Optimization of kNDVI for improved biosphere monitoring with the AVHRR sensor

Vegetation indices (VIs) calculated from spectral values have been widely used for monitoring the terrestrial biosphere. These indices serve as proxies for determining certain biophysical parameters, useful for monitoring vegetation. Due to its ease of application, the Normalized Difference Vegetation Index (NDVI) has been used as a proxy of various biophysical parameters. However, NDVI can only capture linear relationships between the difference of the near infrared (NIR) and red bands with various biophysical parameters. To address this limitation, the NDVI kernel (kNDVI) was proposed with certain reference values for optimal estimation, but it requires to be adjusted according to the object of study. 20 points distributed in different countries of Europe and Africa were chosen considering that the reflectivity information was representative of the whole study area and that it allowed us to distinguish diverse types of vegetation cover. Reflectivity data were obtained from the NIR and red bands of the AVHRR sensor at a temporal resolution every 8 days for the year 2012. Therefore, in an initial analysis, areas with little or no vegetation, such as areas covered by water bodies, glaciers, coastal strips and deserts, were discarded. Subsequently, we performed the optimization of this index using a Bayesian function that provides statistical improvements of its performance in estimating leaf area index (LAI) and compared it with the LAI product obtained from MOD15A2H.   The results showed a good overall performance of the index, since the correlation coefficient of kNDVI and LAI was 0. 55, 0.70 and 0.52 for forests, agricultural areas and pastures, respectively. Likewise, the validation of the equation obtained for the estimation of LAI using the optimized kNDVI, yielded correlation values greater than 0.70 (in agricultural areas). Based on this, we confirm the overall satisfactory performance of the index and highlight its suitability for monitoring terrestrial ecosystems as well as for LAI estimation. However, we recommend a more specific or regionalized analysis according to the different cover types that could contribute to improving the estimates. These results were implemented in Google Earth Engine (GEE), to facilitate the analysis and discussion of the products obtained