Two Decades of Satellite Eyes Reveal Where Mizoram Is Quietly Losing Its Green Cover

Deep in the eastern Himalayan extension, where steep north–south hill ranges rise above deeply incised river valleys, the Indian state of Mizoram has long been regarded as one of the country’s greenest landscapes. Forest cover there occupies roughly 86 percent of the state’s geographical area, according to the Forest Survey of India, and humid tropical to subtropical conditions deliver around 2,500 millimetres of rain each year. Yet a new two-decade assessment shows that even this densely vegetated mountain state is not immune to measurable landscape transformation. Using Landsat satellite imagery and a carefully calibrated vegetation-mapping technique, researchers have documented a net reduction of 275 square kilometres of mapped vegetation across Mizoram between 2005 and 2025, a decline that is far from evenly distributed across the state’s districts.

The study, conducted by K. Lalramngaizuala, Vishwambhar Prasad Sati, and F. C. Kypacharili of the Department of Geography and Resource Management at Mizoram University, was published in the journal Discover Forests. It set out to fill a conspicuous gap: while previous work in Northeast India had examined shifting cultivation, forest-fire susceptibility, livelihood conditions, or localised land-cover change, no consistent state-wide, district-resolved comparison of vegetation persistence, gain, and loss existed for the entire state. The researchers’ answer was to build one from the longest continuous Earth-observation archive available, pairing Landsat 5 Thematic Mapper imagery from 2005 with Landsat 9 Operational Land Imager and Thermal Infrared Sensor imagery from 2025, both processed at a 30-metre spatial resolution.

The technical heart of the study lies in how it handles the Normalized Difference Vegetation Index, or NDVI, one of the most widely used spectral indicators in remote sensing. NDVI exploits a simple but powerful physical contrast: chlorophyll pigments in healthy leaves absorb red radiation strongly, while the internal leaf structure reflects near-infrared light intensely. The index is calculated as the difference between near-infrared and red reflectance divided by their sum, producing values that theoretically range from minus one to plus one. High positive values generally signal dense, healthy vegetation, while values near or below zero correspond to exposed soil, built-up surfaces, or water. But the authors caution that NDVI is influenced by atmospheric effects, soil background, phenological variation, sensor characteristics, and saturation in dense vegetation, which makes naive comparisons across different satellite sensors risky.

That risk is precisely what the team’s methodological innovation addresses. Rather than applying a single fixed threshold to both years, the researchers used image-specific thresholding, extracting vegetation independently for each acquisition year before any temporal comparison. The rationale is that Landsat 5 TM and Landsat 9 OLI/TIRS differ in radiometric sensitivity, spectral response functions, calibration procedures, and detector characteristics, all of which influence the NDVI values recorded for identical land-cover conditions. Applying one universal threshold to both sensors could introduce inter-sensor classification bias and undermine the reliability of long-term change detection. Through iterative refinement against false-colour composite imagery and known landscape characteristics, the team settled on a threshold of NDVI greater than 0.48 for the 2005 Landsat 5 composite and greater than 0.61 for the 2025 Landsat 9 composite.

All imagery was processed in Google Earth Engine, the cloud-computing platform that has transformed the efficiency and reproducibility of regional land-cover analysis. The researchers used Landsat Collection 2 Level-2 Surface Reflectance products, which arrive already atmospherically corrected and analysis-ready. To minimise contamination and seasonal variability, they selected only scenes with less than 20 percent cloud cover acquired during comparable phenological periods. Cloud and cloud-shadow pixels were identified and removed using the quality assessment band supplied with the products, through bitwise masking prior to compositing. Median image composites were then generated separately for each study year, suppressing residual cloud artefacts and random noise while preserving representative surface conditions. Surface reflectance values were rescaled using the standard Collection 2 factors, and all rasters were clipped to Mizoram’s administrative boundary before analysis.

With binary vegetation masks built for each year, the team performed a spatial overlay analysis in a geographic information system to classify every pixel into one of four transition categories: persistent vegetation, vegetation loss, vegetation gain, or persistent non-vegetated land. The headline numbers tell a story of slow but real erosion. Vegetation remained the dominant mapped land-cover category in both years, but its extent fell from 20,558.8 square kilometres, or 97.52 percent of the state, in 2005 to 20,283.8 square kilometres, or 96.20 percent, in 2025. Non-vegetated land correspondingly grew from 522.1 to 797.2 square kilometres. Gross vegetation loss totalled 602.0 square kilometres, outpacing gross vegetation gain of 326.9 square kilometres by roughly 275 square kilometres.

The spatial pattern is where the findings become genuinely striking, because the change was anything but uniform. Vegetation loss clustered around major settlement areas and selected river valleys, including the surroundings of Aizawl, portions of the Tlawng and Tuirial river corridors, and parts of Champhai, Hnahthial, Khawzawl, Lawngtlai, and Saiha districts. At the district level, Lawngtlai recorded the greatest gross vegetation loss at 140.7 square kilometres, followed by Kolasib with 62.3, Aizawl with 60.3, Champhai with 58.2, and Lunglei with 58.0. Lawngtlai also suffered the largest net decline of 105.4 square kilometres, followed by Kolasib, Saiha, Aizawl, and Lunglei. In contrast, Mamit, Champhai, and Saitual registered small net gains of 13.9, 6.3, and 3.1 square kilometres respectively.

One district illustrates why the authors insist that gross gains, gross losses, and net change must always be reported together. Champhai experienced both substantial loss, at 58.2 square kilometres, and the highest gain of any district, at 64.6 square kilometres, yielding a small positive net balance. A district that appears stable in net terms may in fact have undergone considerable churn, with vegetation cleared in some places and regenerated or established in others. The authors also note that because district areas differ substantially, comparisons would be strengthened by area-normalised rates, expressing gain or loss as a percentage of each district’s geographical area or initial vegetation extent. The concentration of loss in accessible valleys and settlement-adjacent terrain is consistent with the spatial logic of land-use transformation in mountainous regions, where roads, settlements, cultivation, and infrastructure concentrate in relatively gentle terrain, but the binary classification cannot identify what replaced the lost vegetation.

That limitation shapes the study’s careful interpretation. The researchers explicitly frame their results as changes in mapped vegetation extent rather than as direct measures of forest condition, biomass, or vegetation health, and they decline to attribute the observed losses to any single driver. Shifting cultivation, settlements, roads, river-valley development, forest fire, and infrastructure expansion are all plausible contributors, since the mapped loss areas overlap with landscapes where these factors are known to operate, but the analysis does not quantify their independent or combined effects. Global meta-analyses of deforestation consistently show that vegetation change reflects interacting proximate and underlying drivers, including agriculture, wood extraction, economic factors, and governance, varying across regions and forest types. Climate variability, too, was not included in the analysis, so the authors warn against attributing the mapped change to climatic change. Field observations collected in April 2025 at representative locations across Mizoram provided qualitative verification of the 2025 classification, though a formal retrospective accuracy assessment was not possible.

What the study delivers, and what its authors argue matters most, is a repeatable geospatial baseline. The district-resolved map of persistence, gain, and loss gives conservation planners and land managers a defensible starting point for prioritising action, with particular attention recommended for Lawngtlai, Kolasib, Saiha, Aizawl, and Lunglei. The coexistence of gain and loss suggests that restoration and protection must proceed together, and that areas of apparent recovery require field assessment before they are celebrated as ecological regeneration, since the NDVI-based method cannot distinguish natural regrowth from plantation establishment. Future work, the researchers propose, should integrate multi-class land-use and land-cover mapping, fragmentation metrics, climate and terrain variables, and spatial driver modelling using tools such as logistic regression, geographically weighted regression, or random forest. In a region whose biodiversity and ecosystem services sit within the ecologically sensitive Indo-Myanmar Mountain landscape, that next step could turn two decades of satellite hindsight into a forward-looking strategy for keeping Mizoram green.

Subject of Research: Two-decade satellite assessment of vegetation extent change in Mizoram, Northeast India

Article Title: Vegetation dynamics in Mizoram Northeast India from 2005 to 2025 using landsat imagery and image specific NDVI thresholding

Article References: Lalramngaizuala, K., Sati, V. P., & Kypacharili, F. C. (2026). Vegetation dynamics in Mizoram Northeast India from 2005 to 2025 using landsat imagery and image specific NDVI thresholding. Discover Forests, 2(1), Article 74. https://doi.org/10.1007/s44415-026-00136-2

Image Credits: AI Generated

DOI: 10.1007/s44415-026-00136-2

Keywords: vegetation change, NDVI, Landsat, Google Earth Engine, remote sensing, Mizoram, Northeast India, forest cover, land-use change, eastern Himalaya, spatial analysis, conservation