TL;DR
This paper uses satellite-derived vegetation data (NDVI) and rainfall records from Senegal to assess land performance and identify areas of degradation. By analyzing the correlation between rainfall and vegetation growth across different land cover types, the study reveals how human activities and environmental factors impact land productivity, offering a valuable tool for regional land management.
What Problem Does This Paper Address?
Land degradation, characterized by a decline in land productivity, is a critical environmental issue, particularly in arid and semi-arid regions like the Sahel. Understanding the causes and extent of this degradation is crucial for effective land management and sustainable development. However, monitoring land performance over large areas and identifying the drivers of change (both natural and human-induced) presents a significant challenge. This paper aims to address this by developing and applying a methodology to evaluate land performance in Senegal using readily available satellite and meteorological data, specifically focusing on the relationship between rainfall and vegetation health.
Methods
The study employs a multi-temporal analysis of Normalized Difference Vegetation Index (NDVI) and annual rainfall data. NDVI, derived from satellite imagery, serves as a proxy for vegetation vigor and productivity. The core of the methodology involves:
- Correlation Analysis: Calculating the correlation coefficient between annual rainfall and season-integrated NDVI (iNDVI) for different land cover types (steppe, agriculture, savanna, woodland/forest). This establishes a baseline relationship, assuming that in healthy ecosystems, vegetation growth is strongly linked to rainfall.
- Land Cover Performance Assessment: Deviations from this expected correlation are interpreted as indicators of land performance. High positive correlations suggest "normal" performance where vegetation responds predictably to rainfall. Low or negative correlations can indicate areas where land resources are naturally unproductive or, more importantly, where human activities have impacted vegetation growth.
- Pixel-by-Pixel Analysis: The correlation analysis is performed at a pixel level (8 km resolution) to identify spatial patterns of land performance across Senegal.
- Standard Error of the Estimate (SEE) Analysis: To further strengthen the detection of anomalies, a "local" SEE is calculated within moving windows. Pixels with values significantly deviating from the regression line (outliers) are identified as potential indicators of anomalous land cover performance, particularly those showing low NDVI despite high rainfall.
- Time Series Analysis: Overlapping 12-year intervals are used to examine trends in land cover performance and identify changes between different periods.
Datasets
- NDVI Data: Half-monthly NDVI data from the Advanced Very High Resolution Radiometer (AVHRR) sensor, processed by the Global Inventory Monitoring and Modeling Studies (GIMMS) group. This data has a spatial resolution of 8 km and covers the period 1982-1997.
- Rainfall Data: Dekadal (10-day) rainfall data from stations across Senegal, obtained from the Centre de Suivi Ecologique, Dakar. This data was processed to generate annual rainfall maps for each year from 1982-1997, using kriging with trend interpolation.
Key Insights and Ideas
- Rainfall-NDVI Link is Strong but Not Universal: The study confirms a strong regional correlation between rainfall and iNDVI across Senegal, particularly in the northern two-thirds. However, this relationship weakens in areas with high rainfall (over 700-800 mm/yr) due to NDVI saturation and in areas with significant human intervention like irrigation.
- Correlation as a Diagnostic Tool: The rainfall-NDVI correlation coefficient proves to be a powerful indicator of land cover performance. Deviations from expected correlations can highlight areas where human activities (e.g., overgrazing, intensive cultivation) are leading to degradation, or where management practices (e.g., irrigation) are enhancing productivity.
- Identifying Degradation Hotspots: The analysis successfully pinpointed specific areas exhibiting signs of land degradation, such as Dendoudi and Touba, which are known to have experienced long-term degradation due to factors like livestock concentration and urban expansion.
- SEE for Anomaly Detection: The SEE analysis provides a complementary method to correlation analysis, effectively highlighting spatially anomalous pixels that might indicate land degradation, even in areas where the correlation itself might be less clear.
Key Results
- Regional Correlation: The study found a strong relationship between annual rainfall and season-integrated NDVI for Senegal, with correlation coefficients ranging from 0.74 to 0.90 (mean of 0.84).
- Land Cover Specific Correlations: Correlation coefficients varied by land cover type: Steppes (0.89), Agriculture (0.71), Savanna (0.73), and Woodland/Forest (0.30). The low correlation for woodland/forest suggests a less direct dependency on rainfall, possibly due to other factors or saturation effects.
- Identified Degradation Areas: The analysis identified three areas of poor performance: Dendoudi, Touba, and Revane, which showed significant negative anomalies in the SEE analysis and were confirmed by ground evidence and independent satellite data.
- Temporal Trends: Positive NDVI trends were observed in the southeast and east-central regions, potentially linked to increased rainfall. Negative trends were localized, particularly in the west-central agricultural region, likely due to soil depletion from continuous cultivation.
Limitations and Future Work
- Spatial Resolution: The 8-km pixel resolution of the AVHRR data is a significant limitation, as many local degradation areas may fall below this threshold. This heterogeneity within a pixel can lead to misclassification or mask fine-scale degradation.
- Woodland/Forest Assessment: The NDVI-rainfall correlation approach is less appropriate for woodland/forest areas due to saturation effects and increased cloud cover/water vapor in southern Senegal.
- Small Sample Size for Temporal Analysis: The use of 12-year intervals for temporal analysis, while maximizing sample size, introduced some uncertainty due to statistical variability.
- Future Work: The authors recommend using higher resolution NDVI data (1-km) for more detailed analysis. They also suggest implementing comprehensive long-term monitoring programs for biomass and land cover change across the Sahel region and examining other test sites to confirm the applicability of their approach. Further investigation is needed to understand the anomalous departures observed in the 6-10 year category of negative SEE outliers, particularly in northern Senegal.
Summary updated