Evaluating pollution‐related damage and restoration success in urban forests with participatory monitoring and digital tools

Peri‐urban forest monitoring requires indicators of vegetation damage. An example is the sacred fir (Abies religiosa) forests surrounding Mexico City, which have been heavily exposed to tropospheric ozone, a harmful pollutant, for over 4 decades. We developed a participatory monitoring system with which local community members and scientists generated data on ozone tree damage. Santa Rosa Xochiac rangers (13) used the digital tool KoboToolBox to record ozone damage to trees, tree height, tree ages, tree condition, tree position, and whether the tree had been planted. Thirty‐five percent of the trees (n = 1765) had ozone damage. Younger trees had a lower percentage of foliage damaged by ozone than older trees (p < 0.0001), and asymptomatic trees tended to be younger (p < 0.0001). Symptomatic trees were taller than asymptomatic trees of the same age (R2c = 0.43, R2m = 0.27). Involving local communities facilitated forest monitoring and using digital technology improved data quality. This participatory system can be used to monitor forest condition change over time and thus aids restoration efforts driven by government or local communities’ interests, facilitating local decision‐making.


INTRODUCTION
Monitoring forest condition has significantly increased since the establishment of international agreements for halting climate change and biodiversity loss (Danielsen et al., 2011;Romijn et al., 2015).Forest monitoring is usually based on satellite image data, including land-cover and land-use images and vegetation inventories (Chirici et al., 2012;Romijn et al., 2015).This type of monitoring is useful for estimating forest cover and carbon stocks.However, a higher resolution is needed at the local level, for instance by including variables that are relevant only for a specific site and that are more useful for decision-making.Such is the case of urban and peri-urban forests, where monitoring strategies should include aspects related to pollution damage, which are usually not considered (Percy & Ferretti, 2004).
Tropospheric ozone is one of the most common and harmful air pollutants (Churkina et al., 2017), and it has large negative effects on peri-urban forests worldwide (Cho et al., 2011;Percy et al., 2007;Wipfler et al., 2005).Ozone pollution in Mexico City (CDMX) is particularly severe because of the city's large population, geographic location, and topography (Bravo-Alvarez & Torres-Jardón, 2002).The CDMX is in a high elevation atmospheric basin at 2240 m asl, at a latitude of 19 • N, and is surrounded by high mountain ridges on 3 sides.Ozone has been a problem at CDMX since the early 1980s (Alvarado-Rosales & Hernández-Tejeda, 2002;O'Neill et al., 2004;Riojas-Rodríguez et al., 2014) (Appendix S1).This pollutant is generated in the city and spread out of the basin by dominant winds through a number of mountain passes across the surrounding mountains in the southwest.Mountains forming this valley are dominated by sacred fir (Abies religiosa) forests, particularly those of the Parque Nacional Desierto de los Leones (PNDL) (CONANP, 2006).We estimated the AOT40 (accumulated dose of ozone over a threshold of 40 ppb) spatial distribution accumulated during the growing season of 2012 for the PNDL with data from continuous measurements of ozone in several up-and downwind monitoring sites.The estimated AOT40 for the PNDL in 2012 was ∼35,000 ppb×h, which is 5 times higher than the recommended level for the protection of forest trees, and this value was 5000 ppb×h from April to September (CLR-TAP, 2014; Klingberg et al., 2014).This high concentration is associated with ozone damage to plants, which includes reddish needles that fall off the tree within the first 3 years of exposure, decreased individual vigor, and increased susceptibility to pests relative to trees in other areas of the species' range (Alvarado-Rosales & Hernández-Tejeda, 2002).Although the ozone-related forest decline in the PNDL was initially reported decades ago (Ciesla & Macías-Sámano, 1987), there is currently no estimate of the total damaged area or the percentage of affected trees.
Other than pollution, the decline of the peri-urban forests at CDMX has been exacerbated by anthropic forest fires.In 1998, ∼1000 ha of mature fir forests were burned within and around the PNDL (DOF, 1998).Burned areas were since declared an ecological recovery zone (DOF, 1998), but restoration success has been limited, in part because pollution-related stress was not considered in the restoration plans.Indeed, there is a systematic lack of data for the affected area, despite the management plan stating that updating and archiving computerized data on the extent of ozone damage was essential to achieve conservation and restoration goals (CONANP, 2006).Part of the delay was attributed to a lack of equipment and qualified personnel.There is also limited accessibility to the scarce gathered data, most of which are stored in poorly systematized reports.This has resulted in disconnected actions among government agencies, academia, and local communities.For instance, local communities near PNDL, such as Bienes Comunales Santa Rosa Xochiac (SRX), have their own forest conservation and restoration activities, which include reforestation with local germplasm selected and produced in their own nursery.These activities sometimes overlap and are uncoordinated with those of local (i.e., SEDEMA [Mexico City Minister of Environment]) or national conservation agencies (i.e., CONANP [Mexican Agency of Natural Protected Areas]), which often exclude local communities from monitoring and research efforts.
To fulfil data needs, we, members of a local community (SRX), 2 government agencies (SEDEMA and CONANP), and academic researchers (CONABIO [National Commission for the knowledge and use of biodiversity], UNAM [National Autonomous University of Mexico]), joined forces to develop a participatory monitoring system.Such an approach (see Danielsen et al. [2009] for a typology of participatory) allows the compilation of large data sets at minimal cost and empowers local communities through direct participation in science (Danielsen et al., 2011;Holck, 2008).Participatory approaches have been criticized because they tend to produce messy data with limited research uses (Dobson et al., 2020).However, incorporating digital technology, such as mobile devices, to aid data collection may increase data quality for monitoring purposes (DeVries et al., 2016;Pratihast et al., 2013).We used mobile devices and open-source digital technology in a participatory forest survey to generate large amounts of high-quality data for use in studying the effects of air pollution on a peri-urban forest in PNDL.We sought to identify damage trends and their relationship with individual tree age and height and healthy adult trees (i.e., without the typical symptoms of ozone damage [Hernández-Tejeda & Benavides-Meza, 2015;Reyes-Galindo, 2019]) that could serve as a germplasm source for forest restoration (Reyes-Galindo, 2019).To do so, we determined how many putatively tolerant individuals existed, where they were located, and what percentage of naturally regenerated and replanted trees they represented.

Study species
Abies religiosa (sacred fir or Mesoamerican fir) is a conifer, native to the mountains of central Mexico (Farjon, 2010), locally known as oyamel.Needles appear during the growing season (beginning of May) after the bud break occurs (April).Bud scars are evident in branches and allow inferring needle age according to their location on the branch (Appendix S2).This species retains needles on branches for up to 6 years, although they fall from ozone-affected trees within ∼3 years after exposure (Alvarado-Rosales & Hernández-Tejeda, 2002).

Participatory monitoring design
The monitoring we describe was part of a broader transdisciplinary project that followed methods and principles of Lang et al. (2012).Initially, we conducted a series of workshops involving all parties at which ozone pollution was identified as one of the main problems hampering reforestation and diminishing overall forest condition (Figure 1).Participatory monitoring was proposed to complement scientific research and address local needs.We codesigned and implemented such a monitoring program following steps shown in Figure 1.All parties agreed that before continuing restoration actions, the following questions needed to be answered (Figure 1).Are there differences in ozone damage between naturally regenerated and replanted trees in terms of number of affected trees and amount of damage per tree?Do trees in open areas show more ozonedamage symptoms than understory trees or trees growing below larger trees or bushes (hereafter nursed trees) (Flores & Jurado,FIGURE 1 Steps in planning and implementing participatory monitoring of tree condition in a forest affected by ozone pollution.Team members assigned for each step: local community Santa Rosa Xochiac (SRX), Mexico City Minister of Environment (SEDEMA), Mexican Agency of Natural Protected Areas (CONANP), and the academic team, which included researchers and students from National Commission for the knowledge and use of biodiversity (CONABIO) and National Autonomous University of Mexico (UNAM).
2003; Schöb et al., 2013)?Do older trees show more ozonedamage symptoms than younger trees?Do healthy trees grow more than individuals of the same age with ozone-damage symptoms?Are there differences in ozone damage between trees growing at different elevations?
To fill information gaps, the academic team proposed collecting data on general tree characteristics (variables in Appendix S3) and pollution damage to trees and taking photographs of trees (Appendix S4) (Figure 1).Tree damage categories were healthy (no apparent damage), ozone damaged (symptoms as described in Alvarado-Rosales and Hernández-Tejeda [2002]), and other.Other included drought damage, as described in Schuldt et al. (2020), and signs of fungal infection, as described by Chastagner (2001), or insect attack, as described by Burleigh et al. (2014).For ozone-damaged trees, the approximate percentage of affected foliage was also included as a categorical variable.For replanted individuals, their provenance was noted.

Devising the KoboToolbox form and local training
KoboToolbox (2019) (hereafter Kobo) is a series of opensource tools for use in collection of field data, especially in challenging environments (https://www.kobotoolbox.org).It includes a web application, where forms can be developed and filled in online, and an application for mobile devices, where forms can be filled in offline (Appendix S5).Recorded data are sent to the server once an internet connection is available.The web application is available on the native servers or can be installed as a stand-alone on local servers for complete governance over the data.We installed Kobo on CONABIO's servers.
The academic team wrote and tested the Kobo forms (Figure 1) based on identified monitoring variables.The team used automated features to prevent, as much as possible, common problems that arise when data are collected by large teams (Appendix S3).The rest of the team, including rangers, provided feedback and helped develop the final version (Figure 1).
The SRX forest rangers were formal members of the community or close relatives of a community member, known as comuneros, and had reforestation, fire prevention, and firefighting training, but they had no previous experience in evaluating plant condition.The community chose 13 rangers to participate in the monitoring: 10 men and 3 women ranging in age from 17 to 54 years and with varied schooling levels (high school to undergraduate degrees).Rangers were paid according to local wages (monitoring was done during the season when rangers are not paid by the government).
To ease understanding of and to minimize mistakes in filling out the form (Figure 1), we used easy-to-evaluate categories of tree diameter (e.g., beer bottle = 10 cm), provided a small visual guide with photos of the most common damage symptoms, and taught them how to count nodes per branch and trunk (Appendix S6).Because of the COVID-19 pandemic, training was performed partially online and then in situ.

Participatory monitoring implementation
Sampling sites were selected by all parties (Figure 1) and included degraded areas with natural regeneration and replanted trees in PNDL and SRX territories (Figure 2a,b).Monitoring was conducted for 10 days in December 2020 and January 2021 (Figure 1h) by the rangers and 1 academic team member.Rangers used their own mobile devices.
Rangers were divided into groups of 3-4 people.They examined 48, 10 × 10 m plots.Sampling sites were separated by >500 m.There were 3 plots per sampling site, and plots were separated by ∼50 m (Figure 2c).Coordinates were taken at plot centers and individual trees.All fir trees taller than 20 cm or of >0.5 cm diameter were recorded in each plot.These included a mixture of natural generated and replanted trees or only natural generated trees.Needle samples were taken from all trees with branches below 2-2.5 m.Needles were stored in silica gel for use in other research.Trees were labeled with biodegradable wood tags (1 × 10 cm) tied with natural-fiber thread.
Data were uploaded to CONABIO's Kobo server after each working day.The academic team member verified successful data uploading and answered follow-up questions in a group chat.Among others, she corroborated that the number of trees registered on the server corresponded to the collected material (envelopes with needles and unused labels) and sent the automated Kobo report to the ranger teams, which included a list of the correctly uploaded records.
Data were validated (Figure 1) by visually comparing records in the Kobo web application with the photos taken for each individual tree.Data that could not be corroborated were excluded from subsequent analyses.To support data validation, we developed a custom R (R Core Team, 2021) script that allowed detection of errors in plot number and missing data and verified that all records were validated.Errors were manually double-checked and corrected in the Kobo web application.
Results of preliminary data analyses were presented to SRX and SEDEMA, and an anonymous survey was sent to the rangers for feedback.A second form, which included survey results and preliminary analyses, was sent to all participants in which they were asked their opinion on the preliminary results, their motivation to participate, and what individual and community benefits they expected (Appendix S7).Results of this form were discussed in a workshop where additional experiences were shared to improve future monitoring and sampling dynamics (Figure 1).

Statistical analyses
We conducted a 2-way analysis of variance (ANOVA) with interaction and blocking to examine the effects of tree condition, origin (replanted or natural), and position (exposed or nursed) on the percentage of damaged trees.We estimated the percentage of damaged trees (ozone or ozone and others) by plot and category, normalized the estimate (log transformed), and performed a 2-way ANOVA that included sites as blocks (n = 16 sites, with 3 plots each).Because not all blocks had replanted trees, we also performed 1-way ANOVAs for tree origin and position independently.Using a linear mixed model (glmer), we tested the effect of tree condition on height growth, considering the interaction with tree origin and position, assuming a Gamma distribution, and including plots as blocks.The analysis was performed only with trees 7-11 years old (corresponding to the first and third quartiles of ozone damaged trees [Figure 4b]), so each category would have the same number of trees (n = 125, 68, and 209) and the same minimum and maximum age.We compared the goodness of fit of the models with Akaike's information criterion.The better model was tree height ∼ tree ages + health status + tree position + replanted + (1| plot).Finally, we searched for differences in ozone damage between trees growing at different elevations with a linear model.Analyses were performed in stats 4.1.2for R (R Core Team, 2021).Results were visualized with ggplot2 3.3.5 (Wickham, 2016) and ggmisc 0.4.4 (Aphalo, 2021).

Data collection
Each ranger registered an average of 141 (SD 19.65) trees, for a total of 1781 records in 48 plots (Appendix S8).After manual validation, we kept 1765 (99%) records.Needles were collected for 1687 (95.6%) of these individual trees (the rest were dead or too tall).Age was confidently inferred for 1564 (87.7%) trees 2-24 years old.Older trees were excluded from age-related analyses because it was impossible to accurately determine their age.

Extent of ozone damage in fir forests
About 489 trees (27.7%) were apparently healthy.There were 125 individuals (7%) showing ozone-related symptoms, and 499 (28.3%) had both ozone symptoms and other types of damage (drought, insect attacks, etc.) (hereafter ozone and other damage).The remaining 950 plants (37%) exhibited damage not related to ozone.Twenty percent had a single symptom of damage other than ozone, 14% had a combination of more than one type of damage other than ozone, and the rest were recorded as dead (Appendices S4 & S9).The percentage of trees with ozone damage per plot ranged from 4.5% to 84%.Only 1 plot had no ozone-damaged trees (Figure 2d; Appendix S8).A 2-way ANOVA revealed significant main effects of tree position (F 1, 95 = 4.15, p = 0.04) and origin (F 1, 95 = 4.65, p = 0.03) and a significant block (plot) effect (F 15, 95 = 3.34, p < 0.001).The interaction effect between tree position and tree origin was nonsignificant (F 1, 95 = 0.07, p = 0.79) (Figure 3).When tree position and tree origin were tested in independent 1-way ANOVAs, similar results were obtained (Appendix S10), suggesting an important spatial effect in tree condition (Appendix S11).We found no correlation between elevation, and ozone damage among plots (R = 0.01) (Appendix S12).Asymptomatic trees tended to be younger than individuals showing ozone-related damage (alone or in combination with other factors; p < 0.001) (Figure 4a,b).Younger trees had a lower percentage of foliage damaged by ozone than older trees (p < 0.0001) (Figure 4c,d).Asymptomatic trees tended to be younger than trees affected by ozone and other damage (p < 0.0001) (Figure 4a,b).Finally, symptomatic trees were taller than asymptomatic trees of the same age (R 2 c = 0.43, R 2 m = 0.27) (Figure 5; Appendix S13).

Lessons for designing participatory monitoring with digital tools
Kobo made data gathering and computerized archiving easier and allowed inclusion of information that was not available elsewhere, such as year of reforestation for replanted individuals.Data quality was improved and validation (including photo quality) eased by in situ training with test forms.According to their feedback, rangers agreed that using Kobo in the field was easier than taking notes on paper for later data entry.
Most rangers (8 of 12) were surprised by the number of damaged trees and the extent of damage, including ozone and other damage.All rangers agreed that incorporating this tool and approach in their regular activities in the park should facilitate evaluating forest condition.They further thought our approach increased their knowledge of threats to forests, which is what they valued the most about participating in the monitoring.This fits well with their motivation to become community forest rangers: a genuine interest in forest conservation.
The SEDEMA participants commented that they currently have data from previous monitoring efforts that have not been included in management plans because of mixed and inadequate archiving (e.g., paper, spreadsheet), which implies that a significant amount of time would be needed to make that data accessible for statistical analyses.According to them, using digital tools, as proposed in this study, will overcome that challenge in future surveys.
Data collection tools similar to Kobo exist (e.g., Teamscope, REDcap), and we are confident that they all could be used easily in similar initiatives.The experience and ability of the local rangers to work in the forest, together with the optimization of data recording with Kobo, allowed for more efficient sampling (1781 trees in 10 days) than that previously conducted by SRX, SEDEMA, or small academic teams (estimate of 250 trees in the same amount of time).
For instance, using the photographs taken during the monitoring allowed us to corroborate the damage (each photo was reviewed by the same person), and customizing the form prevented common errors that occur when working with a diverse group of people of different backgrounds.Finally, using the devices' GPS instead of manually introducing coordinates helped prevent typos (Mapit GIS, 2020).These practices were much more efficient than manual data cleaning, which is not only more time consuming, and error prone, but also more difficult to reproduce and limited to known geographical areas, which makes it impractical for large data sets (Zizka et al., 2019).Likewise, our scripts for detecting atypical data helped us identify and correct errors more quickly, for instance by double checking with the photographs when needed.
Using a digital collection tool also allowed for generation of daily reports, which, as shown elsewhere (Dobson et al., 2020), motivated the field team and increased the chances of success of the participatory monitoring.An advantage of our method is its flexibility for incorporating changes and recommendations for improving data collection.For instance, we planned to update field guides with photos taken by rangers for future monitoring.This will make their work visible to their colleagues and help motivate future teams.Other improvements suggested by rangers include having ladders for collecting cones, improving plots marking, taking precise width and height measures, including unconsidered sources of damage (which were all treated as other), and evaluating other tree species.

Variables influencing ozone damage in CDMX peri-urban forests
Our participatory monitoring allowed updating much needed data for the PNDL management plan.In 1983 aerial photographs, 28% of the fir forest area showed ozone-damage or beetle infestation symptoms (Macías-Sámano & Cibrían-Tovar, 1989).Despite a considerable reduction of ozone pollution in CDMX in the last 2 decades (SEDEMA, 2018) (Appendix S1), ozone damage is ongoing.Although our data did not allow us to estimate the percentage of damaged area, we found that ∼35% of trees showed ozone-related symptoms, and that although affected trees occurred in practically all the sampled areas (47 of 48 plots), damaged trees were not homogeneously distributed throughout the study area (Figure 3; Appendices S10 & S11).Moreover, ozone-related symptoms were often combined with other forms of damage (p < 0.001), which suggests that ozone could be making trees more susceptible to other stressors, as proposed elsewhere (Langebartels et al., 1998).However, it is also possible that other types of stress make trees more susceptible to ozone damage.
The data gathered helped delineate the knowledge gaps identified by SEDEMA, CONANP, and SRX, particularly regarding the effect of reforestation and tree position, but due to the distribution of ozone damage and the spatial component detected in the landscape and the effect of blocks in the analyses (Figure 3; Appendix S11), a larger sample is needed to examine this in more detail.Over time, ozone damage weakened trees; older trees had a higher percentage of their foliage damaged than younger trees (Figure 4c,d), which is consistent with results of controlled experiments showing that ozone damage is cumulative (Grulke & Heath, 2020).However, the effects of natural selection cannot be ruled out; younger trees could have been selected for and carry alleles that confer some sort of resistance to ozone pollution.This is particularly important because juvenile trees affected by ozone could be more vulnerable to other types of stressors as they age, which could compromise forest conservation and increase forest decline in the long term.Healthy trees were smaller than damaged trees of the same age, which suggests that trees may be growing less tall so as to tolerate ozone pollution (Figure 5a).Replanted trees also seem to be growing less than natural regenerated trees (Figure 5b).Exposed trees tended to be taller than nursed individuals (Figure 5c).These results mirror similar findings in other trees (Felzer et al., 2007;Pye, 1988) and herbs (Han et al., 2020;Whitfield et al., 1997) and hint that selective forces and facilitation might be at play.Landscape ecology and genomics analyses are needed to explore this further, particularly because of the spatial effects we detected.
Although the sampling sites and blocks' location had an effect in the ANOVA and glmer analyses, our results suggested that elevation had no effect on ozone damage (Appendix S12), despite previous studies in the same area showing higher ozone incidence at higher elevations (Alvarado-Rosales et al., 2017).This could be explained by other factors, such as wind cur- rents or hill exposure.Such analyses should further include local measures of ozone levels, which could also be obtained with participatory methods.Alternatively, one must consider that historical ozone deposition has been so high in CDMX that saturation has already been attained, which could blur the effect of elevation.

Implications of participatory monitoring for forest management and conservation
We believe that involving local communities facilitated forest monitoring (Danielsen et al., 2011;Holck, 2008) and that using digital technology improved data quality (DeVries et al., 2016;Pratihast et al., 2013).We found that more specific variables could be monitored with this approach than with automated methods based on remote sensing.For instance, we had damage data for nearly 1700 trees, which is the largest sample ever for the PNDL, and the cost was minimal.Our results are already being used by SEDEMA and SRX to improve their conservation and restoration activities, given that data collection was aimed at answering specific questions.Our results will inform the updated version of the PDNL management plan, and the condition of seed source trees for reforestation will be evaluated according to the criteria used herein.
Our participatory approach can be easily adapted to monitor other relevant aspects of peri-urban forests.For instance, ozone levels are not monitored in the forest because stations require either electricity or frequent visits, which increases operating costs and presents safety concerns.With proper training, rangers could collect such data with portable equipment and digital tools.Monitoring restoration efforts could also become easier with a participatory approach.For instance, there are but few records on survival for the more than 60,000 trees planted in ozone-affected zones of PNDL, and their germplasm origin is seldom recorded.Both SRX and SEDEMA are interested in gathering these data so they can adjust their management plans.They wish to use our approach to develop an autonomous local monitoring program (as in Danielsen et al. [2009]) in which customized forms are used in digital collection tools and field visits are recurrent.
Finally, the participatory system and proposed variables to monitor ozone damage can be extended to other polluted forests.In Mexico, participatory monitoring could also be backed up by institutions such as CONABIO, which can provide servers for data storage and tools for visualizing and interpreting results.Other institutions or organizations could offer similar services at a more global level (e.g., Biodiversity International, World Wildlife Fund).Indeed, our experience shows that providing technological infrastructure and creating local capabilities (Webber et al., 2022) can considerably ease the lack of resources that local communities and environmental agencies often face in undertaking forest monitoring (DeVries et al., 2016;Evans et al., 2018;Pratihast et al., 2013).Supporting local monitoring groups and transdisciplinary processes in alliance (or not) with scientific researchers will help local actors develop their own tools to adjust their monitoring needs and guide their own forest management decisions (Lang et al., 2012).Therefore, the transdisciplinary steps we presented here will help obtain relevant information for the management of forests affected by local problems that in many cases fall outside national objectives.

FIGURE 2
FIGURE 2 Location of study area and distribution of forest condition categories in sampling plots: (a) Parque Nacional Desierto de los Leones (PNDL) in Mexico City (CDMX), (b) Santa Rosa Xochiac and other communities near PNDL (c, PNDL monastery; x, Cruz de Coloxtitla), (c) distribution of sampling plots (only some plot numbers are shown for reference), and (d) number of trees sampled in each plot and their condition category.

FIGURE 3
FIGURE 3 Damaged trees per plot relative to tree position and origin (nursed, plants growing below larger trees or bushes; exposed, trees growing in open areas; horizontal line, median; bar ends, range of the data; whiskers, minimum and maximum values within a certain distance from the upper and lower quartiles of the data; points, outliers).

FIGURE 4
FIGURE 4 Tree condition and percentage of ozone damage (i.e., damage to foliage) relative to tree age (estimated by number of nodes): (a) number of asymptomatic trees and trees showing ozone damage and ozone damage in combination with other signs of damage, (b) distribution of tree age across condition classes, (c) number of trees showing different percentages of damage from ozone, and (d) distribution of the percentage of ozone damage (****p < 0.0001; ***p < 0.001; **p < 0.01; *p < 0.05).

FIGURE 5
FIGURE 5 Relationship between tree age and height for different condition classes for (a) all individuals, (b) origin of individuals, and (c) position (nursed, plants growing below larger trees or bushes; exposed, trees growing in open areas) of individuals (shading, 95% bootstrapped CI; R 2 c = 0.43, R 2 m = 0.27).Estimates and associated p values for each variable are in Appendix S13.