<p><strong>Abstract.</strong> Growing season time frames can be estimated and mapped using the vegetation indexes mapping and analysis. This approach brings significant benefit consisted in the ability of detailed (highly discrete in the meaning of spatial resolution) mapping of spatial differences in growing season stage and length. In comparison with interpolation of ground air temperature (applied when using temperature to detect growing seasons), real spatial resolution raises to kilometers per pixel and higher, while nodes of ground observation network can be spaced by thousands of kilometers in some regions. Our ongoing study is devoted to design a processing chain for mapping of growing season time frames basing on vegetation indexes data with close-to-one-day time resolution. We used MOD09GA dataset as an initial data. Data processing was implemented in Google Earth Engine big geospatial data platform.</p>

We applied the Normalized Difference Vegetation Index (NDVI) as an indicator reflecting dynamics of vegetation cover growing. NDVI time series were derived from the Terra/MODIS satellite imagery (MOD09A1 product was used, available from the open archive of the US Geological survey). Data for 2006-2016 were used. The ground test plate is located in the Karelia Republic, Russia (61°,7218 N, 34°,3689 E). Moving average method with three inflection points was applied to ensure a piecewise monotone smoothing of NDVI time series. We also applied meteorological data (surface air temperature, average daily precipitation, average snow cover heightcollected at Petrozavodsk meteorological station in 2006-2016). Used meteorological data were derived from the website of the Russian Federal Service for Hydrometeorology and Environmental Monitoring -ROSHYDROMET (http://meteo.ru). As a result, we designed a methodology for selection and processing (including initial compilation, smoothing and interpolation of time series) of remote sensing and meteorological observations to generate training data and control datasets demanded by artificial neural network. Obtained results demonstrate that neural network trained using meteorological data can predict NDVI values. Less than 15% relative error of the NDVI prediction was gained at our ground test plate.

Climate is a dynamic and extremely multifactorial phenomenon. Various climate models can be implemented to assess and forecast climate change, which incorporate various numbers of climate factors and consider these factors either as independent or as complex (through grouping more or less similar factors). In accordance to such flexibility of climate modeling, issues of models verification and model-based computations evaluation remain most important tasks. Solution for the verification and evaluation problems can be achieved, in particular, through implementation of a complex geographic information system (GIS) capable to provide accumulation and joint analysis of retrospective and real-time arrays of climate and supplementary geospatial data. Our study series is devoted to the joint analysis of the dynamics of climate and vegetation cover parameters in the Northern regions. Currently we are presenting initial model of the geospatial database, which includes ground-based meteorological observations and satellite remote sensing data and allows assessment of various climate parameters (surface air temperature and humidity trends, framing dates and duration of growing seasons, etc.).