Long-term NDVI series analysis in a sylvopastoral irrigated area

Satellite vegetation index NDVI is useful to assess green biomass and monitor crop growth. A series of Landsat images with 30 m resolution are used to estimate vegetation dynamics through the analysis of NDVI data. A time series of June NDVI is elaborated for a mountainous irrigation district in northern Tunisia, from 1984 to 2017. Public irrigation devices were implemented in 1984 for developing arboriculture, mainly apple trees, using groundwater pumping. Days corresponding to cloud cover less than 10% are considered, representing a total of 25 images. The comparison with Sentinel data 2A was achieved on two dates. Assuming three categories for NDVI (sparse, moderate and dense vegetation) the percentage of concordance is acceptable, respectively 95% and 75%. We evaluate the accuracy of NDVI observations in comparison to information obtained through field interviews. Moreover, each year, NDVI spatial variability is analyzed fitting the normal and the two-parameter log-normal distributions. The former is rejected in most cases, while the Log Normal distribution is accepted in 76% of years, and continuously beginning from 1998. The location parameters of the fitted log-normal distributions represent the dominant NDVI corrected by a factor representing the spatial variability of agriculture conditions. The classification of years using the scatterplot of fitted location and scale parameters separates them in the period 1998–2017 into two groups: (C1) from 1998 to 2010 and (C2) from 2011 to 2017. Thus, 2010 appears as a changing year for vegetation productivity. Observed irrigation volumes display an increasing trend, which is reinforced beginning from 2010. Taking the growing season rainfall (from September to April) as an independent variable for predicting the location parameter, a linear regression is fitted for each group of years (C1 and C2). The adjusted linear regressions present the same slope but a jump in the intercept is detected after 2010. It indicates that for the same growing season rainfall conditions, crop productivity increased in recent years, due to the phenological growth and to the increase of irrigation inputs. Indeed, years were classified into 2 categories according to the growing season rainfall. During precipitation shortage years, observed irrigation input is on average 1444 mmha −1 per NDVI unit while it is on average 1269 mmha −1 for surplus years, representing a reduction of only 12% in case of surplus year. A significant predictive logarithmic model of median NDVI using irrigation volumes per unit area is found for years in deficit rainfall conditions, with 62% of explained variance, showing the importance of conducting this study.