On the occurrence and causes of long-term declines in MODIS NDVI within the savanna environment of central Brazil

Abstract

The trajectory analysis of the Normalized Difference Vegetation Index (NDVI), derived from the Moderate Resolution Imaging Spectroradiometer (MODIS), can reveal long-term declines potentially linked to land degradation or decreasing vegetation productivity. In this study, we investigated the occurrence and causes of MODIS NDVI declines in the Southwest Goiás Microregion, one of the oldest agricultural areas in the Brazilian savanna environment (Cerrado). Before conducting the NDVI trajectory analysis with the Trends.Earth tool to identify changing patterns from 2000 to 2020, we first examined land use dynamics from 1985 to 2020 using Landsat imagery. We then employed binary logistic regression to statistically examine various potential factors contributing to NDVI declines. In the logistic regression model, the Aggregate NDVI Trend Indicator was used as the response variable, recoded as a binary outcome: 0 for no decline in NDVI and 1 for decline in NDVI, the long-term event of interest. Fourteen categorical and five continuous predictor variables were considered, encompassing land use and land cover changes, duration of pasture and crop use, fire frequency, precipitation, soil composition, and topography. The results showed a significant overall increase in NDVI across 66% of the study area, with 28% remaining stable. However, statistically significant NDVI declines covered 3364 km², or approximately 6% of the study area, as shown by Trends.Earth analysis. Logistic regression indicated that NDVI declines were primarily driven by two factors: the conversion of savanna to pastures and the soil composition or texture. Approximately 50% of the declines occurred in pastures converted from native savanna vegetation, while 25% were observed in savannas and 14% in crops. NDVI declines were predominantly observed in pastures situated over soils with more than 500 g/kg of sand content. Given the recent expansion of crop areas over existing pastures, detected in our study with Landsat data, the number of recorded declines in NDVI or land degraded areas is likely to increase in near future, particularly if this expansion occurs on sandy soils without adoption of adequate soil and crop management practices. Our study highlights the importance of time series analysis of satellite data in assessing land conditions in the Brazilian savanna environment.