This study explores digital soil mapping (DSM) approaches to monitor the spatial variability of soil resources worldwide, with a focus on evaluating data quality, especially in extensive areas such as Brazil. The objective is to assess quality metrics for a time series spanning from 1985 to 2023, improving the ability to make informed decisions when using these maps. To achieve this objective, standard cross-validation (CV) with random resampling and spatial cross-validation with resampling by spatial clusters were employed. Quality metrics such as Mean Error (ME), Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Nash-Sutcliffe Efficiency (NSE) were calculated at both national and biome levels. This study evaluated SOC stock predictions, revealing a good overall fit (NSE of 0.87 in standard CV), but with reduced performance when spatial dependence was considered (NSE of -0.01 in spatial CV with 20 clusters). Performance variation among biomes showed better results in the Pantanal and Pampa and worse in the Amazon and Cerrado. Spatial CV indicated the need for regional training due to sensitivity to geographic segmentation. The comparison between validation methods highlighted that spatial approaches provide a more realistic assessment, suggesting the average of the metrics for a balanced analysis (NSE of 0.43).