Using remote sensing to address the effects of conservation agriculture in Cambodia

 

Conservation agriculture (CA) is crucial for agricultural sustainability in the long term. The primary components are as follows: (i) no-till, (ii) cover residue, (iii) specific management practices and technology, and (iv) integrated crop health and nutrition planning. 

This strategy has been demonstrated to offer several benefits, including improved soil health, water, and fuel conservation, improved environmental conservation, reduction in both water and wind erosion, etc. Among the most significant benefits, CA also contributes to the preservation and enhancement of soil organic carbon (SOC), especially in the topmost layer of the soil profile, making it a critical strategy for mitigating anthropogenic emissions.

Conventional approaches to ground SOC monitoring require extensive field sampling followed by costly laboratory analysis, posing a barrier to the majority of farming operations worldwide. Quantifying SOC at a lower cost is crucial for delivering accurate data about soil characteristics and informing about the impacts of various management practices, particularly on smallholders and developing countries. 

Earth observation monitoring can assist in this endeavor by allowing for the mapping of crop cover and even residues, as well as providing information about various practices. Usually, greater moisture or SOM in the soil, more light is absorbed, resulting in a darker soil color and decreased reflectance. The wavelengths in the visible, near infrared, and shortwave infrared bands have been demonstrated to correspond with various SOC concentrations in a variety of global contexts. This, in combination with statistical analysis, might enable the investigation of changes in SOC over time in locations where soil monitoring or historical data are unavailable or unfeasible, hence informing agricultural systems and practices. 

All of the above is especially true in nations like Cambodia, where small-scale family farms dominate the terrain. According to USAID, 58 percent of farms only grow one crop, rice, and 26 percent grow two or more, with an average field size of 1.6 hectares for more over 60 percent of households. All these features, combined with socioeconomic factors, make farmers reluctant to take risks, which aggravate food insecurity in many cases. 

Additional data is required to assist these communities in achieving sustainable growth while also protecting the ecosystems.
 

The following are the proposed approaches to conduct the analysis: 

1. The employment of convolutional neural networks to assist in classifying conservation and traditional agriculture. 

   1. The specific methods to accomplishing this goal would be based on the following: 

      1. Create a layer defining the field boundaries. Through the use of imagery with very high spatial resolution and image classification techniques. This layer can be used for a variety of purposes in the future but would be employed in our project as an input layer to address the information over each field. 

      2. Compute time series for each field across the area of interest. This step would focus on evaluating and comprehending each field's temporal dynamics. 

      3. Land use and classification of fields managed according to conservation practices against those under traditional management practices. Train and validate a model based on the knowledge gathered in the preceding steps and on ground truth data. 

2. Using regression models to predict carbon content. This stage would require field data on carbon concentrations and soil parameters to verify and validate the model. This should be conducted over fields under conservation techniques and traditional tillering. 

   1. The specific processes involved in this endeavor would be as follows:

      1. Collecting soil data and carbon content 

      2. Obtaining the required satellite images and weather/ climate information. 

      3. Performing a regression analysis to test the performance of different model in predicting SOC and other properties from satellites.