A Spatial and Temporal Model Adjusting Approach for Large Scale Multi-temporal Rangelands Monitoring

Abstract:

Mapping and monitoring of indicators of soil cover, vegetation structure, and various native and non-native species using remote sensing technology is a critical aspect of rangelands management. Due to the ever-changing landscape over space and time, the universal modeling approach where all available training data from different location and time are combined into one model is usually not an ideal solution for any location or time. As previous studies have shown, a universal modeling approach will have bias at a particular location and time combination due to the different statistical distribution that is used to train the model versus the potentially different condition at the said location and time. The spatial and temporal modeling bias is difficult to correct due to the different sampling density over space and the lack of repeat measurements at the same locations. In this study we present a spatial and temporal explicit approach to correct the universal modeling bias at the pixel level and according to the specific data collection time of satellite data. Our results indicate that a universal modeling approach is indeed appropriate for large scale multi-temporal rangelands monitoring if the correct spatial and temporal bias correction are administered.