Methods and systems for optimizing hidden Markov Model based land change prediction

A system comprising:

• one or more processors;
• a communication interface device;
• one or more internal data storage devices operatively coupled to the one or more processors for storing, wherein the one or more processors are configured to:
• classify remotely sensed data received for a pre-defined interval of time from a plurality of sensors into a plurality of land use land cover classes (LULC) as Vegetation (V), Impervious Surface (I), and Soil (S);
• provide data corresponding to a plurality of socio-economic driver variables for a pre-defined interval of time from a database, wherein the plurality of socio-economic driver variables include a Gross Domestic Product of a scale as national, an Interest Rate Cycle of a scale as national, a Consumer Price Index Inflation of a scale as national, a Gross Fixed Capital Formation of a scale as national, an Urban Population Growth Rate of a type social and a scale as national, a Per Capita Electricity Consumption of a scale as regional, a Total Road Length Added of a scale as regional;
• define at least one Hidden Markov Model (HMM) with the plurality of LULC classes of V, I, S as hidden states and plurality of socio-economic driver variables as observations, wherein each of the socio-economic driver variables is modeled as a Gaussian distribution with a mean and a standard deviation, and each Gaussian distribution is divided into quartiles;
• train the defined at least one HMM with data for the plurality of LULC classes of V, I, S as hidden states and the data corresponding to plurality of socio-economic driver variables as observations from the database for generating at least one HMM state transition probability matrix wherein a plurality of transition sub-models are defined for land cover change types as V to S, V to I, I to V, I to S, S to V, S to I excluding transitions from or to water states and a digital elevation model, road layers are used as a driver variable for each of the transition sub-models, wherein the driver variables are transformed to respective evidence likelihood based transition potential images and fed into a spatio-temporal model to yield a consolidated potential transition image for each of the land cover change types;
• generate data corresponding to a plurality of scenario based temporal variables related to future growth scenarios, steady or negative economic growth, or recession and storing the data corresponding to a plurality of scenario based temporal variables in the database, wherein the plurality of scenario based temporal variables are defined in terms of quartile numbers, and data values for the scenario are created by generating random numbers from the specified quartiles of the socio-economic driver variables;
• train the defined at least one HMM with data for the plurality of LULC classes of V, I, S as hidden states and the data corresponding to a plurality of scenario based temporal variables from the database as observations for generating at least one other set of HMM state transition probability matrix;
• integrate the generated at least one HMM state transition probability matrix or the generated at least one other set of HMM state transition probability matrix with at least one spatio-temporal model to obtain at least one integrated model, wherein the spatio-temporal model deploys the digital elevation model, proximity to primary roads, and slope to model urban growth scenario;
• predict LULC changes to generate at least one prediction image;
• and
• prune a socio-economic driver variable from the plurality of socio-economic driver variables by:
• training the defined at least one HMM with all the socio-economic driver variables;
• re-training the defined at least one HMM without the socio-economic driver variable;
• assessing an impact of the socio-economic driver variable;
• and
• prune the socio-economic driver variable when the re-trained at least one HMM is same as the trained at least one HMM and the pruned socio-economic driver variable have little impact or no impact on the training of the defined at least one HMM.

A computer program product comprising a non-transitory computer readable medium having a computer readable program embodied therein, wherein the computer readable program, when executed on a computing device, causes the computing device to:

• classify remotely sensed data received for a pre-defined interval of time from a plurality of sensors into a plurality of land use land cover classes (LULC) as Vegetation (V), Impervious Surface (I), and Soil (S);
• provide data corresponding to a plurality of socio-economic driver variables for a pre-defined interval of time from a database, wherein the plurality of socio-economic driver variables include a Gross Domestic Product of a scale as national, an Interest Rate Cycle of a scale as national, a Consumer Price Index Inflation of a scale as national, a Gross Fixed Capital Formation of a scale as national, an Urban Population Growth Rate of a type social and a scale as national, a Per Capita Electricity Consumption of a scale as regional, a Total Road Length Added of a scale as regional;
• define at least one Hidden Markov Model (HMM) with the plurality of LULC classes of V, I, S as hidden states and plurality of socio-economic driver variables as observations, wherein each of the socio-economic driver variables is modeled as a Gaussian distribution with a mean and a standard deviation, and each Gaussian distribution is divided into quartiles;
• train the defined at least one HMM with data for the plurality of LULC classes of V, I, S as hidden states and the data corresponding to plurality of socio-economic driver variables as observations from the database for generating at least one HMM state transition probability matrix wherein a plurality of transition sub-models are defined for land cover change types as V to S, V to I, I to V, I to S, S to V, S to I excluding transitions from or to water states and a digital elevation model, road layers are used as a driver variable for each of the transition sub-models, wherein the driver variables are transformed to respective evidence likelihood based transition potential images and fed into a spatio-temporal model to yield a consolidated potential transition image for each of the land cover change types;
• generate data corresponding to a plurality of scenario based temporal variables related to future growth scenarios, steady or negative economic growth, or recession and storing the data corresponding to a plurality of scenario based temporal variables in the database, wherein the plurality of scenario based temporal variables are defined in terms of quartile numbers, and data values for the scenario are created by generating random numbers from the specified quartiles of the socio-economic driver variables;
• train the defined at least one HMM with data for the plurality of LULC classes of V, I, S as hidden states and the data corresponding to a plurality of scenario based temporal variables from the database as observations for generating at least one other set of HMM state transition probability matrix;
• integrate the generated at least one HMM state transition probability matrix or the generated at least one another set of HMM state transition probability matrix with at least one spatio-temporal model to obtain at least one integrated model, wherein the spatio-temporal model deploys the digital elevation model, proximity to primary roads, and slope to model urban growth scenario;
• predict LULC changes to generate at least one prediction image;
• and
• prune a socio-economic driver variable from the plurality of socio-economic driver variables by:
• training the defined at least one HMM with all the socio-economic driver variables;
• re-training the defined at least one HMM without the socio-economic driver variable;
• assessing an impact of the socio-economic driver variable;
• and
• pruning the socio-economic driver variable when the re-trained at least one HMM is same as the trained at least one HMM and the pruned socio-economic driver variable have little impact or no impact on the training of the defined at least one HMM.