Photointerpretation Data set

Published: 7 March 2023| Version 1 | DOI: 10.17632/4f8vdds973.1
Matias Castillo Armijo


In order to study the evolution of land cover change in this desert region, we analyzed land cover change process of four valleys with different levels of agricultural development. Specifically we want to corroborate the hypothesis that in the last two decades, land cover change in the agricultural valleys of Arica and Parinacota region is oriented to intensification and is occurring at a high rate of change, directly related to some common biophysical drivers of the landscape and to the agronomical conditions, like elevation or water quality. We found that agricultural intensification is occurring at a high rate of change of land cover, mainly to the spreading of netting and low crops ind detriment of native shrubland or more complex cover like fruit groves. Results, demonstrate that these valleys are under intensification in differente degrees and landscape simplification is expected in the near future. Supplemetary 1 and 2 are the results about the transition matrices and rate of change supporting our hypothesis about intensification.


Steps to reproduce

This dataset correspond to the photointerpretation of satellite images from 2003 to 2019, using a systematic 400x400-m grid of sampling points. A total of 1827 points were distributed over the four valleys (Lluta: 468, Azapa: 806, Vítor: 218, Camarones: 335). For each sampling point we assigned one of ten land cover classes that correspond to the class observed at the specific coordinates: bare soil (BS), water (WA), buildings (BU), anti-aphid netting and greenhouses (NG), low crops and vegetables (LC), fruit groves, including olives (FG), corn (CN), grassland or pasture (GL), native shrubland (NS) and set-aside or fallow (SA). To accommodate for potential agricultural expansion over years, and to allow for neighborhood analyses, we included in our sampling a minimum of 200-m buffer area around the vegetation limits of each valley in 2003. For this reason, our sample included a disproportionate amount of bare soil (40-50% of each valley). Photointerpretation was conducted using Google Earth Pro satellite imagery from 2003 to 2019, due to its image accuracy, representativeness, and ease of visual interpretation (Barbosa & Campos, 2011). For each year, we superimposed the grid of sample points over one of the available satellite image sets (preferably spring) with the best image quality and determined the land cover for each point manually, through a detailed observation considering texture, color, and landscape patterns. We defined a specific land cover class for every point at each valley. The photointerpretation was conducted using an eye observer elevation of 100 to 1000-m. For better accuracy of the land cover class assignment, ground truthing was available for some sample points (≈200) and was used to corroborate the land cover class assigned. To reduce the effect of many observers or any bias in land cover visual classification (Tarko, 2019), only one trained observer (MC) determined the class of land cover for all the points of the complete imagery set and sample. Land cover transition matrices were obtained with RStudio using the function "xtabs", speifically with the script (using as example Vitor data) xtabs(~ X2003 + X2007, data = vitor_matriz) crosstabs03_07 <- xtabs(~ X2003 + X2007, data = vitor_matriz) Aditionally, accumulative rate of change were obtained using land cover point data of each year. The specific formula for rate of change is in the sheet 2, and was calculated with excel.


Universidad de Chile


Landscape, Agricultural Land, Land Degradation