Scaffold Geospatial Workflows

WORKFLOWS.md

@@ -0,0 +1,53 @@
+# Complex Workflows for EO Analysis
+
+This document outlines two complex workflows that demonstrate the power and flexibility of the `eo-processor` library. These workflows leverage the high-performance Rust UDFs to perform sophisticated Earth Observation analyses that would be slow and memory-intensive in pure Python.
+
+## Workflow 1: Deforestation Monitoring with Automated Change Detection
+
+This workflow automates the detection of deforestation events by combining temporal analysis, spectral indices, and spatial filtering. It is designed to be robust to seasonal changes and sensor noise, providing a reliable and scalable solution for monitoring forest cover.
+
+### Steps
+
+1. **Temporal Compositing:**
+   - A time series of Sentinel-2 images is composited using the `temporal_composite` function with the `median` method. This creates a cloud-free, seasonally representative baseline image.
+   - The same is done for a more recent time series to create a "current" image.
+
+2. **Spectral Index Calculation:**
+   - The Normalized Difference Vegetation Index (NDVI) is calculated for both the baseline and current images using the `ndvi` function. This highlights changes in vegetation cover.
+
+3. **Change Detection:**
+   - The difference in NDVI between the baseline and current images is calculated to create a delta-NDVI image. This highlights areas where vegetation has been lost or gained.
+
+4. **Texture Analysis:**
+   - The `texture_entropy` function is applied to the delta-NDVI image. This helps to differentiate between actual deforestation and other types of change, such as agricultural harvesting, which tend to have a more uniform texture.
+
+5. **Morphological Filtering:**
+   - A `binary_opening` operation is applied to the entropy-filtered image to remove small, isolated pixels that are likely to be noise. This is followed by a `binary_closing` operation to fill in small gaps in the detected deforestation areas.
+
+6. **Zonal Statistics:**
+   - The `zonal_stats` function is used to calculate the total area of deforestation within predefined administrative boundaries or protected areas.
+
+## Workflow 2: Crop Yield Prediction with Multi-Sensor Data Fusion
+
+This workflow combines data from multiple sensors to predict crop yields. It leverages the library's ability to perform complex classifications and temporal analysis to create a robust and accurate prediction model.
+
+### Steps
+
+1. **Data Fusion:**
+   - Time series data from Sentinel-2 and Landsat 8 are combined into a single, harmonized time series. This involves resampling the data to a common spatial resolution and performing radiometric normalization.
+
+2. **Complex Classification:**
+   - The `complex_classification` function is used to identify different crop types within the study area. This is a critical step, as different crops have different growth cycles and yield potentials.
+
+3. **Time-Series Analysis:**
+   - For each crop type, a time series of NDVI is extracted. The `detect_breakpoints` function is used to identify key phenological stages, such as green-up, peak growth, and senescence.
+
+4. **Feature Engineering:**
+   - A set of features is engineered from the NDVI time series, including the maximum NDVI, the length of the growing season, and the rate of green-up.
+   - The `texture_entropy` of the NDVI at peak growth is also calculated as a proxy for crop health and uniformity.
+
+5. **Yield Prediction:**
+   - A machine learning model, such as a random forest or gradient boosting model, is trained to predict crop yields based on the engineered features. The model is trained on historical yield data and validated against a held-out test set.
+
+6. **Zonal Statistics:**
+   - The `zonal_stats` function is used to aggregate the predicted yields to the field or administrative boundary level.