A Markov Chain approach to evaluate the impact of CAP Greening on land use in Lombardy region

A recent reform has redesigned Common Agricultural Policy (CAP) contents over the programming period 2015-2020, introducing important changes in the funding strategies, called “Greening”, which reward “agricultural practices beneficial for the climate and the environment” (EU regulation 1307/2013), namely i) arable crops diversification, ii) maintenance of permanent grassland and iii) ecological focus areas (EFA). These farm practices regard, and potentially influence, farmland allocation, particularly arable land and grassland.

We implemented a method to evaluate ex-post impact of the CAP greening payment on farmland use changes, based on real georeferenced data on farmland allocation, collected in Lombardy Region, in Northern Italy, over the period 2011-2016. On this ground, we pointed our attention on analysing at a very detailed (parcel) level the temporal and spatial dynamics of farmland use transitions before and after the introduction of greening commitments, resulting, all together, into an analysis of the evolution of about 3 millions parcels over 6 years. We thus were working in a Big Data framework.

The system has been modelled as a Markov chain, where each land unit (hectare, group of hectares, or parcel, depending on the spatial scale at which we work) evolves, from one year to the other, into one of 23 relevant cultivation classes which have been identified.

Let us denote by pij(t) the probability that a land unit evolves (i.e. is cultivated) from class i to class j, from year t to year t+1. Our aim was to check if any statistically significant change in the transition probabilities pij(t) and/or in the spatial distribution of the 23 cultivation categories, took place after the introduction of greening (that is between 2014 and 2015). We first applied a test of stationarity (Anderson and Goodman, 1957) both to the spatial distribution of the cultivation categories, and to the transition probabilities pij(t), for t varying from 2011 to 2014, in order to check if they may be assumed constant in time, before the application of greening. Unfortunately the test revealed a strong non stationarity, due to a possible correlation among data. This causes a problem in the statistical analysis, since the “physiological” variability registered before the new CAP must be filtered out for a correct comparison with the changes occurred from 2015 onwards. We solved the problem by introducing a new type of weighted χ2 test (Aletti et al., 2018), in which we determine the correct statistical unit that must be considered to accept the hypothesis of stationarity in a set of panel data. We applied this test to the complete set of available years (2011-2016) and we found evidence of change during 2015 in both the spatial distribution of the 23 cultivation classes and the transition probabilities of many relevant cultivations, like maize, maize for silage, wheat, soybean, etc.

Furthermore we computed the Gini index to measure the heterogeneity of cultivations and the transition probabilities for the cultivation classes, which resulted significant to the weighted χ2 test (see example in Figure 1).

Each quantity was computed in a rectangular grid overlapped to our geographical region of interest, In this way we can visualise the zones of Lombardy which have mainly been affected by the greening policy.

Preliminary results shows that for certain crops the spatial heterogeneity and probability of transition increased after the introduction of greening. Such evidence is particularly relevant for maize (all the uses), other cereals and soybean. Pointing our attention to maize (Figure 1), we can observe that after the introduction of greening the red zone (the core of maize monoculture in Lombardy) has been partly “eroded”, introducing more variability in the crop rotation.


Team and funding: the project has been developed by G. Aletti, D. Bertoni, G. Ferrandi, A. Micheletti, D. Cavicchioli, R. Pretolani and is funded by the Fondazione CARIPLO project CAPTION.


Aletti G, Bertoni D., Ferrandi G., Micheletti A., D. Cavicchioli, Pretolani R., Farmland Use Transitions After the CAP Greening: a Preliminary Analysis Using Markov Chains Approach . Preprint, submitted for publication (2018).

Anderson T.W., Goodman L.A., Statistical inference about Markov chains, Ann. Math. Statist., 28, 1 (1957), 89-110.